code stringlengths 87 55.2k | code_codestyle int64 0 349 | style_context stringlengths 135 49.1k | style_context_codestyle int64 0 349 | label int64 0 1 |
|---|---|---|---|---|
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_flax_available,
is_tf_available,
is_tokenizers_available,
is_torch_available,
)
lowercase__ :int = {
"configuration_blenderbot": [
"BLENDERBOT_PRETRAINED_CONFIG_ARCHIVE_MAP",
"BlenderbotConfig",
"BlenderbotOnnxConfig",
],
"tokenization_blenderbot": ["BlenderbotTokenizer"],
}
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
lowercase__ :Optional[int] = ["BlenderbotTokenizerFast"]
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
lowercase__ :List[str] = [
"BLENDERBOT_PRETRAINED_MODEL_ARCHIVE_LIST",
"BlenderbotForCausalLM",
"BlenderbotForConditionalGeneration",
"BlenderbotModel",
"BlenderbotPreTrainedModel",
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
lowercase__ :Any = [
"TFBlenderbotForConditionalGeneration",
"TFBlenderbotModel",
"TFBlenderbotPreTrainedModel",
]
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
lowercase__ :Union[str, Any] = [
"FlaxBlenderbotForConditionalGeneration",
"FlaxBlenderbotModel",
"FlaxBlenderbotPreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_blenderbot import (
BLENDERBOT_PRETRAINED_CONFIG_ARCHIVE_MAP,
BlenderbotConfig,
BlenderbotOnnxConfig,
)
from .tokenization_blenderbot import BlenderbotTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_blenderbot_fast import BlenderbotTokenizerFast
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_blenderbot import (
BLENDERBOT_PRETRAINED_MODEL_ARCHIVE_LIST,
BlenderbotForCausalLM,
BlenderbotForConditionalGeneration,
BlenderbotModel,
BlenderbotPreTrainedModel,
)
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_blenderbot import (
TFBlenderbotForConditionalGeneration,
TFBlenderbotModel,
TFBlenderbotPreTrainedModel,
)
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_flax_blenderbot import (
FlaxBlenderbotForConditionalGeneration,
FlaxBlenderbotModel,
FlaxBlenderbotPreTrainedModel,
)
else:
import sys
lowercase__ :Union[str, Any] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)
| 101 |
import argparse
import json
import os
import torch
from torch import nn
from transformers import NllbMoeConfig, NllbMoeModel
from transformers.modeling_utils import dtype_byte_size
from transformers.utils import WEIGHTS_INDEX_NAME, WEIGHTS_NAME
def lowercase_ (A : List[str] ):
snake_case__ : Tuple = [
'encoder.version',
'decoder.version',
'model.encoder.version',
'model.decoder.version',
'decoder.output_projection.weight',
'_float_tensor',
'encoder.embed_positions._float_tensor',
'decoder.embed_positions._float_tensor',
]
for k in ignore_keys:
state_dict.pop(A , A )
def lowercase_ (A : str ):
snake_case__ , snake_case__ : Union[str, Any] = emb.weight.shape
snake_case__ : str = nn.Linear(A , A , bias=A )
snake_case__ : str = emb.weight.data
return lin_layer
def lowercase_ (A : Optional[int] , A : Union[str, Any]=None ):
snake_case__ : Any = {}
for old_key in state_dict.keys():
snake_case__ : Tuple = old_key
if "moe_layer.experts." in key:
if expert_idx is not None:
snake_case__ : int = key.replace('moe_layer.experts.0' , F'''ffn.experts.expert_{expert_idx}''' )
else:
snake_case__ : Any = key.replace('moe_layer.experts.' , 'ffn.experts.expert_' )
if "gate" in key:
snake_case__ : Dict = key.replace('.moe_layer.gate.wg' , '.ffn.router.classifier' )
if "fc2" and "experts" not in key:
snake_case__ : str = key.replace('.fc2.' , '.ffn.fc2.' )
if "fc1" and "experts" not in key:
snake_case__ : str = key.replace('.fc1.' , '.ffn.fc1.' )
if ".encoder_attn." in key:
snake_case__ : Tuple = key.replace('.encoder_attn.' , '.cross_attention.' )
if "encoder_attn_layer_norm" in key:
snake_case__ : Tuple = key.replace('encoder_attn_layer_norm' , 'cross_attention_layer_norm' )
if "final_layer_norm" in key:
snake_case__ : Optional[int] = key.replace('final_layer_norm' , 'ff_layer_norm' )
snake_case__ : Dict = state_dict[old_key]
return new_dict
def lowercase_ (A : List[Any] , A : Tuple , A : List[Any] , A : List[str] , A : str = WEIGHTS_NAME ):
snake_case__ : Dict = []
snake_case__ : str = 0
os.makedirs(A , exist_ok=A )
for expert in range(A ):
snake_case__ : Tuple = switch_checkpoint_path + F'''-rank-{expert}.pt'''
if os.path.isfile(A ):
snake_case__ : Optional[Any] = torch.load(A )['model']
remove_ignore_keys_(A )
snake_case__ : Optional[Any] = rename_fairseq_keys(A , A )
snake_case__ : Dict = os.path.join(
A , weights_name.replace('.bin' , F'''-{len(A )+1:05d}-of-???.bin''' ) )
torch.save(A , A )
sharded_state_dicts.append(expert_state.keys() )
total_size += sum([value.numel() for key, value in expert_state.items()] ) * dtype_byte_size(
expert_state[list(A )[0]].dtype )
# Add the last block
snake_case__ : Tuple = os.path.join(A , weights_name.replace('.bin' , F'''-{len(A )+1:05d}-of-???.bin''' ) )
snake_case__ : Union[str, Any] = torch.load(switch_checkpoint_path + '-shared.pt' )['model']
remove_ignore_keys_(A )
snake_case__ : str = rename_fairseq_keys(A , A )
snake_case__ : Any = shared_weights['decoder.embed_tokens.weight']
sharded_state_dicts.append(shared_weights.keys() )
# If we only have the shared weights (dummy model/experts saved on the same file)
if len(A ) == 1:
snake_case__ : Any = os.path.join(A , A )
torch.save(A , A )
return {weights_name: sharded_state_dicts[0]}, None
else:
torch.save(A , A )
# Otherwise, let's build the index
snake_case__ : Tuple = {}
for idx, shard in enumerate(A ):
snake_case__ : Optional[int] = weights_name.replace('.bin' , F'''-{idx+1:05d}-of-{len(A ):05d}.bin''' )
snake_case__ : List[Any] = os.path.join(A , weights_name.replace('.bin' , F'''-{idx+1:05d}-of-???.bin''' ) )
os.rename(A , os.path.join(A , A ) )
for key in shard:
snake_case__ : Any = shard_file
# Add the metadata
snake_case__ : int = {'total_size': total_size}
snake_case__ : Dict = {'metadata': metadata, 'weight_map': weight_map}
with open(os.path.join(A , A ) , 'w' , encoding='utf-8' ) as f:
snake_case__ : Any = json.dumps(A , indent=2 , sort_keys=A ) + '\n'
f.write(A )
return metadata, index
if __name__ == "__main__":
a_ :int = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--nllb_moe_checkpoint_path",
default="/home/arthur_huggingface_co/fairseq/weights/checkpoints/model_moe_54b/checkpoint_2_300000",
type=str,
required=False,
help="Path to a directory containing a folder per layer. Follows the original Google format.",
)
parser.add_argument("--dtype", default="float32", type=str, required=False, help="dtype of the saved model")
parser.add_argument(
"--pytorch_dump_folder_path",
default="/home/arthur_huggingface_co/fairseq/weights/checkpoints/hf-converted-moe-54b",
type=str,
required=False,
help="Path to the output pytorch model.",
)
a_ :Optional[Any] = parser.parse_args()
a_ , a_ :Optional[Any] = shard_on_the_fly(
args.nllb_moe_checkpoint_path,
args.pytorch_dump_folder_path,
128,
args.dtype,
)
a_ :List[str] = NllbMoeConfig.from_pretrained(
"facebook/nllb-200-3.3B", encoder_sparse_step=4, decoder_sparse_step=4, num_experts=128
)
config.save_pretrained(args.pytorch_dump_folder_path)
a_ :int = NllbMoeModel.from_pretrained(args.pytorch_dump_folder_path)
print("Done")
model.save_pretrained(args.pytorch_dump_folder_path)
| 277 | 0 |
"""simple docstring"""
from collections import OrderedDict
from typing import Mapping
from packaging import version
from ...configuration_utils import PretrainedConfig
from ...onnx import OnnxConfig
from ...utils import logging
SCREAMING_SNAKE_CASE : Union[str, Any] = logging.get_logger(__name__)
SCREAMING_SNAKE_CASE : Optional[int] = {
"""sail/poolformer_s12""": """https://huggingface.co/sail/poolformer_s12/resolve/main/config.json""",
# See all PoolFormer models at https://huggingface.co/models?filter=poolformer
}
class _UpperCAmelCase ( __snake_case ):
'''simple docstring'''
lowerCamelCase__ ='poolformer'
def __init__(self , a_=3 , a_=16 , a_=16 , a_=3 , a_=4.0 , a_=[2, 2, 6, 2] , a_=[64, 1_28, 3_20, 5_12] , a_=[7, 3, 3, 3] , a_=[4, 2, 2, 2] , a_=[2, 1, 1, 1] , a_=4 , a_=0.0 , a_="gelu" , a_=True , a_=1E-5 , a_=0.02 , **a_ , ):
'''simple docstring'''
__snake_case : int = num_channels
__snake_case : Union[str, Any] = patch_size
__snake_case : Optional[int] = stride
__snake_case : List[Any] = padding
__snake_case : Optional[Any] = pool_size
__snake_case : List[Any] = hidden_sizes
__snake_case : Union[str, Any] = mlp_ratio
__snake_case : List[Any] = depths
__snake_case : Dict = patch_sizes
__snake_case : str = strides
__snake_case : Dict = num_encoder_blocks
__snake_case : List[str] = drop_path_rate
__snake_case : int = hidden_act
__snake_case : List[str] = use_layer_scale
__snake_case : List[str] = layer_scale_init_value
__snake_case : List[Any] = initializer_range
super().__init__(**a_ )
class _UpperCAmelCase ( __snake_case ):
'''simple docstring'''
lowerCamelCase__ =version.parse('1.11' )
@property
def SCREAMING_SNAKE_CASE (self ):
'''simple docstring'''
return OrderedDict(
[
('''pixel_values''', {0: '''batch''', 1: '''num_channels''', 2: '''height''', 3: '''width'''}),
] )
@property
def SCREAMING_SNAKE_CASE (self ):
'''simple docstring'''
return 2E-3
| 102 |
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_sentencepiece_available,
is_tokenizers_available,
is_torch_available,
)
a_ :Optional[Any] = {"configuration_reformer": ["REFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP", "ReformerConfig"]}
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :str = ["ReformerTokenizer"]
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :int = ["ReformerTokenizerFast"]
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :List[str] = [
"REFORMER_PRETRAINED_MODEL_ARCHIVE_LIST",
"ReformerAttention",
"ReformerForMaskedLM",
"ReformerForQuestionAnswering",
"ReformerForSequenceClassification",
"ReformerLayer",
"ReformerModel",
"ReformerModelWithLMHead",
"ReformerPreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_reformer import REFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, ReformerConfig
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_reformer import ReformerTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_reformer_fast import ReformerTokenizerFast
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_reformer import (
REFORMER_PRETRAINED_MODEL_ARCHIVE_LIST,
ReformerAttention,
ReformerForMaskedLM,
ReformerForQuestionAnswering,
ReformerForSequenceClassification,
ReformerLayer,
ReformerModel,
ReformerModelWithLMHead,
ReformerPreTrainedModel,
)
else:
import sys
a_ :Optional[Any] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)
| 277 | 0 |
import re
def UpperCamelCase( __UpperCamelCase : str ):
if len(re.findall('''[ATCG]''' ,__UpperCamelCase ) ) != len(__UpperCamelCase ):
raise ValueError('''Invalid Strand''' )
return dna.translate(dna.maketrans('''ATCG''' ,'''TAGC''' ) )
if __name__ == "__main__":
import doctest
doctest.testmod()
| 103 |
import itertools
import random
import unittest
import numpy as np
from transformers import BatchFeature, SpeechTaFeatureExtractor
from transformers.testing_utils import require_torch
from transformers.utils.import_utils import is_torch_available
from ...test_sequence_feature_extraction_common import SequenceFeatureExtractionTestMixin
if is_torch_available():
import torch
a_ :Any = random.Random()
def lowercase_ (A : int , A : Union[str, Any]=1.0 , A : List[str]=None , A : Any=None ):
if rng is None:
snake_case__ : List[str] = global_rng
snake_case__ : int = []
for batch_idx in range(shape[0] ):
values.append([] )
for _ in range(shape[1] ):
values[-1].append(rng.random() * scale )
return values
@require_torch
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
def __init__( self : Optional[Any], _snake_case : List[str], _snake_case : Tuple=7, _snake_case : Union[str, Any]=4_0_0, _snake_case : Any=2_0_0_0, _snake_case : Dict=1, _snake_case : Optional[Any]=0.0, _snake_case : List[Any]=1_6_0_0_0, _snake_case : List[Any]=True, _snake_case : List[Any]=8_0, _snake_case : Dict=1_6, _snake_case : str=6_4, _snake_case : Tuple="hann_window", _snake_case : Union[str, Any]=8_0, _snake_case : Optional[Any]=7_6_0_0, _snake_case : str=1e-10, _snake_case : Any=True, ) ->Union[str, Any]:
snake_case__ : Optional[int] = parent
snake_case__ : Optional[Any] = batch_size
snake_case__ : List[Any] = min_seq_length
snake_case__ : List[Any] = max_seq_length
snake_case__ : Any = (self.max_seq_length - self.min_seq_length) // (self.batch_size - 1)
snake_case__ : Tuple = feature_size
snake_case__ : List[Any] = padding_value
snake_case__ : Any = sampling_rate
snake_case__ : Dict = do_normalize
snake_case__ : Union[str, Any] = num_mel_bins
snake_case__ : Any = hop_length
snake_case__ : Any = win_length
snake_case__ : Any = win_function
snake_case__ : Optional[int] = fmin
snake_case__ : int = fmax
snake_case__ : Union[str, Any] = mel_floor
snake_case__ : Union[str, Any] = return_attention_mask
def lowercase_ ( self : Optional[int] ) ->List[str]:
return {
"feature_size": self.feature_size,
"padding_value": self.padding_value,
"sampling_rate": self.sampling_rate,
"do_normalize": self.do_normalize,
"num_mel_bins": self.num_mel_bins,
"hop_length": self.hop_length,
"win_length": self.win_length,
"win_function": self.win_function,
"fmin": self.fmin,
"fmax": self.fmax,
"mel_floor": self.mel_floor,
"return_attention_mask": self.return_attention_mask,
}
def lowercase_ ( self : Any, _snake_case : Optional[Any]=False, _snake_case : List[str]=False ) ->Union[str, Any]:
def _flatten(_snake_case : List[str] ):
return list(itertools.chain(*_snake_case ) )
if equal_length:
snake_case__ : Any = floats_list((self.batch_size, self.max_seq_length) )
else:
# make sure that inputs increase in size
snake_case__ : int = [
_flatten(floats_list((x, self.feature_size) ) )
for x in range(self.min_seq_length, self.max_seq_length, self.seq_length_diff )
]
if numpify:
snake_case__ : Any = [np.asarray(_snake_case ) for x in speech_inputs]
return speech_inputs
def lowercase_ ( self : Union[str, Any], _snake_case : str=False, _snake_case : Dict=False ) ->List[str]:
if equal_length:
snake_case__ : Optional[Any] = [floats_list((self.max_seq_length, self.num_mel_bins) ) for _ in range(self.batch_size )]
else:
# make sure that inputs increase in size
snake_case__ : List[str] = [
floats_list((x, self.num_mel_bins) )
for x in range(self.min_seq_length, self.max_seq_length, self.seq_length_diff )
]
if numpify:
snake_case__ : int = [np.asarray(_snake_case ) for x in speech_inputs]
return speech_inputs
@require_torch
class snake_case__ ( lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = SpeechTaFeatureExtractor
def lowercase_ ( self : int ) ->Union[str, Any]:
snake_case__ : List[str] = SpeechTaFeatureExtractionTester(self )
def lowercase_ ( self : Any, _snake_case : Dict ) ->Any:
self.assertTrue(np.all(np.mean(_snake_case, axis=0 ) < 1e-3 ) )
self.assertTrue(np.all(np.abs(np.var(_snake_case, axis=0 ) - 1 ) < 1e-3 ) )
def lowercase_ ( self : List[Any] ) ->Union[str, Any]:
# Tests that all call wrap to encode_plus and batch_encode_plus
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
# create three inputs of length 800, 1000, and 1200
snake_case__ : int = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Tuple = [np.asarray(_snake_case ) for speech_input in speech_inputs]
# Test not batched input
snake_case__ : str = feat_extract(speech_inputs[0], return_tensors='np' ).input_values
snake_case__ : List[str] = feat_extract(np_speech_inputs[0], return_tensors='np' ).input_values
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test batched
snake_case__ : Any = feat_extract(_snake_case, return_tensors='np' ).input_values
snake_case__ : Union[str, Any] = feat_extract(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
def lowercase_ ( self : int ) ->Optional[int]:
snake_case__ : List[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Tuple = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : int = ['longest', 'max_length', 'do_not_pad']
snake_case__ : List[str] = [None, 1_6_0_0, None]
for max_length, padding in zip(_snake_case, _snake_case ):
snake_case__ : Optional[int] = feat_extract(_snake_case, padding=_snake_case, max_length=_snake_case, return_tensors='np' )
snake_case__ : Optional[int] = processed.input_values
self._check_zero_mean_unit_variance(input_values[0][:8_0_0] )
self.assertTrue(input_values[0][8_0_0:].sum() < 1e-6 )
self._check_zero_mean_unit_variance(input_values[1][:1_0_0_0] )
self.assertTrue(input_values[0][1_0_0_0:].sum() < 1e-6 )
self._check_zero_mean_unit_variance(input_values[2][:1_2_0_0] )
def lowercase_ ( self : Union[str, Any] ) ->Optional[Any]:
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Tuple = range(8_0_0, 1_4_0_0, 2_0_0 )
snake_case__ : Optional[Any] = [floats_list((1, x) )[0] for x in lengths]
snake_case__ : Union[str, Any] = ['longest', 'max_length', 'do_not_pad']
snake_case__ : str = [None, 1_6_0_0, None]
for max_length, padding in zip(_snake_case, _snake_case ):
snake_case__ : List[str] = feat_extract(_snake_case, max_length=_snake_case, padding=_snake_case )
snake_case__ : Tuple = processed.input_values
self._check_zero_mean_unit_variance(input_values[0][:8_0_0] )
self._check_zero_mean_unit_variance(input_values[1][:1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2][:1_2_0_0] )
def lowercase_ ( self : List[Any] ) ->Optional[Any]:
snake_case__ : Any = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : str = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Optional[Any] = feat_extract(
_snake_case, truncation=_snake_case, max_length=1_0_0_0, padding='max_length', return_tensors='np' )
snake_case__ : int = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1] )
self._check_zero_mean_unit_variance(input_values[2] )
def lowercase_ ( self : int ) ->Union[str, Any]:
snake_case__ : str = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Dict = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : str = feat_extract(
_snake_case, truncation=_snake_case, max_length=1_0_0_0, padding='longest', return_tensors='np' )
snake_case__ : Dict = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1, :1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2] )
# make sure that if max_length < longest -> then pad to max_length
self.assertTrue(input_values.shape == (3, 1_0_0_0) )
snake_case__ : Tuple = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : List[str] = feat_extract(
_snake_case, truncation=_snake_case, max_length=2_0_0_0, padding='longest', return_tensors='np' )
snake_case__ : Optional[Any] = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1, :1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2] )
# make sure that if max_length > longest -> then pad to longest
self.assertTrue(input_values.shape == (3, 1_2_0_0) )
def lowercase_ ( self : List[str] ) ->Dict:
snake_case__ : Dict = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : List[Any] = np.random.rand(1_0_0 ).astype(np.floataa )
snake_case__ : int = np_speech_inputs.tolist()
for inputs in [py_speech_inputs, np_speech_inputs]:
snake_case__ : int = feature_extractor.pad([{'input_values': inputs}], return_tensors='np' )
self.assertTrue(np_processed.input_values.dtype == np.floataa )
snake_case__ : Optional[int] = feature_extractor.pad([{'input_values': inputs}], return_tensors='pt' )
self.assertTrue(pt_processed.input_values.dtype == torch.floataa )
def lowercase_ ( self : Optional[int] ) ->Optional[Any]:
# Tests that all call wrap to encode_plus and batch_encode_plus
snake_case__ : str = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
# create three inputs of length 800, 1000, and 1200
snake_case__ : List[Any] = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Dict = [np.asarray(_snake_case ) for speech_input in speech_inputs]
# Test feature size
snake_case__ : Optional[int] = feature_extractor(audio_target=_snake_case, padding=_snake_case, return_tensors='np' ).input_values
self.assertTrue(input_values.ndim == 3 )
self.assertTrue(input_values.shape[-1] == feature_extractor.num_mel_bins )
# Test not batched input
snake_case__ : Dict = feature_extractor(speech_inputs[0], return_tensors='np' ).input_values
snake_case__ : Any = feature_extractor(np_speech_inputs[0], return_tensors='np' ).input_values
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test batched
snake_case__ : Dict = feature_extractor(_snake_case, return_tensors='np' ).input_values
snake_case__ : Dict = feature_extractor(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test 2-D numpy arrays are batched.
snake_case__ : Optional[Any] = [floats_list((1, x) )[0] for x in (8_0_0, 8_0_0, 8_0_0)]
snake_case__ : int = np.asarray(_snake_case )
snake_case__ : Union[str, Any] = feature_extractor(_snake_case, return_tensors='np' ).input_values
snake_case__ : Union[str, Any] = feature_extractor(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
def lowercase_ ( self : Union[str, Any] ) ->str:
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : List[str] = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : Tuple = BatchFeature({input_name: speech_inputs} )
self.assertTrue(all(len(_snake_case ) == len(_snake_case ) for x, y in zip(_snake_case, processed_features[input_name] ) ) )
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target(equal_length=_snake_case )
snake_case__ : Union[str, Any] = BatchFeature({input_name: speech_inputs}, tensor_type='np' )
snake_case__ : Dict = processed_features[input_name]
if len(batch_features_input.shape ) < 3:
snake_case__ : List[str] = batch_features_input[:, :, None]
self.assertTrue(
batch_features_input.shape
== (self.feat_extract_tester.batch_size, len(speech_inputs[0] ), self.feat_extract_tester.num_mel_bins) )
@require_torch
def lowercase_ ( self : List[str] ) ->Any:
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target(equal_length=_snake_case )
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Tuple = feat_extract.model_input_names[0]
snake_case__ : List[Any] = BatchFeature({input_name: speech_inputs}, tensor_type='pt' )
snake_case__ : Tuple = processed_features[input_name]
if len(batch_features_input.shape ) < 3:
snake_case__ : Any = batch_features_input[:, :, None]
self.assertTrue(
batch_features_input.shape
== (self.feat_extract_tester.batch_size, len(speech_inputs[0] ), self.feat_extract_tester.num_mel_bins) )
@require_torch
def lowercase_ ( self : Optional[int] ) ->Tuple:
snake_case__ : Dict = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Union[str, Any] = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : List[str] = BatchFeature({input_name: speech_inputs} )
snake_case__ : int = feat_extract.num_mel_bins # hack!
snake_case__ : Tuple = feat_extract.pad(_snake_case, padding='longest', return_tensors='np' )[input_name]
snake_case__ : Union[str, Any] = feat_extract.pad(_snake_case, padding='longest', return_tensors='pt' )[input_name]
self.assertTrue(abs(input_np.astype(np.floataa ).sum() - input_pt.numpy().astype(np.floataa ).sum() ) < 1e-2 )
def lowercase_ ( self : int ) ->Any:
snake_case__ : Any = self.feat_extract_dict
snake_case__ : List[Any] = True
snake_case__ : Union[str, Any] = self.feature_extraction_class(**_snake_case )
snake_case__ : Any = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : List[Any] = [len(_snake_case ) for x in speech_inputs]
snake_case__ : Union[str, Any] = feat_extract.model_input_names[0]
snake_case__ : Optional[int] = BatchFeature({input_name: speech_inputs} )
snake_case__ : List[str] = feat_extract.num_mel_bins # hack!
snake_case__ : str = feat_extract.pad(_snake_case, padding='longest', return_tensors='np' )
self.assertIn('attention_mask', _snake_case )
self.assertListEqual(list(processed.attention_mask.shape ), list(processed[input_name].shape[:2] ) )
self.assertListEqual(processed.attention_mask.sum(-1 ).tolist(), _snake_case )
def lowercase_ ( self : Optional[int] ) ->str:
snake_case__ : int = self.feat_extract_dict
snake_case__ : List[str] = True
snake_case__ : Tuple = self.feature_extraction_class(**_snake_case )
snake_case__ : List[str] = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : str = [len(_snake_case ) for x in speech_inputs]
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : Optional[int] = BatchFeature({input_name: speech_inputs} )
snake_case__ : Optional[Any] = min(_snake_case )
snake_case__ : Union[str, Any] = feat_extract.num_mel_bins # hack!
snake_case__ : Tuple = feat_extract.pad(
_snake_case, padding='max_length', max_length=_snake_case, truncation=_snake_case, return_tensors='np' )
self.assertIn('attention_mask', _snake_case )
self.assertListEqual(
list(processed_pad.attention_mask.shape ), [processed_pad[input_name].shape[0], max_length] )
self.assertListEqual(
processed_pad.attention_mask[:, :max_length].sum(-1 ).tolist(), [max_length for x in speech_inputs] )
def lowercase_ ( self : List[Any], _snake_case : Optional[int] ) ->Optional[Any]:
from datasets import load_dataset
snake_case__ : str = load_dataset('hf-internal-testing/librispeech_asr_dummy', 'clean', split='validation' )
# automatic decoding with librispeech
snake_case__ : Dict = ds.sort('id' ).select(range(_snake_case ) )[:num_samples]['audio']
return [x["array"] for x in speech_samples]
def lowercase_ ( self : str ) ->str:
# fmt: off
snake_case__ : List[Any] = torch.tensor(
[2.3804e-03, 2.0752e-03, 1.9836e-03, 2.1057e-03, 1.6174e-03,
3.0518e-04, 9.1553e-05, 3.3569e-04, 9.7656e-04, 1.8311e-03,
2.0142e-03, 2.1057e-03, 1.7395e-03, 4.5776e-04, -3.9673e-04,
4.5776e-04, 1.0071e-03, 9.1553e-05, 4.8828e-04, 1.1597e-03,
7.3242e-04, 9.4604e-04, 1.8005e-03, 1.8311e-03, 8.8501e-04,
4.2725e-04, 4.8828e-04, 7.3242e-04, 1.0986e-03, 2.1057e-03] )
# fmt: on
snake_case__ : Union[str, Any] = self._load_datasamples(1 )
snake_case__ : Optional[int] = SpeechTaFeatureExtractor()
snake_case__ : List[Any] = feature_extractor(_snake_case, return_tensors='pt' ).input_values
self.assertEquals(input_values.shape, (1, 9_3_6_8_0) )
self.assertTrue(torch.allclose(input_values[0, :3_0], _snake_case, atol=1e-6 ) )
def lowercase_ ( self : Any ) ->str:
# fmt: off
snake_case__ : Optional[Any] = torch.tensor(
[-2.6_8_7_0, -3.0_1_0_4, -3.1_3_5_6, -3.5_3_5_2, -3.0_0_4_4, -3.0_3_5_3, -3.4_7_1_9, -3.6_7_7_7,
-3.1_5_2_0, -2.9_4_3_5, -2.6_5_5_3, -2.8_7_9_5, -2.9_9_4_4, -2.5_9_2_1, -3.0_2_7_9, -3.0_3_8_6,
-3.0_8_6_4, -3.1_2_9_1, -3.2_3_5_3, -2.7_4_4_4, -2.6_8_3_1, -2.7_2_8_7, -3.1_7_6_1, -3.1_5_7_1,
-3.2_7_2_6, -3.0_5_8_2, -3.1_0_0_7, -3.4_5_3_3, -3.4_6_9_5, -3.0_9_9_8] )
# fmt: on
snake_case__ : List[str] = self._load_datasamples(1 )
snake_case__ : str = SpeechTaFeatureExtractor()
snake_case__ : Optional[Any] = feature_extractor(audio_target=_snake_case, return_tensors='pt' ).input_values
self.assertEquals(input_values.shape, (1, 3_6_6, 8_0) )
self.assertTrue(torch.allclose(input_values[0, 0, :3_0], _snake_case, atol=1e-4 ) )
| 277 | 0 |
'''simple docstring'''
import inspect
import tempfile
import unittest
from huggingface_hub import hf_hub_download
from transformers import is_torch_available
from transformers.testing_utils import is_flaky, require_torch, slow, torch_device
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
lowerCAmelCase__ = 1e-4
if is_torch_available():
import torch
from transformers import AutoformerConfig, AutoformerForPrediction, AutoformerModel
from transformers.models.autoformer.modeling_autoformer import AutoformerDecoder, AutoformerEncoder
@require_torch
class lowercase_ :
"""simple docstring"""
def __init__( self : Any ,lowercase__ : Optional[Any] ,lowercase__ : List[str]=1_6 ,lowercase__ : List[Any]=1_3 ,lowercase__ : str=7 ,lowercase__ : int=1_4 ,lowercase__ : Dict=1_0 ,lowercase__ : Any=1_9 ,lowercase__ : Optional[Any]=5 ,lowercase__ : Tuple=4 ,lowercase__ : int=True ,lowercase__ : Any=1_6 ,lowercase__ : Tuple=2 ,lowercase__ : str=4 ,lowercase__ : str=4 ,lowercase__ : Dict="gelu" ,lowercase__ : str=0.1 ,lowercase__ : Tuple=0.1 ,lowercase__ : List[Any]=[1, 2, 3, 4, 5] ,lowercase__ : Optional[int]=2_5 ,lowercase__ : str=5 ,):
__lowercase = d_model
__lowercase = parent
__lowercase = batch_size
__lowercase = prediction_length
__lowercase = context_length
__lowercase = cardinality
__lowercase = num_time_features
__lowercase = lags_sequence
__lowercase = embedding_dimension
__lowercase = is_training
__lowercase = hidden_size
__lowercase = num_hidden_layers
__lowercase = num_attention_heads
__lowercase = intermediate_size
__lowercase = hidden_act
__lowercase = hidden_dropout_prob
__lowercase = attention_probs_dropout_prob
__lowercase = context_length
__lowercase = prediction_length + label_length
__lowercase = label_length
__lowercase = moving_average
__lowercase = autocorrelation_factor
def SCREAMING_SNAKE_CASE ( self : Tuple ):
return AutoformerConfig(
d_model=self.d_model ,encoder_layers=self.num_hidden_layers ,decoder_layers=self.num_hidden_layers ,encoder_attention_heads=self.num_attention_heads ,decoder_attention_heads=self.num_attention_heads ,encoder_ffn_dim=self.intermediate_size ,decoder_ffn_dim=self.intermediate_size ,dropout=self.hidden_dropout_prob ,attention_dropout=self.attention_probs_dropout_prob ,prediction_length=self.prediction_length ,context_length=self.context_length ,label_length=self.label_length ,lags_sequence=self.lags_sequence ,num_time_features=self.num_time_features ,num_static_categorical_features=1 ,cardinality=[self.cardinality] ,embedding_dimension=[self.embedding_dimension] ,moving_average=self.moving_average ,)
def SCREAMING_SNAKE_CASE ( self : Dict ,lowercase__ : List[str] ):
__lowercase = config.context_length + max(config.lags_sequence )
__lowercase = ids_tensor([self.batch_size, 1] ,config.cardinality[0] )
__lowercase = floats_tensor([self.batch_size, _past_length, config.num_time_features] )
__lowercase = floats_tensor([self.batch_size, _past_length] )
__lowercase = floats_tensor([self.batch_size, _past_length] ) > 0.5
# decoder inputs
__lowercase = floats_tensor([self.batch_size, config.prediction_length, config.num_time_features] )
__lowercase = floats_tensor([self.batch_size, config.prediction_length] )
__lowercase = {
'''past_values''': past_values,
'''static_categorical_features''': static_categorical_features,
'''past_time_features''': past_time_features,
'''past_observed_mask''': past_observed_mask,
'''future_time_features''': future_time_features,
'''future_values''': future_values,
}
return inputs_dict
def SCREAMING_SNAKE_CASE ( self : Optional[Any] ):
__lowercase = self.get_config()
__lowercase = self.prepare_autoformer_inputs_dict(lowercase__ )
return config, inputs_dict
def SCREAMING_SNAKE_CASE ( self : Union[str, Any] ):
__lowercase , __lowercase = self.prepare_config_and_inputs()
return config, inputs_dict
def SCREAMING_SNAKE_CASE ( self : Any ,lowercase__ : int ,lowercase__ : Any ):
__lowercase = AutoformerModel(config=lowercase__ ).to(lowercase__ ).eval()
__lowercase = model(**lowercase__ )
__lowercase = outputs.encoder_last_hidden_state
__lowercase = outputs.last_hidden_state
with tempfile.TemporaryDirectory() as tmpdirname:
__lowercase = model.get_encoder()
encoder.save_pretrained(lowercase__ )
__lowercase = AutoformerEncoder.from_pretrained(lowercase__ ).to(lowercase__ )
__lowercase , __lowercase , __lowercase , __lowercase , __lowercase = model.create_network_inputs(**lowercase__ )
__lowercase , __lowercase = model.decomposition_layer(transformer_inputs[:, : config.context_length, ...] )
__lowercase = torch.cat(
(transformer_inputs[:, : config.context_length, ...], feature[:, : config.context_length, ...]) ,dim=-1 ,)
__lowercase = encoder(inputs_embeds=lowercase__ )[0]
self.parent.assertTrue((encoder_last_hidden_state_a - encoder_last_hidden_state).abs().max().item() < 1e-3 )
__lowercase = (
torch.mean(transformer_inputs[:, : config.context_length, ...] ,dim=1 )
.unsqueeze(1 )
.repeat(1 ,config.prediction_length ,1 )
)
__lowercase = torch.zeros(
[transformer_inputs.shape[0], config.prediction_length, transformer_inputs.shape[2]] ,device=enc_input.device ,)
__lowercase = torch.cat(
(
torch.cat((seasonal_input[:, -config.label_length :, ...], zeros) ,dim=1 ),
feature[:, config.context_length - config.label_length :, ...],
) ,dim=-1 ,)
__lowercase = torch.cat(
(
torch.cat((trend_input[:, -config.label_length :, ...], mean) ,dim=1 ),
feature[:, config.context_length - config.label_length :, ...],
) ,dim=-1 ,)
with tempfile.TemporaryDirectory() as tmpdirname:
__lowercase = model.get_decoder()
decoder.save_pretrained(lowercase__ )
__lowercase = AutoformerDecoder.from_pretrained(lowercase__ ).to(lowercase__ )
__lowercase = decoder(
trend=lowercase__ ,inputs_embeds=lowercase__ ,encoder_hidden_states=lowercase__ ,)[0]
self.parent.assertTrue((last_hidden_state_a - last_hidden_state).abs().max().item() < 1e-3 )
@require_torch
class lowercase_ (lowerCamelCase__ , lowerCamelCase__ , unittest.TestCase ):
"""simple docstring"""
SCREAMING_SNAKE_CASE : Union[str, Any] = (AutoformerModel, AutoformerForPrediction) if is_torch_available() else ()
SCREAMING_SNAKE_CASE : Any = (AutoformerForPrediction,) if is_torch_available() else ()
SCREAMING_SNAKE_CASE : List[Any] = {'feature-extraction': AutoformerModel} if is_torch_available() else {}
SCREAMING_SNAKE_CASE : Union[str, Any] = False
SCREAMING_SNAKE_CASE : int = False
SCREAMING_SNAKE_CASE : str = False
SCREAMING_SNAKE_CASE : Optional[int] = False
SCREAMING_SNAKE_CASE : Optional[Any] = False
SCREAMING_SNAKE_CASE : List[Any] = False
def SCREAMING_SNAKE_CASE ( self : Optional[Any] ):
__lowercase = AutoformerModelTester(self )
__lowercase = ConfigTester(self ,config_class=lowercase__ ,has_text_modality=lowercase__ )
def SCREAMING_SNAKE_CASE ( self : Tuple ):
self.config_tester.run_common_tests()
def SCREAMING_SNAKE_CASE ( self : List[Any] ):
__lowercase , __lowercase = self.model_tester.prepare_config_and_inputs()
for model_class in self.all_model_classes:
__lowercase = model_class(lowercase__ )
with tempfile.TemporaryDirectory() as tmpdirname:
model.save_pretrained(lowercase__ )
__lowercase , __lowercase = model_class.from_pretrained(lowercase__ ,output_loading_info=lowercase__ )
self.assertEqual(info['''missing_keys'''] ,[] )
def SCREAMING_SNAKE_CASE ( self : Dict ):
__lowercase = self.model_tester.prepare_config_and_inputs_for_common()
self.model_tester.check_encoder_decoder_model_standalone(*lowercase__ )
@unittest.skip(reason='''Model has no tokens embeddings''' )
def SCREAMING_SNAKE_CASE ( self : List[str] ):
pass
def SCREAMING_SNAKE_CASE ( self : List[str] ):
__lowercase = inspect.signature(getattr(lowercase__ ,'''forward''' ) )
# The main input is the name of the argument after `self`
__lowercase = list(model_signature.parameters.keys() )[1]
self.assertEqual(AutoformerModel.main_input_name ,lowercase__ )
def SCREAMING_SNAKE_CASE ( self : List[Any] ):
__lowercase , __lowercase = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
__lowercase = model_class(lowercase__ )
__lowercase = inspect.signature(model.forward )
# signature.parameters is an OrderedDict => so arg_names order is deterministic
__lowercase = [*signature.parameters.keys()]
__lowercase = [
'''past_values''',
'''past_time_features''',
'''past_observed_mask''',
'''static_categorical_features''',
'''static_real_features''',
'''future_values''',
'''future_time_features''',
]
if model.__class__.__name__ in ["AutoformerForPrediction"]:
expected_arg_names.append('''future_observed_mask''' )
expected_arg_names.extend(
[
'''decoder_attention_mask''',
'''head_mask''',
'''decoder_head_mask''',
'''cross_attn_head_mask''',
'''encoder_outputs''',
'''past_key_values''',
'''output_hidden_states''',
'''output_attentions''',
'''use_cache''',
'''return_dict''',
] )
self.assertListEqual(arg_names[: len(lowercase__ )] ,lowercase__ )
def SCREAMING_SNAKE_CASE ( self : Optional[Any] ):
__lowercase , __lowercase = self.model_tester.prepare_config_and_inputs_for_common()
__lowercase = True
__lowercase = getattr(self.model_tester ,'''seq_length''' ,lowercase__ )
__lowercase = getattr(self.model_tester ,'''decoder_seq_length''' ,lowercase__ )
__lowercase = getattr(self.model_tester ,'''encoder_seq_length''' ,lowercase__ )
__lowercase = getattr(self.model_tester ,'''d_model''' ,lowercase__ )
__lowercase = getattr(self.model_tester ,'''num_attention_heads''' ,lowercase__ )
__lowercase = d_model // num_attention_heads
for model_class in self.all_model_classes:
__lowercase = True
__lowercase = False
__lowercase = True
__lowercase = model_class(lowercase__ )
model.to(lowercase__ )
model.eval()
with torch.no_grad():
__lowercase = model(**self._prepare_for_class(lowercase__ ,lowercase__ ) )
__lowercase = outputs.encoder_attentions if config.is_encoder_decoder else outputs.attentions
self.assertEqual(len(lowercase__ ) ,self.model_tester.num_hidden_layers )
# check that output_attentions also work using config
del inputs_dict["output_attentions"]
__lowercase = True
__lowercase = model_class(lowercase__ )
model.to(lowercase__ )
model.eval()
with torch.no_grad():
__lowercase = model(**self._prepare_for_class(lowercase__ ,lowercase__ ) )
__lowercase = outputs.encoder_attentions
self.assertEqual(len(lowercase__ ) ,self.model_tester.num_hidden_layers )
self.assertListEqual(
list(attentions[0].shape[-3:] ) ,[self.model_tester.num_attention_heads, encoder_seq_length, dim] ,)
__lowercase = len(lowercase__ )
__lowercase = 7
if "last_hidden_state" in outputs:
correct_outlen += 1
if "trend" in outputs:
correct_outlen += 1
if "past_key_values" in outputs:
correct_outlen += 1 # past_key_values have been returned
if "loss" in outputs:
correct_outlen += 1
if "params" in outputs:
correct_outlen += 1
self.assertEqual(lowercase__ ,lowercase__ )
# decoder attentions
__lowercase = outputs.decoder_attentions
self.assertIsInstance(lowercase__ ,(list, tuple) )
self.assertEqual(len(lowercase__ ) ,self.model_tester.num_hidden_layers )
self.assertListEqual(
list(decoder_attentions[0].shape[-3:] ) ,[self.model_tester.num_attention_heads, decoder_seq_length, dim] ,)
# cross attentions
__lowercase = outputs.cross_attentions
self.assertIsInstance(lowercase__ ,(list, tuple) )
self.assertEqual(len(lowercase__ ) ,self.model_tester.num_hidden_layers )
self.assertListEqual(
list(cross_attentions[0].shape[-3:] ) ,[self.model_tester.num_attention_heads, decoder_seq_length, dim] ,)
# Check attention is always last and order is fine
__lowercase = True
__lowercase = True
__lowercase = model_class(lowercase__ )
model.to(lowercase__ )
model.eval()
with torch.no_grad():
__lowercase = model(**self._prepare_for_class(lowercase__ ,lowercase__ ) )
self.assertEqual(out_len + 2 ,len(lowercase__ ) )
__lowercase = outputs.encoder_attentions if config.is_encoder_decoder else outputs.attentions
self.assertEqual(len(lowercase__ ) ,self.model_tester.num_hidden_layers )
self.assertListEqual(
list(self_attentions[0].shape[-3:] ) ,[self.model_tester.num_attention_heads, encoder_seq_length, dim] ,)
@is_flaky()
def SCREAMING_SNAKE_CASE ( self : Union[str, Any] ):
super().test_retain_grad_hidden_states_attentions()
def _A ( A__="train-batch.pt" ):
"""simple docstring"""
__lowercase = hf_hub_download(repo_id='''hf-internal-testing/tourism-monthly-batch''' , filename=A__ , repo_type='''dataset''' )
__lowercase = torch.load(A__ , map_location=A__ )
return batch
@require_torch
@slow
class lowercase_ (unittest.TestCase ):
"""simple docstring"""
def SCREAMING_SNAKE_CASE ( self : Any ):
__lowercase = AutoformerModel.from_pretrained('''huggingface/autoformer-tourism-monthly''' ).to(lowercase__ )
__lowercase = prepare_batch()
with torch.no_grad():
__lowercase = model(
past_values=batch['''past_values'''] ,past_time_features=batch['''past_time_features'''] ,past_observed_mask=batch['''past_observed_mask'''] ,static_categorical_features=batch['''static_categorical_features'''] ,future_values=batch['''future_values'''] ,future_time_features=batch['''future_time_features'''] ,)[0]
__lowercase = torch.Size(
(6_4, model.config.prediction_length + model.config.label_length, model.config.feature_size) )
self.assertEqual(output.shape ,lowercase__ )
__lowercase = torch.tensor(
[[0.3_5_9_3, -1.3_3_9_8, 0.6_3_3_0], [0.2_2_7_9, 1.5_3_9_6, -0.1_7_9_2], [0.0_4_5_0, 1.3_2_2_5, -0.2_3_3_5]] ,device=lowercase__ )
self.assertTrue(torch.allclose(output[0, :3, :3] ,lowercase__ ,atol=lowercase__ ) )
def SCREAMING_SNAKE_CASE ( self : Union[str, Any] ):
__lowercase = AutoformerForPrediction.from_pretrained('''huggingface/autoformer-tourism-monthly''' ).to(lowercase__ )
__lowercase = prepare_batch('''val-batch.pt''' )
with torch.no_grad():
__lowercase = model(
past_values=batch['''past_values'''] ,past_time_features=batch['''past_time_features'''] ,past_observed_mask=batch['''past_observed_mask'''] ,static_categorical_features=batch['''static_categorical_features'''] ,).encoder_last_hidden_state
__lowercase = torch.Size((6_4, model.config.context_length, model.config.d_model) )
self.assertEqual(output.shape ,lowercase__ )
__lowercase = torch.tensor(
[[-0.0_7_3_4, -0.9_0_3_6, 0.8_3_5_8], [4.7_1_8_6, 2.4_1_1_3, 1.9_5_8_1], [1.7_9_5_3, 2.3_5_5_8, 1.2_9_7_0]] ,device=lowercase__ )
self.assertTrue(torch.allclose(output[0, :3, :3] ,lowercase__ ,atol=lowercase__ ) )
def SCREAMING_SNAKE_CASE ( self : str ):
__lowercase = AutoformerForPrediction.from_pretrained('''huggingface/autoformer-tourism-monthly''' ).to(lowercase__ )
__lowercase = prepare_batch('''val-batch.pt''' )
with torch.no_grad():
__lowercase = model.generate(
static_categorical_features=batch['''static_categorical_features'''] ,past_time_features=batch['''past_time_features'''] ,past_values=batch['''past_values'''] ,future_time_features=batch['''future_time_features'''] ,past_observed_mask=batch['''past_observed_mask'''] ,)
__lowercase = torch.Size((6_4, model.config.num_parallel_samples, model.config.prediction_length) )
self.assertEqual(outputs.sequences.shape ,lowercase__ )
__lowercase = torch.tensor([3_1_3_0.6_7_6_3, 4_0_5_6.5_2_9_3, 7_0_5_3.0_7_8_6] ,device=lowercase__ )
__lowercase = outputs.sequences.mean(dim=1 )
self.assertTrue(torch.allclose(mean_prediction[0, -3:] ,lowercase__ ,rtol=1e-1 ) )
| 104 |
# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from ..models.auto import AutoModelForSeqaSeqLM, AutoTokenizer
from .base import PipelineTool
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """philschmid/bart-large-cnn-samsum"""
_SCREAMING_SNAKE_CASE = (
"""This is a tool that summarizes an English text. It takes an input `text` containing the text to summarize, """
"""and returns a summary of the text."""
)
_SCREAMING_SNAKE_CASE = """summarizer"""
_SCREAMING_SNAKE_CASE = AutoTokenizer
_SCREAMING_SNAKE_CASE = AutoModelForSeqaSeqLM
_SCREAMING_SNAKE_CASE = ["""text"""]
_SCREAMING_SNAKE_CASE = ["""text"""]
def lowercase_ ( self : Optional[Any], _snake_case : str ) ->Any:
return self.pre_processor(_snake_case, return_tensors='pt', truncation=_snake_case )
def lowercase_ ( self : int, _snake_case : List[Any] ) ->Any:
return self.model.generate(**_snake_case )[0]
def lowercase_ ( self : int, _snake_case : int ) ->str:
return self.pre_processor.decode(_snake_case, skip_special_tokens=_snake_case, clean_up_tokenization_spaces=_snake_case )
| 277 | 0 |
"""simple docstring"""
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_flax_available,
is_tf_available,
is_tokenizers_available,
is_torch_available,
is_vision_available,
)
a : Optional[Any] = {
'''configuration_clip''': [
'''CLIP_PRETRAINED_CONFIG_ARCHIVE_MAP''',
'''CLIPConfig''',
'''CLIPOnnxConfig''',
'''CLIPTextConfig''',
'''CLIPVisionConfig''',
],
'''processing_clip''': ['''CLIPProcessor'''],
'''tokenization_clip''': ['''CLIPTokenizer'''],
}
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a : Any = ['''CLIPTokenizerFast''']
try:
if not is_vision_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a : Any = ['''CLIPFeatureExtractor''']
a : List[Any] = ['''CLIPImageProcessor''']
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a : Any = [
'''CLIP_PRETRAINED_MODEL_ARCHIVE_LIST''',
'''CLIPModel''',
'''CLIPPreTrainedModel''',
'''CLIPTextModel''',
'''CLIPTextModelWithProjection''',
'''CLIPVisionModel''',
'''CLIPVisionModelWithProjection''',
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a : Any = [
'''TF_CLIP_PRETRAINED_MODEL_ARCHIVE_LIST''',
'''TFCLIPModel''',
'''TFCLIPPreTrainedModel''',
'''TFCLIPTextModel''',
'''TFCLIPVisionModel''',
]
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a : Tuple = [
'''FlaxCLIPModel''',
'''FlaxCLIPPreTrainedModel''',
'''FlaxCLIPTextModel''',
'''FlaxCLIPTextPreTrainedModel''',
'''FlaxCLIPVisionModel''',
'''FlaxCLIPVisionPreTrainedModel''',
]
if TYPE_CHECKING:
from .configuration_clip import (
CLIP_PRETRAINED_CONFIG_ARCHIVE_MAP,
CLIPConfig,
CLIPOnnxConfig,
CLIPTextConfig,
CLIPVisionConfig,
)
from .processing_clip import CLIPProcessor
from .tokenization_clip import CLIPTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_clip_fast import CLIPTokenizerFast
try:
if not is_vision_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .feature_extraction_clip import CLIPFeatureExtractor
from .image_processing_clip import CLIPImageProcessor
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_clip import (
CLIP_PRETRAINED_MODEL_ARCHIVE_LIST,
CLIPModel,
CLIPPreTrainedModel,
CLIPTextModel,
CLIPTextModelWithProjection,
CLIPVisionModel,
CLIPVisionModelWithProjection,
)
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_clip import (
TF_CLIP_PRETRAINED_MODEL_ARCHIVE_LIST,
TFCLIPModel,
TFCLIPPreTrainedModel,
TFCLIPTextModel,
TFCLIPVisionModel,
)
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_flax_clip import (
FlaxCLIPModel,
FlaxCLIPPreTrainedModel,
FlaxCLIPTextModel,
FlaxCLIPTextPreTrainedModel,
FlaxCLIPVisionModel,
FlaxCLIPVisionPreTrainedModel,
)
else:
import sys
a : Tuple = _LazyModule(__name__, globals()['''__file__'''], _import_structure, module_spec=__spec__)
| 105 |
import argparse
import torch
from transformers import LxmertConfig, LxmertForPreTraining, load_tf_weights_in_lxmert
from transformers.utils import logging
logging.set_verbosity_info()
def lowercase_ (A : str , A : List[Any] , A : Any ):
# Initialise PyTorch model
snake_case__ : List[Any] = LxmertConfig.from_json_file(A )
print(F'''Building PyTorch model from configuration: {config}''' )
snake_case__ : List[str] = LxmertForPreTraining(A )
# Load weights from tf checkpoint
load_tf_weights_in_lxmert(A , A , A )
# Save pytorch-model
print(F'''Save PyTorch model to {pytorch_dump_path}''' )
torch.save(model.state_dict() , A )
if __name__ == "__main__":
a_ :Union[str, Any] = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--tf_checkpoint_path", default=None, type=str, required=True, help="Path to the TensorFlow checkpoint path."
)
parser.add_argument(
"--config_file",
default=None,
type=str,
required=True,
help="The config json file corresponding to the pre-trained model. \nThis specifies the model architecture.",
)
parser.add_argument(
"--pytorch_dump_path", default=None, type=str, required=True, help="Path to the output PyTorch model."
)
a_ :Optional[int] = parser.parse_args()
convert_tf_checkpoint_to_pytorch(args.tf_checkpoint_path, args.config_file, args.pytorch_dump_path)
| 277 | 0 |
"""simple docstring"""
def __SCREAMING_SNAKE_CASE ( ):
return 1
def __SCREAMING_SNAKE_CASE ( A_ ):
return 0 if x < 0 else two_pence(x - 2 ) + one_pence()
def __SCREAMING_SNAKE_CASE ( A_ ):
return 0 if x < 0 else five_pence(x - 5 ) + two_pence(A_ )
def __SCREAMING_SNAKE_CASE ( A_ ):
return 0 if x < 0 else ten_pence(x - 10 ) + five_pence(A_ )
def __SCREAMING_SNAKE_CASE ( A_ ):
return 0 if x < 0 else twenty_pence(x - 20 ) + ten_pence(A_ )
def __SCREAMING_SNAKE_CASE ( A_ ):
return 0 if x < 0 else fifty_pence(x - 50 ) + twenty_pence(A_ )
def __SCREAMING_SNAKE_CASE ( A_ ):
return 0 if x < 0 else one_pound(x - 1_00 ) + fifty_pence(A_ )
def __SCREAMING_SNAKE_CASE ( A_ ):
return 0 if x < 0 else two_pound(x - 2_00 ) + one_pound(A_ )
def __SCREAMING_SNAKE_CASE ( A_ = 2_00 ):
return two_pound(A_ )
if __name__ == "__main__":
print(solution(int(input().strip())))
| 106 |
import argparse
import fairseq
import torch
from torch import nn
from transformers import (
MBartaaTokenizer,
MBartConfig,
MBartForCausalLM,
SpeechEncoderDecoderConfig,
SpeechEncoderDecoderModel,
WavaVecaConfig,
WavaVecaFeatureExtractor,
WavaVecaModel,
logging,
)
logging.set_verbosity_info()
a_ :Tuple = logging.get_logger(__name__)
a_ :List[Any] = {
"post_extract_proj": "feature_projection.projection",
"encoder.pos_conv.0": "encoder.pos_conv_embed.conv",
"self_attn.k_proj": "encoder.layers.*.attention.k_proj",
"self_attn.v_proj": "encoder.layers.*.attention.v_proj",
"self_attn.q_proj": "encoder.layers.*.attention.q_proj",
"self_attn.out_proj": "encoder.layers.*.attention.out_proj",
"self_attn_layer_norm": "encoder.layers.*.layer_norm",
"fc1": "encoder.layers.*.feed_forward.intermediate_dense",
"fc2": "encoder.layers.*.feed_forward.output_dense",
"final_layer_norm": "encoder.layers.*.final_layer_norm",
"encoder.layer_norm": "encoder.layer_norm",
"w2v_model.layer_norm": "feature_projection.layer_norm",
"quantizer.weight_proj": "quantizer.weight_proj",
"quantizer.vars": "quantizer.codevectors",
"project_q": "project_q",
"final_proj": "project_hid",
"w2v_encoder.proj": "lm_head",
"mask_emb": "masked_spec_embed",
}
a_ :Optional[int] = [
"lm_head",
"quantizer.weight_proj",
"quantizer.codevectors",
"project_q",
"project_hid",
]
def lowercase_ (A : Union[str, Any] , A : str , A : Dict , A : Optional[Any] , A : Optional[Any] ):
for attribute in key.split('.' ):
snake_case__ : Any = getattr(A , A )
if weight_type is not None:
snake_case__ : Optional[Any] = getattr(A , A ).shape
else:
snake_case__ : Optional[int] = hf_pointer.shape
assert hf_shape == value.shape, (
F'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
F''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
snake_case__ : Tuple = value
elif weight_type == "weight_g":
snake_case__ : Tuple = value
elif weight_type == "weight_v":
snake_case__ : List[Any] = value
elif weight_type == "bias":
snake_case__ : List[Any] = value
else:
snake_case__ : Optional[Any] = value
logger.info(F'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def lowercase_ (A : str , A : Any ):
snake_case__ : Union[str, Any] = []
snake_case__ : Union[str, Any] = fairseq_model.state_dict()
snake_case__ : Union[str, Any] = hf_model.feature_extractor
snake_case__ : Any = hf_model.adapter
for name, value in fairseq_dict.items():
snake_case__ : Any = False
if "conv_layers" in name:
load_conv_layer(
A , A , A , A , hf_model.config.feat_extract_norm == 'group' , )
snake_case__ : List[Any] = True
elif any(x in name for x in ['adaptor', 'w2v_encoder.proj.', 'w2v_proj_ln.'] ):
load_adapter(A , A , A , A )
snake_case__ : Optional[Any] = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split('w2v_model.' )[-1] == name.split('.' )[0]:
snake_case__ : Tuple = True
if "*" in mapped_key:
snake_case__ : List[Any] = name.split(A )[0].split('.' )[-2]
snake_case__ : Optional[int] = mapped_key.replace('*' , A )
if "weight_g" in name:
snake_case__ : Optional[int] = 'weight_g'
elif "weight_v" in name:
snake_case__ : Optional[Any] = 'weight_v'
elif "bias" in name:
snake_case__ : Union[str, Any] = 'bias'
elif "weight" in name:
snake_case__ : Optional[int] = 'weight'
else:
snake_case__ : Tuple = None
set_recursively(A , A , A , A , A )
continue
if not is_used:
unused_weights.append(A )
logger.warning(F'''Unused weights: {unused_weights}''' )
def lowercase_ (A : Union[str, Any] , A : Any , A : str , A : str , A : int ):
snake_case__ : str = full_name.split('conv_layers.' )[-1]
snake_case__ : Optional[int] = name.split('.' )
snake_case__ : Tuple = int(items[0] )
snake_case__ : Any = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
snake_case__ : Union[str, Any] = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
snake_case__ : Union[str, Any] = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
F'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
snake_case__ : Optional[int] = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
snake_case__ : Optional[Any] = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(A )
def lowercase_ (A : Optional[Any] , A : Any , A : Tuple , A : Any ):
snake_case__ : List[str] = full_name.split('adaptor.' )[-1]
snake_case__ : Tuple = name.split('.' )
if items[1].isdigit():
snake_case__ : Optional[int] = int(items[1] )
else:
snake_case__ : Any = None
if "adaptor" not in full_name:
if "proj_ln" in full_name:
# has to be layer norm
if "bias" in name:
assert (
value.shape == adapter.proj_layer_norm.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj_layer_norm.bias.data.shape} was found.'''
snake_case__ : List[Any] = value
logger.info(F'''Adapter proj layer norm bias was initialized from {full_name}.''' )
if "weight" in name:
assert (
value.shape == adapter.proj_layer_norm.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj_layer_norm.weight.data.shape} was found.'''
snake_case__ : int = value
else:
# has to be projection layer
if "bias" in name:
assert (
value.shape == adapter.proj.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj.bias.data.shape} was found.'''
snake_case__ : str = value
logger.info(F'''Adapter proj layer bias was initialized from {full_name}.''' )
if "weight" in name:
assert (
value.shape == adapter.proj.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj.weight.data.shape} was found.'''
snake_case__ : Dict = value
logger.info(F'''Adapter proj layer weight was initialized from {full_name}.''' )
elif isinstance(A , A ):
if "bias" in name:
assert (
value.shape == adapter.layers[layer_id].conv.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.layers[layer_id].conv.bias.data.shape} was found.'''
snake_case__ : List[str] = value
logger.info(F'''Adapter layer {layer_id} bias was initialized from {full_name}.''' )
elif "weight" in name:
assert (
value.shape == adapter.layers[layer_id].conv.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.layers[layer_id].conv.weight.data.shape} was found.'''
snake_case__ : List[str] = value
logger.info(F'''Adapter layer {layer_id} bias was initialized from {full_name}.''' )
else:
unused_weights.append(A )
def lowercase_ (A : int ):
snake_case__ , snake_case__ : Union[str, Any] = emb.weight.shape
snake_case__ : int = nn.Linear(A , A , bias=A )
snake_case__ : Optional[Any] = emb.weight.data
return lin_layer
@torch.no_grad()
def lowercase_ (A : Tuple , A : Tuple , A : Any , A : Optional[Any] , A : int , A : Optional[Any] , A : Union[str, Any] , A : Union[str, Any] , A : Optional[Any] , A : List[Any] , A : Union[str, Any] , ):
snake_case__ : Optional[Any] = WavaVecaConfig.from_pretrained(
A , add_adapter=A , adapter_stride=A , adapter_kernel_size=A , use_auth_token=A , output_hidden_size=A , )
snake_case__ : Dict = MBartConfig.from_pretrained(A )
# load model
snake_case__ , snake_case__ , snake_case__ : Any = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] , arg_overrides={
'config_yaml': config_yaml_path,
'data': '/'.join(dict_path.split('/' )[:-1] ),
'w2v_path': checkpoint_path,
'load_pretrained_decoder_from': None,
} , )
snake_case__ : List[Any] = model[0].eval()
# load feature extractor
snake_case__ : str = WavaVecaFeatureExtractor.from_pretrained(A , use_auth_token=A )
# set weights for wav2vec2 encoder
snake_case__ : List[str] = WavaVecaModel(A )
recursively_load_weights_wavaveca(model.encoder , A )
# load decoder weights
snake_case__ : Any = MBartForCausalLM(A )
snake_case__ , snake_case__ : int = hf_decoder.model.decoder.load_state_dict(model.decoder.state_dict() , strict=A )
logger.warning(F'''The following keys are missing when loading the decoder weights: {missing_keys}''' )
logger.warning(F'''The following keys are unexpected when loading the decoder weights: {unexpected_keys}''' )
snake_case__ : Union[str, Any] = SpeechEncoderDecoderModel(encoder=A , decoder=A )
snake_case__ : str = False
snake_case__ : int = MBartaaTokenizer(A )
tokenizer.save_pretrained(A )
snake_case__ : Any = hf_wavavec.config.to_dict()
snake_case__ : Tuple = tokenizer.pad_token_id
snake_case__ : Union[str, Any] = tokenizer.bos_token_id
snake_case__ : Dict = tokenizer.eos_token_id
snake_case__ : Optional[int] = 'mbart50'
snake_case__ : Union[str, Any] = 'wav2vec2'
snake_case__ : List[str] = tokenizer.eos_token_id
snake_case__ : Union[str, Any] = 2_5_0_0_0_4
snake_case__ : int = tokenizer.eos_token_id
snake_case__ : Union[str, Any] = SpeechEncoderDecoderConfig.from_dict(A )
hf_wavavec.save_pretrained(A )
feature_extractor.save_pretrained(A )
if __name__ == "__main__":
a_ :str = argparse.ArgumentParser()
parser.add_argument("--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model.")
parser.add_argument("--checkpoint_path", default=None, type=str, help="Path to fairseq checkpoint")
parser.add_argument("--dict_path", default=None, type=str, help="Path to dict of fine-tuned model")
parser.add_argument("--config_yaml_path", default=None, type=str, help="Path to yaml file of fine-tuned model")
parser.add_argument(
"--encoder_config_path",
default="facebook/wav2vec2-xls-r-1b",
type=str,
help="Path to hf encoder wav2vec2 checkpoint config",
)
parser.add_argument(
"--decoder_config_path",
default="facebook/mbart-large-50-one-to-many-mmt",
type=str,
help="Path to hf decoder checkpoint config",
)
parser.add_argument("--add_adapter", default=True, type=bool, help="whethere to add model adapter layers")
parser.add_argument("--adapter_stride", default=2, type=int, help="stride of adapter layers")
parser.add_argument("--adapter_kernel_size", default=3, type=int, help="kernel size of adapter layers")
parser.add_argument("--encoder_output_dim", default=1_024, type=int, help="encoder output dim")
parser.add_argument("--start_token_id", default=250_004, type=int, help="`decoder_start_token_id` of model config")
a_ :Union[str, Any] = parser.parse_args()
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.dict_path,
args.config_yaml_path,
encoder_config_path=args.encoder_config_path,
decoder_config_path=args.decoder_config_path,
add_adapter=args.add_adapter,
adapter_kernel_size=args.adapter_kernel_size,
adapter_stride=args.adapter_stride,
decoder_start_token_id=args.start_token_id,
encoder_output_dim=args.encoder_output_dim,
)
| 277 | 0 |
import datasets
import faiss
import numpy as np
import streamlit as st
import torch
from elasticsearch import Elasticsearch
from elia_utils import (
embed_questions_for_retrieval,
make_qa_sas_model,
qa_sas_generate,
query_es_index,
query_qa_dense_index,
)
import transformers
from transformers import AutoModel, AutoModelForSeqaSeqLM, AutoTokenizer
__lowerCAmelCase : Any = 'bart'
__lowerCAmelCase : List[str] = True
@st.cache(allow_output_mutation=A )
def __magic_name__ ( ):
'''simple docstring'''
if LOAD_DENSE_INDEX:
a = AutoTokenizer.from_pretrained("yjernite/retribert-base-uncased" )
a = AutoModel.from_pretrained("yjernite/retribert-base-uncased" ).to("cuda:0" )
a = qar_model.eval()
else:
a , a = (None, None)
if MODEL_TYPE == "bart":
a = AutoTokenizer.from_pretrained("yjernite/bart_eli5" )
a = AutoModelForSeqaSeqLM.from_pretrained("yjernite/bart_eli5" ).to("cuda:0" )
a = torch.load("seq2seq_models/eli5_bart_model_blm_2.pth" )
sas_model.load_state_dict(save_dict["model"] )
a = sas_model.eval()
else:
a , a = make_qa_sas_model(
model_name="t5-small", from_file="seq2seq_models/eli5_t5_model_1024_4.pth", device="cuda:0" )
return (qar_tokenizer, qar_model, sas_tokenizer, sas_model)
@st.cache(allow_output_mutation=A )
def __magic_name__ ( ):
'''simple docstring'''
if LOAD_DENSE_INDEX:
a = faiss.StandardGpuResources()
a = datasets.load_dataset(path="wiki_snippets", name="wiki40b_en_100_0" )["train"]
a = np.memmap(
"wiki40b_passages_reps_32_l-8_h-768_b-512-512.dat", dtype="float32", mode="r", shape=(wikiaab_passages.num_rows, 128), )
a = faiss.IndexFlatIP(128 )
a = faiss.index_cpu_to_gpu(A, 1, A )
wikiaab_gpu_index_flat.add(A ) # TODO fix for larger GPU
else:
a , a = (None, None)
a = Elasticsearch([{"host": "localhost", "port": "9200"}] )
return (wikiaab_passages, wikiaab_gpu_index_flat, es_client)
@st.cache(allow_output_mutation=A )
def __magic_name__ ( ):
'''simple docstring'''
a = datasets.load_dataset("eli5", name="LFQA_reddit" )
a = elia["train_eli5"]
a = np.memmap(
"eli5_questions_reps.dat", dtype="float32", mode="r", shape=(elia_train.num_rows, 128) )
a = faiss.IndexFlatIP(128 )
eli5_train_q_index.add(A )
return (elia_train, eli5_train_q_index)
__lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase : int = load_indexes()
__lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase , __lowerCAmelCase : Any = load_models()
__lowerCAmelCase , __lowerCAmelCase : List[Any] = load_train_data()
def __magic_name__ ( A : Optional[int], A : Any=10 ):
'''simple docstring'''
a = embed_questions_for_retrieval([question], A, A )
a , a = eli5_train_q_index.search(A, A )
a = [elia_train[int(A )] for i in I[0]]
return nn_examples
def __magic_name__ ( A : Tuple, A : Optional[Any]="wiki40b", A : Union[str, Any]="dense", A : Dict=10 ):
'''simple docstring'''
if source == "none":
a , a = (" <P> ".join(["" for _ in range(11 )] ).strip(), [])
else:
if method == "dense":
a , a = query_qa_dense_index(
A, A, A, A, A, A )
else:
a , a = query_es_index(
A, A, index_name="english_wiki40b_snippets_100w", n_results=A, )
a = [
(res["article_title"], res["section_title"].strip(), res["score"], res["passage_text"]) for res in hit_lst
]
a = "question: {} context: {}".format(A, A )
return question_doc, support_list
@st.cache(
hash_funcs={
torch.Tensor: (lambda A : None),
transformers.models.bart.tokenization_bart.BartTokenizer: (lambda A : None),
} )
def __magic_name__ ( A : Optional[Any], A : int, A : str, A : Any=64, A : List[str]=256, A : Union[str, Any]=False, A : Optional[Any]=2, A : Tuple=0.95, A : List[str]=0.8 ):
'''simple docstring'''
with torch.no_grad():
a = qa_sas_generate(
A, A, A, num_answers=1, num_beams=A, min_len=A, max_len=A, do_sample=A, temp=A, top_p=A, top_k=A, max_input_length=1024, device="cuda:0", )[0]
return (answer, support_list)
st.title('Long Form Question Answering with ELI5')
# Start sidebar
__lowerCAmelCase : Tuple = '<img src=\'https://huggingface.co/front/assets/huggingface_logo.svg\'>'
__lowerCAmelCase : Optional[Any] = '\n<html>\n <head>\n <style>\n .img-container {\n padding-left: 90px;\n padding-right: 90px;\n padding-top: 50px;\n padding-bottom: 50px;\n background-color: #f0f3f9;\n }\n </style>\n </head>\n <body>\n <span class="img-container"> <!-- Inline parent element -->\n %s\n </span>\n </body>\n</html>\n' % (
header_html,
)
st.sidebar.markdown(
header_full,
unsafe_allow_html=True,
)
# Long Form QA with ELI5 and Wikipedia
__lowerCAmelCase : List[str] = '\nThis demo presents a model trained to [provide long-form answers to open-domain questions](https://yjernite.github.io/lfqa.html).\nFirst, a document retriever fetches a set of relevant Wikipedia passages given the question from the [Wiki40b](https://research.google/pubs/pub49029/) dataset,\na pre-processed fixed snapshot of Wikipedia.\n'
st.sidebar.markdown(description, unsafe_allow_html=True)
__lowerCAmelCase : Any = [
'Answer the question',
'View the retrieved document only',
'View the most similar ELI5 question and answer',
'Show me everything, please!',
]
__lowerCAmelCase : Optional[int] = st.sidebar.checkbox('Demo options')
if demo_options:
__lowerCAmelCase : List[Any] = st.sidebar.selectbox(
'',
action_list,
index=3,
)
__lowerCAmelCase : Optional[Any] = action_list.index(action_st)
__lowerCAmelCase : Optional[int] = st.sidebar.selectbox(
'',
['Show full text of passages', 'Show passage section titles'],
index=0,
)
__lowerCAmelCase : Optional[int] = show_type == 'Show full text of passages'
else:
__lowerCAmelCase : Union[str, Any] = 3
__lowerCAmelCase : List[Any] = True
__lowerCAmelCase : List[str] = st.sidebar.checkbox('Retrieval options')
if retrieval_options:
__lowerCAmelCase : Tuple = '\n ### Information retriever options\n\n The **sparse** retriever uses ElasticSearch, while the **dense** retriever uses max-inner-product search between a question and passage embedding\n trained using the [ELI5](https://arxiv.org/abs/1907.09190) questions-answer pairs.\n The answer is then generated by sequence to sequence model which takes the question and retrieved document as input.\n '
st.sidebar.markdown(retriever_info)
__lowerCAmelCase : List[str] = st.sidebar.selectbox('Which Wikipedia format should the model use?', ['wiki40b', 'none'])
__lowerCAmelCase : Optional[Any] = st.sidebar.selectbox('Which Wikipedia indexer should the model use?', ['dense', 'sparse', 'mixed'])
else:
__lowerCAmelCase : int = 'wiki40b'
__lowerCAmelCase : str = 'dense'
__lowerCAmelCase : List[str] = 'beam'
__lowerCAmelCase : Optional[Any] = 2
__lowerCAmelCase : Optional[int] = 64
__lowerCAmelCase : List[str] = 256
__lowerCAmelCase : List[Any] = None
__lowerCAmelCase : Optional[Any] = None
__lowerCAmelCase : str = st.sidebar.checkbox('Generation options')
if generate_options:
__lowerCAmelCase : Optional[int] = '\n ### Answer generation options\n\n The sequence-to-sequence model was initialized with [BART](https://huggingface.co/facebook/bart-large)\n weights and fine-tuned on the ELI5 QA pairs and retrieved documents. You can use the model for greedy decoding with\n **beam** search, or **sample** from the decoder\'s output probabilities.\n '
st.sidebar.markdown(generate_info)
__lowerCAmelCase : Optional[Any] = st.sidebar.selectbox('Would you like to use beam search or sample an answer?', ['beam', 'sampled'])
__lowerCAmelCase : List[Any] = st.sidebar.slider(
'Minimum generation length', min_value=8, max_value=256, value=64, step=8, format=None, key=None
)
__lowerCAmelCase : Dict = st.sidebar.slider(
'Maximum generation length', min_value=64, max_value=512, value=256, step=16, format=None, key=None
)
if sampled == "beam":
__lowerCAmelCase : Dict = st.sidebar.slider('Beam size', min_value=1, max_value=8, value=2, step=None, format=None, key=None)
else:
__lowerCAmelCase : Optional[Any] = st.sidebar.slider(
'Nucleus sampling p', min_value=0.1, max_value=1.0, value=0.9_5, step=0.0_1, format=None, key=None
)
__lowerCAmelCase : Optional[int] = st.sidebar.slider(
'Temperature', min_value=0.1, max_value=1.0, value=0.7, step=0.0_1, format=None, key=None
)
__lowerCAmelCase : Tuple = None
# start main text
__lowerCAmelCase : List[str] = [
'<MY QUESTION>',
'How do people make chocolate?',
'Why do we get a fever when we are sick?',
'How can different animals perceive different colors?',
'What is natural language processing?',
'What\'s the best way to treat a sunburn?',
'What exactly are vitamins ?',
'How does nuclear energy provide electricity?',
'What\'s the difference between viruses and bacteria?',
'Why are flutes classified as woodwinds when most of them are made out of metal ?',
'Why do people like drinking coffee even though it tastes so bad?',
'What happens when wine ages? How does it make the wine taste better?',
'If an animal is an herbivore, where does it get the protein that it needs to survive if it only eats grass?',
'How can we set a date to the beginning or end of an artistic period? Doesn\'t the change happen gradually?',
'How does New Zealand have so many large bird predators?',
]
__lowerCAmelCase : List[Any] = st.selectbox(
'What would you like to ask? ---- select <MY QUESTION> to enter a new query',
questions_list,
index=1,
)
if question_s == "<MY QUESTION>":
__lowerCAmelCase : Optional[Any] = st.text_input('Enter your question here:', '')
else:
__lowerCAmelCase : Union[str, Any] = question_s
if st.button('Show me!'):
if action in [0, 1, 3]:
if index_type == "mixed":
__lowerCAmelCase , __lowerCAmelCase : int = make_support(question, source=wiki_source, method='dense', n_results=10)
__lowerCAmelCase , __lowerCAmelCase : Union[str, Any] = make_support(question, source=wiki_source, method='sparse', n_results=10)
__lowerCAmelCase : int = []
for res_d, res_s in zip(support_list_dense, support_list_sparse):
if tuple(res_d) not in support_list:
support_list += [tuple(res_d)]
if tuple(res_s) not in support_list:
support_list += [tuple(res_s)]
__lowerCAmelCase : List[str] = support_list[:10]
__lowerCAmelCase : List[str] = '<P> ' + ' <P> '.join([res[-1] for res in support_list])
else:
__lowerCAmelCase , __lowerCAmelCase : Any = make_support(question, source=wiki_source, method=index_type, n_results=10)
if action in [0, 3]:
__lowerCAmelCase , __lowerCAmelCase : Any = answer_question(
question_doc,
sas_model,
sas_tokenizer,
min_len=min_len,
max_len=int(max_len),
sampling=(sampled == 'sampled'),
n_beams=n_beams,
top_p=top_p,
temp=temp,
)
st.markdown('### The model generated answer is:')
st.write(answer)
if action in [0, 1, 3] and wiki_source != "none":
st.markdown('--- \n ### The model is drawing information from the following Wikipedia passages:')
for i, res in enumerate(support_list):
__lowerCAmelCase : Union[str, Any] = 'https://en.wikipedia.org/wiki/{}'.format(res[0].replace(' ', '_'))
__lowerCAmelCase : Tuple = res[1].strip()
if sec_titles == "":
__lowerCAmelCase : Any = '[{}]({})'.format(res[0], wiki_url)
else:
__lowerCAmelCase : Tuple = sec_titles.split(' & ')
__lowerCAmelCase : Any = ' & '.join(
['[{}]({}#{})'.format(sec.strip(), wiki_url, sec.strip().replace(' ', '_')) for sec in sec_list]
)
st.markdown(
'{0:02d} - **Article**: {1:<18} <br> _Section_: {2}'.format(i + 1, res[0], sections),
unsafe_allow_html=True,
)
if show_passages:
st.write(
'> <span style="font-family:arial; font-size:10pt;">' + res[-1] + '</span>', unsafe_allow_html=True
)
if action in [2, 3]:
__lowerCAmelCase : int = find_nearest_training(question)
__lowerCAmelCase : List[Any] = nn_train_list[0]
st.markdown(
'--- \n ### The most similar question in the ELI5 training set was: \n\n {}'.format(train_exple['title'])
)
__lowerCAmelCase : str = [
'{}. {}'.format(i + 1, ' \n'.join([line.strip() for line in ans.split('\n') if line.strip() != '']))
for i, (ans, sc) in enumerate(zip(train_exple['answers']['text'], train_exple['answers']['score']))
if i == 0 or sc > 2
]
st.markdown('##### Its answers were: \n\n {}'.format('\n'.join(answers_st)))
__lowerCAmelCase : Dict = '\n---\n\n**Disclaimer**\n\n*The intent of this app is to provide some (hopefully entertaining) insights into the behavior of a current LFQA system.\nEvaluating biases of such a model and ensuring factual generations are still very much open research problems.\nTherefore, until some significant progress is achieved, we caution against using the generated answers for practical purposes.*\n'
st.sidebar.markdown(disclaimer, unsafe_allow_html=True)
| 107 |
from collections import OrderedDict
from typing import TYPE_CHECKING, Any, Mapping, Optional, Union
from ...configuration_utils import PretrainedConfig
from ...onnx import OnnxConfig
from ...utils import logging
if TYPE_CHECKING:
from ... import FeatureExtractionMixin, PreTrainedTokenizerBase, TensorType
a_ :Tuple = logging.get_logger(__name__)
a_ :Union[str, Any] = {
"microsoft/deberta-v2-xlarge": "https://huggingface.co/microsoft/deberta-v2-xlarge/resolve/main/config.json",
"microsoft/deberta-v2-xxlarge": "https://huggingface.co/microsoft/deberta-v2-xxlarge/resolve/main/config.json",
"microsoft/deberta-v2-xlarge-mnli": (
"https://huggingface.co/microsoft/deberta-v2-xlarge-mnli/resolve/main/config.json"
),
"microsoft/deberta-v2-xxlarge-mnli": (
"https://huggingface.co/microsoft/deberta-v2-xxlarge-mnli/resolve/main/config.json"
),
}
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """deberta-v2"""
def __init__( self : Union[str, Any], _snake_case : Dict=1_2_8_1_0_0, _snake_case : Any=1_5_3_6, _snake_case : Tuple=2_4, _snake_case : int=2_4, _snake_case : Optional[int]=6_1_4_4, _snake_case : Optional[int]="gelu", _snake_case : Optional[int]=0.1, _snake_case : List[str]=0.1, _snake_case : str=5_1_2, _snake_case : Optional[int]=0, _snake_case : Optional[int]=0.0_2, _snake_case : Dict=1e-7, _snake_case : int=False, _snake_case : Any=-1, _snake_case : List[str]=0, _snake_case : Tuple=True, _snake_case : Any=None, _snake_case : Union[str, Any]=0, _snake_case : Tuple="gelu", **_snake_case : Union[str, Any], ) ->Optional[int]:
super().__init__(**_snake_case )
snake_case__ : Dict = hidden_size
snake_case__ : Optional[int] = num_hidden_layers
snake_case__ : Any = num_attention_heads
snake_case__ : List[Any] = intermediate_size
snake_case__ : List[Any] = hidden_act
snake_case__ : Union[str, Any] = hidden_dropout_prob
snake_case__ : Dict = attention_probs_dropout_prob
snake_case__ : List[str] = max_position_embeddings
snake_case__ : List[str] = type_vocab_size
snake_case__ : Optional[Any] = initializer_range
snake_case__ : Optional[int] = relative_attention
snake_case__ : Tuple = max_relative_positions
snake_case__ : Union[str, Any] = pad_token_id
snake_case__ : Optional[int] = position_biased_input
# Backwards compatibility
if type(_snake_case ) == str:
snake_case__ : int = [x.strip() for x in pos_att_type.lower().split('|' )]
snake_case__ : List[str] = pos_att_type
snake_case__ : Union[str, Any] = vocab_size
snake_case__ : Optional[int] = layer_norm_eps
snake_case__ : Optional[int] = kwargs.get('pooler_hidden_size', _snake_case )
snake_case__ : int = pooler_dropout
snake_case__ : str = pooler_hidden_act
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
@property
def lowercase_ ( self : Optional[int] ) ->Mapping[str, Mapping[int, str]]:
if self.task == "multiple-choice":
snake_case__ : List[Any] = {0: 'batch', 1: 'choice', 2: 'sequence'}
else:
snake_case__ : int = {0: 'batch', 1: 'sequence'}
if self._config.type_vocab_size > 0:
return OrderedDict(
[('input_ids', dynamic_axis), ('attention_mask', dynamic_axis), ('token_type_ids', dynamic_axis)] )
else:
return OrderedDict([('input_ids', dynamic_axis), ('attention_mask', dynamic_axis)] )
@property
def lowercase_ ( self : Dict ) ->int:
return 1_2
def lowercase_ ( self : Tuple, _snake_case : Union["PreTrainedTokenizerBase", "FeatureExtractionMixin"], _snake_case : int = -1, _snake_case : int = -1, _snake_case : int = -1, _snake_case : bool = False, _snake_case : Optional["TensorType"] = None, _snake_case : int = 3, _snake_case : int = 4_0, _snake_case : int = 4_0, _snake_case : "PreTrainedTokenizerBase" = None, ) ->Mapping[str, Any]:
snake_case__ : Union[str, Any] = super().generate_dummy_inputs(preprocessor=_snake_case, framework=_snake_case )
if self._config.type_vocab_size == 0 and "token_type_ids" in dummy_inputs:
del dummy_inputs["token_type_ids"]
return dummy_inputs
| 277 | 0 |
"""simple docstring"""
import unittest
from transformers import BigBirdTokenizer, BigBirdTokenizerFast
from transformers.testing_utils import get_tests_dir, require_sentencepiece, require_tokenizers, require_torch, slow
from transformers.utils import cached_property
from ...test_tokenization_common import TokenizerTesterMixin
lowerCAmelCase__ = '''▁'''
lowerCAmelCase__ = get_tests_dir('''fixtures/test_sentencepiece.model''')
@require_sentencepiece
@require_tokenizers
class SCREAMING_SNAKE_CASE__ ( lowercase , unittest.TestCase ):
"""simple docstring"""
a : str =BigBirdTokenizer
a : Union[str, Any] =BigBirdTokenizerFast
a : Tuple =True
a : Any =True
def lowercase__ ( self ):
"""simple docstring"""
super().setUp()
lowerCAmelCase : str = self.tokenizer_class(snake_case__ , keep_accents=snake_case__ )
tokenizer.save_pretrained(self.tmpdirname )
def lowercase__ ( self ):
"""simple docstring"""
lowerCAmelCase : str = "<s>"
lowerCAmelCase : Optional[int] = 1
self.assertEqual(self.get_tokenizer()._convert_token_to_id(snake_case__ ) , snake_case__ )
self.assertEqual(self.get_tokenizer()._convert_id_to_token(snake_case__ ) , snake_case__ )
def lowercase__ ( self ):
"""simple docstring"""
lowerCAmelCase : Dict = list(self.get_tokenizer().get_vocab().keys() )
self.assertEqual(vocab_keys[0] , "<unk>" )
self.assertEqual(vocab_keys[1] , "<s>" )
self.assertEqual(vocab_keys[-1] , "[MASK]" )
self.assertEqual(len(snake_case__ ) , 1_004 )
def lowercase__ ( self ):
"""simple docstring"""
self.assertEqual(self.get_tokenizer().vocab_size , 1_000 )
def lowercase__ ( self ):
"""simple docstring"""
if not self.test_rust_tokenizer:
return
lowerCAmelCase : Tuple = self.get_tokenizer()
lowerCAmelCase : Optional[int] = self.get_rust_tokenizer()
lowerCAmelCase : Tuple = "I was born in 92000, and this is falsé."
lowerCAmelCase : Optional[int] = tokenizer.tokenize(snake_case__ )
lowerCAmelCase : int = rust_tokenizer.tokenize(snake_case__ )
self.assertListEqual(snake_case__ , snake_case__ )
lowerCAmelCase : Tuple = tokenizer.encode(snake_case__ , add_special_tokens=snake_case__ )
lowerCAmelCase : int = rust_tokenizer.encode(snake_case__ , add_special_tokens=snake_case__ )
self.assertListEqual(snake_case__ , snake_case__ )
lowerCAmelCase : List[Any] = self.get_rust_tokenizer()
lowerCAmelCase : Tuple = tokenizer.encode(snake_case__ )
lowerCAmelCase : List[Any] = rust_tokenizer.encode(snake_case__ )
self.assertListEqual(snake_case__ , snake_case__ )
def lowercase__ ( self ):
"""simple docstring"""
lowerCAmelCase : Any = BigBirdTokenizer(snake_case__ , keep_accents=snake_case__ )
lowerCAmelCase : Union[str, Any] = tokenizer.tokenize("This is a test" )
self.assertListEqual(snake_case__ , ["▁This", "▁is", "▁a", "▁t", "est"] )
self.assertListEqual(
tokenizer.convert_tokens_to_ids(snake_case__ ) , [285, 46, 10, 170, 382] , )
lowerCAmelCase : Any = tokenizer.tokenize("I was born in 92000, and this is falsé." )
self.assertListEqual(
snake_case__ , [
SPIECE_UNDERLINE + "I",
SPIECE_UNDERLINE + "was",
SPIECE_UNDERLINE + "b",
"or",
"n",
SPIECE_UNDERLINE + "in",
SPIECE_UNDERLINE + "",
"9",
"2",
"0",
"0",
"0",
",",
SPIECE_UNDERLINE + "and",
SPIECE_UNDERLINE + "this",
SPIECE_UNDERLINE + "is",
SPIECE_UNDERLINE + "f",
"al",
"s",
"é",
".",
] , )
lowerCAmelCase : str = tokenizer.convert_tokens_to_ids(snake_case__ )
self.assertListEqual(
snake_case__ , [8, 21, 84, 55, 24, 19, 7, 0, 602, 347, 347, 347, 3, 12, 66, 46, 72, 80, 6, 0, 4] , )
lowerCAmelCase : List[str] = tokenizer.convert_ids_to_tokens(snake_case__ )
self.assertListEqual(
snake_case__ , [
SPIECE_UNDERLINE + "I",
SPIECE_UNDERLINE + "was",
SPIECE_UNDERLINE + "b",
"or",
"n",
SPIECE_UNDERLINE + "in",
SPIECE_UNDERLINE + "",
"<unk>",
"2",
"0",
"0",
"0",
",",
SPIECE_UNDERLINE + "and",
SPIECE_UNDERLINE + "this",
SPIECE_UNDERLINE + "is",
SPIECE_UNDERLINE + "f",
"al",
"s",
"<unk>",
".",
] , )
@cached_property
def lowercase__ ( self ):
"""simple docstring"""
return BigBirdTokenizer.from_pretrained("google/bigbird-roberta-base" )
@slow
def lowercase__ ( self ):
"""simple docstring"""
lowerCAmelCase : List[Any] = "Hello World!"
lowerCAmelCase : Any = [65, 18_536, 2_260, 101, 66]
self.assertListEqual(snake_case__ , self.big_tokenizer.encode(snake_case__ ) )
@slow
def lowercase__ ( self ):
"""simple docstring"""
lowerCAmelCase : Union[str, Any] = (
"This is a very long text with a lot of weird characters, such as: . , ~ ? ( ) \" [ ] ! : - . Also we will"
" add words that should not exsist and be tokenized to <unk>, such as saoneuhaoesuth"
)
# fmt: off
lowerCAmelCase : List[str] = [65, 871, 419, 358, 946, 991, 2_521, 452, 358, 1_357, 387, 7_751, 3_536, 112, 985, 456, 126, 865, 938, 5_400, 5_734, 458, 1_368, 467, 786, 2_462, 5_246, 1_159, 633, 865, 4_519, 457, 582, 852, 2_557, 427, 916, 508, 405, 34_324, 497, 391, 408, 11_342, 1_244, 385, 100, 938, 985, 456, 574, 362, 12_597, 3_200, 3_129, 1_172, 66] # noqa: E231
# fmt: on
self.assertListEqual(snake_case__ , self.big_tokenizer.encode(snake_case__ ) )
@require_torch
@slow
def lowercase__ ( self ):
"""simple docstring"""
import torch
from transformers import BigBirdConfig, BigBirdModel
# Build sequence
lowerCAmelCase : Dict = list(self.big_tokenizer.get_vocab().keys() )[:10]
lowerCAmelCase : int = " ".join(snake_case__ )
lowerCAmelCase : Dict = self.big_tokenizer.encode_plus(snake_case__ , return_tensors="pt" , return_token_type_ids=snake_case__ )
lowerCAmelCase : Any = self.big_tokenizer.batch_encode_plus(
[sequence + " " + sequence] , return_tensors="pt" , return_token_type_ids=snake_case__ )
lowerCAmelCase : str = BigBirdConfig(attention_type="original_full" )
lowerCAmelCase : Any = BigBirdModel(snake_case__ )
assert model.get_input_embeddings().weight.shape[0] >= self.big_tokenizer.vocab_size
with torch.no_grad():
model(**snake_case__ )
model(**snake_case__ )
@slow
def lowercase__ ( self ):
"""simple docstring"""
lowerCAmelCase : List[str] = BigBirdTokenizer.from_pretrained("google/bigbird-roberta-base" )
lowerCAmelCase : Union[str, Any] = tokenizer.decode(tokenizer("Paris is the [MASK]." ).input_ids )
self.assertTrue(decoded_text == "[CLS] Paris is the[MASK].[SEP]" )
@slow
def lowercase__ ( self ):
"""simple docstring"""
lowerCAmelCase : Any = {"input_ids": [[65, 39_286, 458, 36_335, 2_001, 456, 13_073, 13_266, 455, 113, 7_746, 1_741, 11_157, 391, 13_073, 13_266, 455, 113, 3_967, 35_412, 113, 4_936, 109, 3_870, 2_377, 113, 30_084, 45_720, 458, 134, 17_496, 112, 503, 11_672, 113, 118, 112, 5_665, 13_347, 38_687, 112, 1_496, 31_389, 112, 3_268, 47_264, 134, 962, 112, 16_377, 8_035, 23_130, 430, 12_169, 15_518, 28_592, 458, 146, 41_697, 109, 391, 12_169, 15_518, 16_689, 458, 146, 41_358, 109, 452, 726, 4_034, 111, 763, 35_412, 5_082, 388, 1_903, 111, 9_051, 391, 2_870, 48_918, 1_900, 1_123, 550, 998, 112, 9_586, 15_985, 455, 391, 410, 22_955, 37_636, 114, 66], [65, 448, 17_496, 419, 3_663, 385, 763, 113, 27_533, 2_870, 3_283, 13_043, 1_639, 24_713, 523, 656, 24_013, 18_550, 2_521, 517, 27_014, 21_244, 420, 1_212, 1_465, 391, 927, 4_833, 388, 578, 11_786, 114, 66, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [65, 484, 2_169, 7_687, 21_932, 18_146, 726, 363, 17_032, 3_391, 114, 66, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]], "attention_mask": [[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]} # noqa: E501
# fmt: on
self.tokenizer_integration_test_util(
expected_encoding=snake_case__ , model_name="google/bigbird-roberta-base" , revision="215c99f1600e06f83acce68422f2035b2b5c3510" , )
| 108 |
import argparse
import json
import pickle
from pathlib import Path
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import MaskFormerConfig, MaskFormerForInstanceSegmentation, MaskFormerImageProcessor, SwinConfig
from transformers.utils import logging
logging.set_verbosity_info()
a_ :str = logging.get_logger(__name__)
def lowercase_ (A : str ):
snake_case__ : Tuple = SwinConfig.from_pretrained(
'microsoft/swin-tiny-patch4-window7-224' , out_features=['stage1', 'stage2', 'stage3', 'stage4'] )
snake_case__ : List[Any] = MaskFormerConfig(backbone_config=A )
snake_case__ : Union[str, Any] = 'huggingface/label-files'
if "ade20k-full" in model_name:
# this should be ok
snake_case__ : Dict = 8_4_7
snake_case__ : List[str] = 'maskformer-ade20k-full-id2label.json'
elif "ade" in model_name:
# this should be ok
snake_case__ : Union[str, Any] = 1_5_0
snake_case__ : Any = 'ade20k-id2label.json'
elif "coco-stuff" in model_name:
# this should be ok
snake_case__ : List[str] = 1_7_1
snake_case__ : Union[str, Any] = 'maskformer-coco-stuff-id2label.json'
elif "coco" in model_name:
# TODO
snake_case__ : Dict = 1_3_3
snake_case__ : str = 'coco-panoptic-id2label.json'
elif "cityscapes" in model_name:
# this should be ok
snake_case__ : List[str] = 1_9
snake_case__ : Union[str, Any] = 'cityscapes-id2label.json'
elif "vistas" in model_name:
# this should be ok
snake_case__ : Tuple = 6_5
snake_case__ : List[str] = 'mapillary-vistas-id2label.json'
snake_case__ : Dict = json.load(open(hf_hub_download(A , A , repo_type='dataset' ) , 'r' ) )
snake_case__ : List[str] = {int(A ): v for k, v in idalabel.items()}
return config
def lowercase_ (A : Any ):
snake_case__ : Optional[int] = []
# stem
# fmt: off
rename_keys.append(('backbone.patch_embed.proj.weight', 'model.pixel_level_module.encoder.model.embeddings.patch_embeddings.projection.weight') )
rename_keys.append(('backbone.patch_embed.proj.bias', 'model.pixel_level_module.encoder.model.embeddings.patch_embeddings.projection.bias') )
rename_keys.append(('backbone.patch_embed.norm.weight', 'model.pixel_level_module.encoder.model.embeddings.norm.weight') )
rename_keys.append(('backbone.patch_embed.norm.bias', 'model.pixel_level_module.encoder.model.embeddings.norm.bias') )
# stages
for i in range(len(config.backbone_config.depths ) ):
for j in range(config.backbone_config.depths[i] ):
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm1.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_before.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm1.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_before.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.relative_position_bias_table''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.self.relative_position_bias_table''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.relative_position_index''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.self.relative_position_index''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.proj.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.output.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.proj.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.output.dense.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm2.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_after.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm2.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_after.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc1.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.intermediate.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc1.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.intermediate.dense.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc2.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.output.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc2.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.output.dense.bias''') )
if i < 3:
rename_keys.append((F'''backbone.layers.{i}.downsample.reduction.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.reduction.weight''') )
rename_keys.append((F'''backbone.layers.{i}.downsample.norm.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.norm.weight''') )
rename_keys.append((F'''backbone.layers.{i}.downsample.norm.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.norm.bias''') )
rename_keys.append((F'''backbone.norm{i}.weight''', F'''model.pixel_level_module.encoder.hidden_states_norms.{i}.weight''') )
rename_keys.append((F'''backbone.norm{i}.bias''', F'''model.pixel_level_module.encoder.hidden_states_norms.{i}.bias''') )
# FPN
rename_keys.append(('sem_seg_head.layer_4.weight', 'model.pixel_level_module.decoder.fpn.stem.0.weight') )
rename_keys.append(('sem_seg_head.layer_4.norm.weight', 'model.pixel_level_module.decoder.fpn.stem.1.weight') )
rename_keys.append(('sem_seg_head.layer_4.norm.bias', 'model.pixel_level_module.decoder.fpn.stem.1.bias') )
for source_index, target_index in zip(range(3 , 0 , -1 ) , range(0 , 3 ) ):
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.0.weight''') )
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.norm.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.1.weight''') )
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.norm.bias''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.1.bias''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.0.weight''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.norm.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.1.weight''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.norm.bias''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.1.bias''') )
rename_keys.append(('sem_seg_head.mask_features.weight', 'model.pixel_level_module.decoder.mask_projection.weight') )
rename_keys.append(('sem_seg_head.mask_features.bias', 'model.pixel_level_module.decoder.mask_projection.bias') )
# Transformer decoder
for idx in range(config.decoder_config.decoder_layers ):
# self-attention out projection
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.out_proj.weight''', F'''model.transformer_module.decoder.layers.{idx}.self_attn.out_proj.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.out_proj.bias''', F'''model.transformer_module.decoder.layers.{idx}.self_attn.out_proj.bias''') )
# cross-attention out projection
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.out_proj.weight''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn.out_proj.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.out_proj.bias''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn.out_proj.bias''') )
# MLP 1
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear1.weight''', F'''model.transformer_module.decoder.layers.{idx}.fc1.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear1.bias''', F'''model.transformer_module.decoder.layers.{idx}.fc1.bias''') )
# MLP 2
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear2.weight''', F'''model.transformer_module.decoder.layers.{idx}.fc2.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear2.bias''', F'''model.transformer_module.decoder.layers.{idx}.fc2.bias''') )
# layernorm 1 (self-attention layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm1.weight''', F'''model.transformer_module.decoder.layers.{idx}.self_attn_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm1.bias''', F'''model.transformer_module.decoder.layers.{idx}.self_attn_layer_norm.bias''') )
# layernorm 2 (cross-attention layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm2.weight''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm2.bias''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn_layer_norm.bias''') )
# layernorm 3 (final layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm3.weight''', F'''model.transformer_module.decoder.layers.{idx}.final_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm3.bias''', F'''model.transformer_module.decoder.layers.{idx}.final_layer_norm.bias''') )
rename_keys.append(('sem_seg_head.predictor.transformer.decoder.norm.weight', 'model.transformer_module.decoder.layernorm.weight') )
rename_keys.append(('sem_seg_head.predictor.transformer.decoder.norm.bias', 'model.transformer_module.decoder.layernorm.bias') )
# heads on top
rename_keys.append(('sem_seg_head.predictor.query_embed.weight', 'model.transformer_module.queries_embedder.weight') )
rename_keys.append(('sem_seg_head.predictor.input_proj.weight', 'model.transformer_module.input_projection.weight') )
rename_keys.append(('sem_seg_head.predictor.input_proj.bias', 'model.transformer_module.input_projection.bias') )
rename_keys.append(('sem_seg_head.predictor.class_embed.weight', 'class_predictor.weight') )
rename_keys.append(('sem_seg_head.predictor.class_embed.bias', 'class_predictor.bias') )
for i in range(3 ):
rename_keys.append((F'''sem_seg_head.predictor.mask_embed.layers.{i}.weight''', F'''mask_embedder.{i}.0.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.mask_embed.layers.{i}.bias''', F'''mask_embedder.{i}.0.bias''') )
# fmt: on
return rename_keys
def lowercase_ (A : Tuple , A : Tuple , A : Optional[Any] ):
snake_case__ : Optional[int] = dct.pop(A )
snake_case__ : Union[str, Any] = val
def lowercase_ (A : Optional[Any] , A : Tuple ):
snake_case__ : Optional[int] = [int(backbone_config.embed_dim * 2**i ) for i in range(len(backbone_config.depths ) )]
for i in range(len(backbone_config.depths ) ):
snake_case__ : Optional[int] = num_features[i]
for j in range(backbone_config.depths[i] ):
# fmt: off
# read in weights + bias of input projection layer (in original implementation, this is a single matrix + bias)
snake_case__ : int = state_dict.pop(F'''backbone.layers.{i}.blocks.{j}.attn.qkv.weight''' )
snake_case__ : Tuple = state_dict.pop(F'''backbone.layers.{i}.blocks.{j}.attn.qkv.bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : str = in_proj_weight[:dim, :]
snake_case__ : int = in_proj_bias[: dim]
snake_case__ : List[Any] = in_proj_weight[
dim : dim * 2, :
]
snake_case__ : List[str] = in_proj_bias[
dim : dim * 2
]
snake_case__ : List[Any] = in_proj_weight[
-dim :, :
]
snake_case__ : Dict = in_proj_bias[-dim :]
# fmt: on
def lowercase_ (A : List[str] , A : List[Any] ):
# fmt: off
snake_case__ : str = config.decoder_config.hidden_size
for idx in range(config.decoder_config.decoder_layers ):
# read in weights + bias of self-attention input projection layer (in the original implementation, this is a single matrix + bias)
snake_case__ : List[Any] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.in_proj_weight''' )
snake_case__ : int = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.in_proj_bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : Any = in_proj_weight[: hidden_size, :]
snake_case__ : Tuple = in_proj_bias[:config.hidden_size]
snake_case__ : List[str] = in_proj_weight[hidden_size : hidden_size * 2, :]
snake_case__ : Dict = in_proj_bias[hidden_size : hidden_size * 2]
snake_case__ : Any = in_proj_weight[-hidden_size :, :]
snake_case__ : int = in_proj_bias[-hidden_size :]
# read in weights + bias of cross-attention input projection layer (in the original implementation, this is a single matrix + bias)
snake_case__ : List[Any] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.in_proj_weight''' )
snake_case__ : List[str] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.in_proj_bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : Optional[int] = in_proj_weight[: hidden_size, :]
snake_case__ : Optional[Any] = in_proj_bias[:config.hidden_size]
snake_case__ : int = in_proj_weight[hidden_size : hidden_size * 2, :]
snake_case__ : List[str] = in_proj_bias[hidden_size : hidden_size * 2]
snake_case__ : List[str] = in_proj_weight[-hidden_size :, :]
snake_case__ : str = in_proj_bias[-hidden_size :]
# fmt: on
def lowercase_ ():
snake_case__ : Any = 'http://images.cocodataset.org/val2017/000000039769.jpg'
snake_case__ : int = Image.open(requests.get(A , stream=A ).raw )
return im
@torch.no_grad()
def lowercase_ (A : str , A : str , A : str , A : bool = False ):
snake_case__ : Optional[int] = get_maskformer_config(A )
# load original state_dict
with open(A , 'rb' ) as f:
snake_case__ : List[Any] = pickle.load(A )
snake_case__ : Optional[int] = data['model']
# for name, param in state_dict.items():
# print(name, param.shape)
# rename keys
snake_case__ : List[str] = create_rename_keys(A )
for src, dest in rename_keys:
rename_key(A , A , A )
read_in_swin_q_k_v(A , config.backbone_config )
read_in_decoder_q_k_v(A , A )
# update to torch tensors
for key, value in state_dict.items():
snake_case__ : int = torch.from_numpy(A )
# load 🤗 model
snake_case__ : str = MaskFormerForInstanceSegmentation(A )
model.eval()
for name, param in model.named_parameters():
print(A , param.shape )
snake_case__ , snake_case__ : Union[str, Any] = model.load_state_dict(A , strict=A )
assert missing_keys == [
"model.pixel_level_module.encoder.model.layernorm.weight",
"model.pixel_level_module.encoder.model.layernorm.bias",
]
assert len(A ) == 0, F'''Unexpected keys: {unexpected_keys}'''
# verify results
snake_case__ : Optional[Any] = prepare_img()
if "vistas" in model_name:
snake_case__ : int = 6_5
elif "cityscapes" in model_name:
snake_case__ : Dict = 6_5_5_3_5
else:
snake_case__ : Tuple = 2_5_5
snake_case__ : Optional[int] = True if 'ade' in model_name else False
snake_case__ : Dict = MaskFormerImageProcessor(ignore_index=A , reduce_labels=A )
snake_case__ : Any = image_processor(A , return_tensors='pt' )
snake_case__ : Any = model(**A )
print('Logits:' , outputs.class_queries_logits[0, :3, :3] )
if model_name == "maskformer-swin-tiny-ade":
snake_case__ : Tuple = torch.tensor(
[[3.6353, -4.4770, -2.6065], [0.5081, -4.2394, -3.5343], [2.1909, -5.0353, -1.9323]] )
assert torch.allclose(outputs.class_queries_logits[0, :3, :3] , A , atol=1e-4 )
print('Looks ok!' )
if pytorch_dump_folder_path is not None:
print(F'''Saving model and image processor to {pytorch_dump_folder_path}''' )
Path(A ).mkdir(exist_ok=A )
model.save_pretrained(A )
image_processor.save_pretrained(A )
if push_to_hub:
print('Pushing model and image processor to the hub...' )
model.push_to_hub(F'''nielsr/{model_name}''' )
image_processor.push_to_hub(F'''nielsr/{model_name}''' )
if __name__ == "__main__":
a_ :Optional[int] = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--model_name",
default="maskformer-swin-tiny-ade",
type=str,
help=("Name of the MaskFormer model you'd like to convert",),
)
parser.add_argument(
"--checkpoint_path",
default="/Users/nielsrogge/Documents/MaskFormer_checkpoints/MaskFormer-Swin-tiny-ADE20k/model.pkl",
type=str,
help="Path to the original state dict (.pth file).",
)
parser.add_argument(
"--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model directory."
)
parser.add_argument(
"--push_to_hub", action="store_true", help="Whether or not to push the converted model to the 🤗 hub."
)
a_ :Dict = parser.parse_args()
convert_maskformer_checkpoint(
args.model_name, args.checkpoint_path, args.pytorch_dump_folder_path, args.push_to_hub
)
| 277 | 0 |
"""simple docstring"""
from math import pow, sqrt
def _snake_case ( *UpperCamelCase : float ):
UpperCAmelCase : str = len(UpperCamelCase ) > 0 and all(value > 0.0 for value in values )
return result
def _snake_case ( UpperCamelCase : float , UpperCamelCase : float ):
return (
round(sqrt(molar_mass_a / molar_mass_a ) , 6 )
if validate(UpperCamelCase , UpperCamelCase )
else ValueError("""Input Error: Molar mass values must greater than 0.""" )
)
def _snake_case ( UpperCamelCase : float , UpperCamelCase : float , UpperCamelCase : float ):
return (
round(effusion_rate * sqrt(molar_mass_a / molar_mass_a ) , 6 )
if validate(UpperCamelCase , UpperCamelCase , UpperCamelCase )
else ValueError(
"""Input Error: Molar mass and effusion rate values must greater than 0.""" )
)
def _snake_case ( UpperCamelCase : float , UpperCamelCase : float , UpperCamelCase : float ):
return (
round(effusion_rate / sqrt(molar_mass_a / molar_mass_a ) , 6 )
if validate(UpperCamelCase , UpperCamelCase , UpperCamelCase )
else ValueError(
"""Input Error: Molar mass and effusion rate values must greater than 0.""" )
)
def _snake_case ( UpperCamelCase : float , UpperCamelCase : float , UpperCamelCase : float ):
return (
round(molar_mass / pow(effusion_rate_a / effusion_rate_a , 2 ) , 6 )
if validate(UpperCamelCase , UpperCamelCase , UpperCamelCase )
else ValueError(
"""Input Error: Molar mass and effusion rate values must greater than 0.""" )
)
def _snake_case ( UpperCamelCase : float , UpperCamelCase : float , UpperCamelCase : float ):
return (
round(pow(effusion_rate_a / effusion_rate_a , 2 ) / molar_mass , 6 )
if validate(UpperCamelCase , UpperCamelCase , UpperCamelCase )
else ValueError(
"""Input Error: Molar mass and effusion rate values must greater than 0.""" )
)
| 109 |
import unittest
from transformers import is_torch_available
from transformers.testing_utils import require_torch, slow, torch_device
from ...generation.test_utils import GenerationTesterMixin
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers import (
OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST,
OpenAIGPTConfig,
OpenAIGPTDoubleHeadsModel,
OpenAIGPTForSequenceClassification,
OpenAIGPTLMHeadModel,
OpenAIGPTModel,
)
class snake_case__ :
"""simple docstring"""
def __init__( self : List[str], _snake_case : Any, _snake_case : int=1_3, _snake_case : Optional[int]=7, _snake_case : int=True, _snake_case : Optional[Any]=True, _snake_case : Optional[Any]=True, _snake_case : Union[str, Any]=9_9, _snake_case : Optional[Any]=3_2, _snake_case : Tuple=5, _snake_case : str=4, _snake_case : Any=3_7, _snake_case : int="gelu", _snake_case : Optional[Any]=0.1, _snake_case : str=0.1, _snake_case : str=5_1_2, _snake_case : Dict=1_6, _snake_case : str=2, _snake_case : Union[str, Any]=0.0_2, _snake_case : Optional[int]=3, _snake_case : Union[str, Any]=4, _snake_case : Tuple=None, ) ->Optional[Any]:
snake_case__ : Optional[int] = parent
snake_case__ : List[Any] = batch_size
snake_case__ : Tuple = seq_length
snake_case__ : str = is_training
snake_case__ : Optional[int] = use_token_type_ids
snake_case__ : Any = use_labels
snake_case__ : Dict = vocab_size
snake_case__ : str = hidden_size
snake_case__ : Union[str, Any] = num_hidden_layers
snake_case__ : List[str] = num_attention_heads
snake_case__ : Union[str, Any] = intermediate_size
snake_case__ : List[Any] = hidden_act
snake_case__ : int = hidden_dropout_prob
snake_case__ : str = attention_probs_dropout_prob
snake_case__ : Any = max_position_embeddings
snake_case__ : Union[str, Any] = type_vocab_size
snake_case__ : Optional[Any] = type_sequence_label_size
snake_case__ : Optional[int] = initializer_range
snake_case__ : Optional[int] = num_labels
snake_case__ : str = num_choices
snake_case__ : int = scope
snake_case__ : List[str] = self.vocab_size - 1
def lowercase_ ( self : Union[str, Any] ) ->Tuple:
snake_case__ : List[str] = ids_tensor([self.batch_size, self.seq_length], self.vocab_size )
snake_case__ : List[str] = None
if self.use_token_type_ids:
snake_case__ : Optional[Any] = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size )
snake_case__ : Tuple = None
snake_case__ : str = None
snake_case__ : List[Any] = None
if self.use_labels:
snake_case__ : Dict = ids_tensor([self.batch_size], self.type_sequence_label_size )
snake_case__ : int = ids_tensor([self.batch_size, self.seq_length], self.num_labels )
snake_case__ : List[str] = ids_tensor([self.batch_size], self.num_choices )
snake_case__ : Union[str, Any] = OpenAIGPTConfig(
vocab_size=self.vocab_size, n_embd=self.hidden_size, n_layer=self.num_hidden_layers, n_head=self.num_attention_heads, n_positions=self.max_position_embeddings, pad_token_id=self.pad_token_id, )
snake_case__ : List[str] = ids_tensor([self.num_hidden_layers, self.num_attention_heads], 2 )
return (
config,
input_ids,
head_mask,
token_type_ids,
sequence_labels,
token_labels,
choice_labels,
)
def lowercase_ ( self : Any, _snake_case : List[str], _snake_case : Any, _snake_case : List[Any], _snake_case : Tuple, *_snake_case : Optional[Any] ) ->Tuple:
snake_case__ : Union[str, Any] = OpenAIGPTModel(config=_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Optional[Any] = model(_snake_case, token_type_ids=_snake_case, head_mask=_snake_case )
snake_case__ : Union[str, Any] = model(_snake_case, token_type_ids=_snake_case )
snake_case__ : Optional[Any] = model(_snake_case )
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size) )
def lowercase_ ( self : Optional[int], _snake_case : Optional[Any], _snake_case : Union[str, Any], _snake_case : Optional[int], _snake_case : List[Any], *_snake_case : Dict ) ->Optional[int]:
snake_case__ : Optional[Any] = OpenAIGPTLMHeadModel(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Tuple = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.loss.shape, () )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size) )
def lowercase_ ( self : int, _snake_case : Tuple, _snake_case : List[str], _snake_case : List[Any], _snake_case : List[Any], *_snake_case : List[Any] ) ->Optional[int]:
snake_case__ : List[str] = OpenAIGPTDoubleHeadsModel(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Optional[Any] = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.loss.shape, () )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size) )
def lowercase_ ( self : Optional[int], _snake_case : Tuple, _snake_case : Dict, _snake_case : List[str], _snake_case : Optional[Any], *_snake_case : Union[str, Any] ) ->str:
snake_case__ : List[str] = self.num_labels
snake_case__ : Dict = OpenAIGPTForSequenceClassification(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : List[str] = ids_tensor([self.batch_size], self.type_sequence_label_size )
snake_case__ : List[str] = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels) )
def lowercase_ ( self : Dict ) ->int:
snake_case__ : List[Any] = self.prepare_config_and_inputs()
(
(
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) ,
) : Optional[Any] = config_and_inputs
snake_case__ : str = {
'input_ids': input_ids,
'token_type_ids': token_type_ids,
'head_mask': head_mask,
}
return config, inputs_dict
@require_torch
class snake_case__ ( lowerCAmelCase_ , lowerCAmelCase_ , lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = (
(OpenAIGPTModel, OpenAIGPTLMHeadModel, OpenAIGPTDoubleHeadsModel, OpenAIGPTForSequenceClassification)
if is_torch_available()
else ()
)
_SCREAMING_SNAKE_CASE = (
(OpenAIGPTLMHeadModel,) if is_torch_available() else ()
) # TODO (PVP): Add Double HeadsModel when generate() function is changed accordingly
_SCREAMING_SNAKE_CASE = (
{
"""feature-extraction""": OpenAIGPTModel,
"""text-classification""": OpenAIGPTForSequenceClassification,
"""text-generation""": OpenAIGPTLMHeadModel,
"""zero-shot""": OpenAIGPTForSequenceClassification,
}
if is_torch_available()
else {}
)
def lowercase_ ( self : Optional[int], _snake_case : Union[str, Any], _snake_case : int, _snake_case : Tuple, _snake_case : Tuple, _snake_case : List[str] ) ->Optional[Any]:
if pipeline_test_casse_name == "ZeroShotClassificationPipelineTests":
# Get `tokenizer does not have a padding token` error for both fast/slow tokenizers.
# `OpenAIGPTConfig` was never used in pipeline tests, either because of a missing checkpoint or because a
# tiny config could not be created.
return True
return False
def lowercase_ ( self : Optional[Any], _snake_case : Union[str, Any], _snake_case : List[str], _snake_case : Any=False ) ->Tuple:
snake_case__ : Optional[int] = super()._prepare_for_class(_snake_case, _snake_case, return_labels=_snake_case )
if return_labels:
if model_class.__name__ == "OpenAIGPTDoubleHeadsModel":
snake_case__ : Union[str, Any] = torch.zeros(
(self.model_tester.batch_size, self.model_tester.num_choices, self.model_tester.seq_length), dtype=torch.long, device=_snake_case, )
snake_case__ : List[Any] = inputs_dict['labels']
snake_case__ : List[Any] = inputs_dict['labels']
snake_case__ : Any = torch.zeros(
(self.model_tester.batch_size, self.model_tester.num_choices), dtype=torch.long, device=_snake_case, )
snake_case__ : Tuple = torch.zeros(
self.model_tester.batch_size, dtype=torch.long, device=_snake_case )
return inputs_dict
def lowercase_ ( self : Union[str, Any] ) ->List[str]:
snake_case__ : List[str] = OpenAIGPTModelTester(self )
snake_case__ : Any = ConfigTester(self, config_class=_snake_case, n_embd=3_7 )
def lowercase_ ( self : Optional[int] ) ->str:
self.config_tester.run_common_tests()
def lowercase_ ( self : int ) ->Tuple:
snake_case__ : str = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_openai_gpt_model(*_snake_case )
def lowercase_ ( self : Tuple ) ->List[str]:
snake_case__ : Optional[int] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_lm_head_model(*_snake_case )
def lowercase_ ( self : Dict ) ->int:
snake_case__ : Optional[Any] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_double_lm_head_model(*_snake_case )
def lowercase_ ( self : int ) ->str:
snake_case__ : List[Any] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_openai_gpt_for_sequence_classification(*_snake_case )
@slow
def lowercase_ ( self : Optional[Any] ) ->str:
for model_name in OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
snake_case__ : Optional[int] = OpenAIGPTModel.from_pretrained(_snake_case )
self.assertIsNotNone(_snake_case )
@require_torch
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
@slow
def lowercase_ ( self : Tuple ) ->Optional[int]:
snake_case__ : Union[str, Any] = OpenAIGPTLMHeadModel.from_pretrained('openai-gpt' )
model.to(_snake_case )
snake_case__ : Tuple = torch.tensor([[4_8_1, 4_7_3_5, 5_4_4]], dtype=torch.long, device=_snake_case ) # the president is
snake_case__ : int = [
4_8_1,
4_7_3_5,
5_4_4,
2_4_6,
9_6_3,
8_7_0,
7_6_2,
2_3_9,
2_4_4,
4_0_4_7_7,
2_4_4,
2_4_9,
7_1_9,
8_8_1,
4_8_7,
5_4_4,
2_4_0,
2_4_4,
6_0_3,
4_8_1,
] # the president is a very good man. " \n " i\'m sure he is, " said the
snake_case__ : Optional[int] = model.generate(_snake_case, do_sample=_snake_case )
self.assertListEqual(output_ids[0].tolist(), _snake_case )
| 277 | 0 |
from __future__ import absolute_import, division, print_function, unicode_literals
from torch import nn
from torch.nn import CrossEntropyLoss, MSELoss
from transformers import RobertaConfig
from transformers.file_utils import add_start_docstrings, add_start_docstrings_to_model_forward
from transformers.models.roberta.modeling_roberta import (
ROBERTA_INPUTS_DOCSTRING,
ROBERTA_START_DOCSTRING,
RobertaEmbeddings,
)
from .modeling_highway_bert import BertPreTrainedModel, DeeBertModel, HighwayException, entropy
@add_start_docstrings(
"""The RoBERTa Model transformer with early exiting (DeeRoBERTa). """ , lowerCAmelCase_ , )
class __lowerCAmelCase ( lowerCAmelCase_):
_lowercase : Union[str, Any] = RobertaConfig
_lowercase : Tuple = """roberta"""
def __init__( self , lowerCAmelCase__ ) -> List[Any]:
'''simple docstring'''
super().__init__(_snake_case )
a__ : Union[str, Any] =RobertaEmbeddings(_snake_case )
self.init_weights()
@add_start_docstrings(
"""RoBERTa Model (with early exiting - DeeRoBERTa) with a classifier on top,
also takes care of multi-layer training. """ , lowerCAmelCase_ , )
class __lowerCAmelCase ( lowerCAmelCase_):
_lowercase : Optional[int] = RobertaConfig
_lowercase : List[Any] = """roberta"""
def __init__( self , lowerCAmelCase__ ) -> str:
'''simple docstring'''
super().__init__(_snake_case )
a__ : Dict =config.num_labels
a__ : Tuple =config.num_hidden_layers
a__ : Optional[int] =DeeRobertaModel(_snake_case )
a__ : Any =nn.Dropout(config.hidden_dropout_prob )
a__ : str =nn.Linear(config.hidden_size , self.config.num_labels )
@add_start_docstrings_to_model_forward(_snake_case )
def _lowercase ( self , lowerCAmelCase__=None , lowerCAmelCase__=None , lowerCAmelCase__=None , lowerCAmelCase__=None , lowerCAmelCase__=None , lowerCAmelCase__=None , lowerCAmelCase__=None , lowerCAmelCase__=-1 , lowerCAmelCase__=False , ) -> List[str]:
'''simple docstring'''
a__ : Optional[Any] =self.num_layers
try:
a__ : Union[str, Any] =self.roberta(
_snake_case , attention_mask=_snake_case , token_type_ids=_snake_case , position_ids=_snake_case , head_mask=_snake_case , inputs_embeds=_snake_case , )
a__ : List[str] =outputs[1]
a__ : int =self.dropout(_snake_case )
a__ : Union[str, Any] =self.classifier(_snake_case )
a__ : Tuple =(logits,) + outputs[2:] # add hidden states and attention if they are here
except HighwayException as e:
a__ : str =e.message
a__ : Any =e.exit_layer
a__ : Any =outputs[0]
if not self.training:
a__ : Optional[int] =entropy(_snake_case )
a__ : Dict =[]
a__ : Any =[]
if labels is not None:
if self.num_labels == 1:
# We are doing regression
a__ : List[str] =MSELoss()
a__ : Optional[Any] =loss_fct(logits.view(-1 ) , labels.view(-1 ) )
else:
a__ : str =CrossEntropyLoss()
a__ : List[Any] =loss_fct(logits.view(-1 , self.num_labels ) , labels.view(-1 ) )
# work with highway exits
a__ : Optional[int] =[]
for highway_exit in outputs[-1]:
a__ : List[Any] =highway_exit[0]
if not self.training:
highway_logits_all.append(_snake_case )
highway_entropy.append(highway_exit[2] )
if self.num_labels == 1:
# We are doing regression
a__ : str =MSELoss()
a__ : List[str] =loss_fct(highway_logits.view(-1 ) , labels.view(-1 ) )
else:
a__ : int =CrossEntropyLoss()
a__ : Any =loss_fct(highway_logits.view(-1 , self.num_labels ) , labels.view(-1 ) )
highway_losses.append(_snake_case )
if train_highway:
a__ : Optional[Any] =(sum(highway_losses[:-1] ),) + outputs
# exclude the final highway, of course
else:
a__ : Optional[int] =(loss,) + outputs
if not self.training:
a__ : Optional[int] =outputs + ((original_entropy, highway_entropy), exit_layer)
if output_layer >= 0:
a__ : str =(
(outputs[0],) + (highway_logits_all[output_layer],) + outputs[2:]
) # use the highway of the last layer
return outputs # (loss), logits, (hidden_states), (attentions), entropy
| 95 |
import os
import unittest
from transformers.models.transfo_xl.tokenization_transfo_xl import VOCAB_FILES_NAMES, TransfoXLTokenizer
from ...test_tokenization_common import TokenizerTesterMixin
class snake_case__ ( lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = TransfoXLTokenizer
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
def lowercase_ ( self : Optional[int] ) ->Any:
super().setUp()
snake_case__ : Tuple = [
'<unk>',
'[CLS]',
'[SEP]',
'want',
'unwanted',
'wa',
'un',
'running',
',',
'low',
'l',
]
snake_case__ : Any = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['vocab_file'] )
with open(self.vocab_file, 'w', encoding='utf-8' ) as vocab_writer:
vocab_writer.write(''.join([x + '\n' for x in vocab_tokens] ) )
def lowercase_ ( self : Union[str, Any], **_snake_case : List[Any] ) ->Dict:
snake_case__ : str = True
return TransfoXLTokenizer.from_pretrained(self.tmpdirname, **_snake_case )
def lowercase_ ( self : Optional[Any], _snake_case : str ) ->Dict:
snake_case__ : List[Any] = '<unk> UNwanted , running'
snake_case__ : List[Any] = '<unk> unwanted, running'
return input_text, output_text
def lowercase_ ( self : List[Any] ) ->Tuple:
snake_case__ : Dict = TransfoXLTokenizer(vocab_file=self.vocab_file, lower_case=_snake_case )
snake_case__ : str = tokenizer.tokenize('<unk> UNwanted , running' )
self.assertListEqual(_snake_case, ['<unk>', 'unwanted', ',', 'running'] )
self.assertListEqual(tokenizer.convert_tokens_to_ids(_snake_case ), [0, 4, 8, 7] )
def lowercase_ ( self : List[str] ) ->List[Any]:
snake_case__ : str = TransfoXLTokenizer(lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHeLLo ! how \n Are yoU ? ' ), ['hello', '!', 'how', 'are', 'you', '?'] )
def lowercase_ ( self : Optional[int] ) ->Optional[Any]:
snake_case__ : Optional[int] = TransfoXLTokenizer(lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHeLLo ! how \n Are yoU ? ' ), ['HeLLo', '!', 'how', 'Are', 'yoU', '?'] )
def lowercase_ ( self : Optional[int] ) ->Union[str, Any]:
snake_case__ : List[Any] = TransfoXLTokenizer(lower_case=_snake_case )
snake_case__ : Dict = 'Hello (bracket) and side-scrolled [and] Henry\'s $5,000 with 3.34 m. What\'s up!?'
snake_case__ : List[Any] = [
'Hello',
'(',
'bracket',
')',
'and',
'side',
'@-@',
'scrolled',
'[',
'and',
']',
'Henry',
'\'s',
'$',
'5',
'@,@',
'000',
'with',
'3',
'@.@',
'34',
'm',
'.',
'What',
'\'s',
'up',
'!',
'?',
]
self.assertListEqual(tokenizer.tokenize(_snake_case ), _snake_case )
self.assertEqual(tokenizer.convert_tokens_to_string(_snake_case ), _snake_case )
def lowercase_ ( self : Dict ) ->Any:
snake_case__ : Dict = self.get_tokenizer()
snake_case__ : Optional[Any] = len(_snake_case )
tokenizer.add_tokens(['new1', 'new2'] )
tokenizer.move_added_token('new1', 1 )
# Check that moved token is not copied (duplicate)
self.assertEqual(len(_snake_case ), original_len + 2 )
# Check that token is moved to specified id
self.assertEqual(tokenizer.encode('new1' ), [1] )
self.assertEqual(tokenizer.decode([1] ), 'new1' )
| 277 | 0 |
def a ( snake_case__: List[str] ):
'''simple docstring'''
lowercase_ = len(snake_case__ )
for i in range(length - 1 ):
lowercase_ = i
for k in range(i + 1 , snake_case__ ):
if collection[k] < collection[least]:
lowercase_ = k
if least != i:
lowercase_ = (collection[i], collection[least])
return collection
if __name__ == "__main__":
__a = input('Enter numbers separated by a comma:\n').strip()
__a = [int(item) for item in user_input.split(',')]
print(selection_sort(unsorted))
| 30 |
from ...configuration_utils import PretrainedConfig
from ...utils import logging
a_ :Optional[int] = logging.get_logger(__name__)
a_ :Dict = {"openai-gpt": "https://huggingface.co/openai-gpt/resolve/main/config.json"}
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """openai-gpt"""
_SCREAMING_SNAKE_CASE = {
"""max_position_embeddings""": """n_positions""",
"""hidden_size""": """n_embd""",
"""num_attention_heads""": """n_head""",
"""num_hidden_layers""": """n_layer""",
}
def __init__( self : Optional[int], _snake_case : Dict=4_0_4_7_8, _snake_case : str=5_1_2, _snake_case : int=7_6_8, _snake_case : Tuple=1_2, _snake_case : Any=1_2, _snake_case : str="gelu", _snake_case : List[str]=0.1, _snake_case : Any=0.1, _snake_case : Dict=0.1, _snake_case : int=1e-5, _snake_case : Optional[Any]=0.0_2, _snake_case : List[Any]="cls_index", _snake_case : Any=True, _snake_case : Any=None, _snake_case : int=True, _snake_case : Optional[Any]=0.1, **_snake_case : List[Any], ) ->Optional[int]:
snake_case__ : int = vocab_size
snake_case__ : Dict = n_positions
snake_case__ : str = n_embd
snake_case__ : str = n_layer
snake_case__ : List[Any] = n_head
snake_case__ : List[Any] = afn
snake_case__ : Optional[Any] = resid_pdrop
snake_case__ : List[str] = embd_pdrop
snake_case__ : List[Any] = attn_pdrop
snake_case__ : Optional[int] = layer_norm_epsilon
snake_case__ : str = initializer_range
snake_case__ : List[str] = summary_type
snake_case__ : Optional[int] = summary_use_proj
snake_case__ : List[str] = summary_activation
snake_case__ : Optional[Any] = summary_first_dropout
snake_case__ : int = summary_proj_to_labels
super().__init__(**_snake_case )
| 277 | 0 |
"""simple docstring"""
import unittest
import numpy as np
from transformers.testing_utils import require_torch, require_vision
from transformers.utils import is_torch_available, is_vision_available
from ...test_image_processing_common import ImageProcessingSavingTestMixin, prepare_image_inputs
if is_torch_available():
import torch
if is_vision_available():
from PIL import Image
from transformers import MobileNetVaImageProcessor
class __lowerCAmelCase ( unittest.TestCase ):
'''simple docstring'''
def __init__( self , _a , _a=7 , _a=3 , _a=18 , _a=30 , _a=400 , _a=True , _a=None , _a=True , _a=None , ):
__a = size if size is not None else {'shortest_edge': 20}
__a = crop_size if crop_size is not None else {'height': 18, 'width': 18}
__a = parent
__a = batch_size
__a = num_channels
__a = image_size
__a = min_resolution
__a = max_resolution
__a = do_resize
__a = size
__a = do_center_crop
__a = crop_size
def __UpperCAmelCase ( self ):
return {
"do_resize": self.do_resize,
"size": self.size,
"do_center_crop": self.do_center_crop,
"crop_size": self.crop_size,
}
@require_torch
@require_vision
class __lowerCAmelCase ( lowerCAmelCase_ , unittest.TestCase ):
'''simple docstring'''
__UpperCAmelCase : Optional[Any] = MobileNetVaImageProcessor if is_vision_available() else None
def __UpperCAmelCase ( self ):
__a = MobileNetVaImageProcessingTester(self )
@property
def __UpperCAmelCase ( self ):
return self.image_processor_tester.prepare_image_processor_dict()
def __UpperCAmelCase ( self ):
__a = self.image_processing_class(**self.image_processor_dict )
self.assertTrue(hasattr(_snake_case , '''do_resize''' ) )
self.assertTrue(hasattr(_snake_case , '''size''' ) )
self.assertTrue(hasattr(_snake_case , '''do_center_crop''' ) )
self.assertTrue(hasattr(_snake_case , '''crop_size''' ) )
def __UpperCAmelCase ( self ):
__a = self.image_processing_class.from_dict(self.image_processor_dict )
self.assertEqual(image_processor.size , {'''shortest_edge''': 20} )
self.assertEqual(image_processor.crop_size , {'''height''': 18, '''width''': 18} )
__a = self.image_processing_class.from_dict(self.image_processor_dict , size=42 , crop_size=84 )
self.assertEqual(image_processor.size , {'''shortest_edge''': 42} )
self.assertEqual(image_processor.crop_size , {'''height''': 84, '''width''': 84} )
def __UpperCAmelCase ( self ):
pass
def __UpperCAmelCase ( self ):
# Initialize image_processing
__a = self.image_processing_class(**self.image_processor_dict )
# create random PIL images
__a = prepare_image_inputs(self.image_processor_tester , equal_resolution=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case , Image.Image )
# Test not batched input
__a = image_processing(image_inputs[0] , return_tensors='''pt''' ).pixel_values
self.assertEqual(
encoded_images.shape , (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['''height'''],
self.image_processor_tester.crop_size['''width'''],
) , )
# Test batched
__a = image_processing(_snake_case , return_tensors='''pt''' ).pixel_values
self.assertEqual(
encoded_images.shape , (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['''height'''],
self.image_processor_tester.crop_size['''width'''],
) , )
def __UpperCAmelCase ( self ):
# Initialize image_processing
__a = self.image_processing_class(**self.image_processor_dict )
# create random numpy tensors
__a = prepare_image_inputs(self.image_processor_tester , equal_resolution=_snake_case , numpify=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case , np.ndarray )
# Test not batched input
__a = image_processing(image_inputs[0] , return_tensors='''pt''' ).pixel_values
self.assertEqual(
encoded_images.shape , (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['''height'''],
self.image_processor_tester.crop_size['''width'''],
) , )
# Test batched
__a = image_processing(_snake_case , return_tensors='''pt''' ).pixel_values
self.assertEqual(
encoded_images.shape , (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['''height'''],
self.image_processor_tester.crop_size['''width'''],
) , )
def __UpperCAmelCase ( self ):
# Initialize image_processing
__a = self.image_processing_class(**self.image_processor_dict )
# create random PyTorch tensors
__a = prepare_image_inputs(self.image_processor_tester , equal_resolution=_snake_case , torchify=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case , torch.Tensor )
# Test not batched input
__a = image_processing(image_inputs[0] , return_tensors='''pt''' ).pixel_values
self.assertEqual(
encoded_images.shape , (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['''height'''],
self.image_processor_tester.crop_size['''width'''],
) , )
# Test batched
__a = image_processing(_snake_case , return_tensors='''pt''' ).pixel_values
self.assertEqual(
encoded_images.shape , (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['''height'''],
self.image_processor_tester.crop_size['''width'''],
) , )
| 45 |
import argparse
import logging
import os
from datetime import datetime
import numpy as np
import torch
from torch import nn
from torch.utils.data import DataLoader, RandomSampler, TensorDataset
from tqdm import tqdm
from transformers import GPTaLMHeadModel
a_ :Optional[Any] = logging.getLogger(__name__)
def lowercase_ (A : List[Any] , A : List[Any] ):
# save results
if os.path.exists(A ):
if os.path.exists(os.path.join(A , 'config.json' ) ) and os.path.isfile(
os.path.join(A , 'config.json' ) ):
os.remove(os.path.join(A , 'config.json' ) )
if os.path.exists(os.path.join(A , 'pytorch_model.bin' ) ) and os.path.isfile(
os.path.join(A , 'pytorch_model.bin' ) ):
os.remove(os.path.join(A , 'pytorch_model.bin' ) )
else:
os.makedirs(A )
model.save_pretrained(A )
def lowercase_ (A : Any , A : Optional[Any]=False ):
snake_case__ : str = 2
if unlogit:
snake_case__ : Dict = torch.pow(A , A )
snake_case__ : Any = p * torch.log(A )
snake_case__ : Tuple = 0
return -plogp.sum(dim=-1 )
def lowercase_ (A : List[str] ):
logger.info('lv, h >\t' + '\t'.join(F'''{x + 1}''' for x in range(len(A ) ) ) )
for row in range(len(A ) ):
if tensor.dtype != torch.long:
logger.info(F'''layer {row + 1}:\t''' + '\t'.join(F'''{x:.5f}''' for x in tensor[row].cpu().data ) )
else:
logger.info(F'''layer {row + 1}:\t''' + '\t'.join(F'''{x:d}''' for x in tensor[row].cpu().data ) )
def lowercase_ (A : Tuple , A : Optional[Any] , A : str , A : int=True , A : Optional[int]=True , A : Any=None , A : int=False ):
snake_case__ , snake_case__ : Optional[Any] = model.config.num_hidden_layers, model.config.num_attention_heads
snake_case__ : int = torch.zeros(A , A ).to(args.device )
snake_case__ : Any = torch.zeros(A , A ).to(args.device )
if head_mask is None:
snake_case__ : Dict = torch.ones(A , A ).to(args.device )
head_mask.requires_grad_(requires_grad=A )
# If actually pruned attention multi-head, set head mask to None to avoid shape mismatch
if actually_pruned:
snake_case__ : Optional[int] = None
snake_case__ : List[Any] = 0.0
snake_case__ : str = 0.0
for step, inputs in enumerate(tqdm(A , desc='Iteration' , disable=args.local_rank not in [-1, 0] ) ):
snake_case__ : Union[str, Any] = tuple(t.to(args.device ) for t in inputs )
((snake_case__) , ) : Optional[Any] = inputs
# Do a forward pass (not with torch.no_grad() since we need gradients for importance score - see below)
snake_case__ : Union[str, Any] = model(A , labels=A , head_mask=A )
# (loss), lm_logits, presents, (all hidden_states), (attentions)
snake_case__ , snake_case__ , snake_case__ : Dict = (
outputs[0],
outputs[1],
outputs[-1],
) # Loss and logits are the first, attention the last
loss.backward() # Backpropagate to populate the gradients in the head mask
total_loss += loss.detach().cpu().numpy()
if compute_entropy:
for layer, attn in enumerate(A ):
snake_case__ : Optional[Any] = entropy(attn.detach() , A )
attn_entropy[layer] += masked_entropy.sum(-1 ).sum(0 ).sum(0 ).detach()
if compute_importance:
head_importance += head_mask.grad.abs().detach()
tot_tokens += torch.ones_like(A ).float().detach().sum().data
# Normalize
attn_entropy /= tot_tokens
head_importance /= tot_tokens
# Layerwise importance normalization
if not args.dont_normalize_importance_by_layer:
snake_case__ : Union[str, Any] = 2
snake_case__ : List[Any] = torch.pow(torch.pow(A , A ).sum(-1 ) , 1 / exponent )
head_importance /= norm_by_layer.unsqueeze(-1 ) + 1e-20
if not args.dont_normalize_global_importance:
snake_case__ : Tuple = (head_importance - head_importance.min()) / (head_importance.max() - head_importance.min())
# Print matrices
if compute_entropy:
logger.info('Attention entropies' )
print_ad_tensor(A )
if compute_importance:
logger.info('Head importance scores' )
print_ad_tensor(A )
logger.info('Head ranked by importance scores' )
snake_case__ : Tuple = torch.zeros(head_importance.numel() , dtype=torch.long , device=args.device )
snake_case__ : Union[str, Any] = torch.arange(
head_importance.numel() , device=args.device )
snake_case__ : str = head_ranks.view_as(A )
print_ad_tensor(A )
return attn_entropy, head_importance, total_loss
def lowercase_ (A : Optional[int] , A : Dict , A : Optional[int] ):
snake_case__ , snake_case__ , snake_case__ : Any = compute_heads_importance(A , A , A , compute_entropy=A )
snake_case__ : Tuple = 1 / loss # instead of downsteam score use the LM loss
logger.info('Pruning: original score: %f, threshold: %f' , A , original_score * args.masking_threshold )
snake_case__ : Optional[Any] = torch.ones_like(A )
snake_case__ : Union[str, Any] = max(1 , int(new_head_mask.numel() * args.masking_amount ) )
snake_case__ : Dict = original_score
while current_score >= original_score * args.masking_threshold:
snake_case__ : int = new_head_mask.clone().detach() # save current head mask
# heads from least important to most - keep only not-masked heads
snake_case__ : List[Any] = float('Inf' )
snake_case__ : Union[str, Any] = head_importance.view(-1 ).sort()[1]
if len(A ) <= num_to_mask:
print('BREAK BY num_to_mask' )
break
# mask heads
snake_case__ : int = current_heads_to_mask[:num_to_mask]
logger.info('Heads to mask: %s' , str(current_heads_to_mask.tolist() ) )
snake_case__ : int = new_head_mask.view(-1 )
snake_case__ : int = 0.0
snake_case__ : Union[str, Any] = new_head_mask.view_as(A )
snake_case__ : List[str] = new_head_mask.clone().detach()
print_ad_tensor(A )
# Compute metric and head importance again
snake_case__ , snake_case__ , snake_case__ : Any = compute_heads_importance(
A , A , A , compute_entropy=A , head_mask=A )
snake_case__ : Dict = 1 / loss
logger.info(
'Masking: current score: %f, remaining heads %d (%.1f percents)' , A , new_head_mask.sum() , new_head_mask.sum() / new_head_mask.numel() * 1_0_0 , )
logger.info('Final head mask' )
print_ad_tensor(A )
np.save(os.path.join(args.output_dir , 'head_mask.npy' ) , head_mask.detach().cpu().numpy() )
return head_mask
def lowercase_ (A : List[str] , A : Tuple , A : Optional[Any] , A : int ):
snake_case__ : Any = datetime.now()
snake_case__ , snake_case__ , snake_case__ : str = compute_heads_importance(
A , A , A , compute_entropy=A , compute_importance=A , head_mask=A )
snake_case__ : Tuple = 1 / loss
snake_case__ : Dict = datetime.now() - before_time
snake_case__ : Union[str, Any] = sum(p.numel() for p in model.parameters() )
snake_case__ : Optional[Any] = {
layer: (1 - head_mask[layer].long()).nonzero().squeeze().tolist() for layer in range(len(A ) )
}
for k, v in heads_to_prune.items():
if isinstance(A , A ):
snake_case__ : Any = [
v,
]
assert sum(len(A ) for h in heads_to_prune.values() ) == (1 - head_mask.long()).sum().item()
model.prune_heads(A )
snake_case__ : Dict = sum(p.numel() for p in model.parameters() )
snake_case__ : Tuple = datetime.now()
snake_case__ , snake_case__ , snake_case__ : Dict = compute_heads_importance(
A , A , A , compute_entropy=A , compute_importance=A , head_mask=A , actually_pruned=A , )
snake_case__ : Any = 1 / loss
snake_case__ : int = datetime.now() - before_time
logger.info(
'Pruning: original num of params: %.2e, after pruning %.2e (%.1f percents)' , A , A , pruned_num_params / original_num_params * 1_0_0 , )
logger.info('Pruning: score with masking: %f score with pruning: %f' , A , A )
logger.info('Pruning: speed ratio (original timing / new timing): %f percents' , original_time / new_time * 1_0_0 )
save_model(A , args.output_dir )
def lowercase_ ():
snake_case__ : str = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
'--data_dir' , default=A , type=A , required=A , help='The input data dir. Should contain the .tsv files (or other data files) for the task.' , )
parser.add_argument(
'--model_name_or_path' , default=A , type=A , required=A , help='Path to pretrained model or model identifier from huggingface.co/models' , )
parser.add_argument(
'--output_dir' , default=A , type=A , required=A , help='The output directory where the model predictions and checkpoints will be written.' , )
# Other parameters
parser.add_argument(
'--config_name' , default='' , type=A , help='Pretrained config name or path if not the same as model_name_or_path' , )
parser.add_argument(
'--tokenizer_name' , default='' , type=A , help='Pretrained tokenizer name or path if not the same as model_name_or_path' , )
parser.add_argument(
'--cache_dir' , default=A , type=A , help='Where do you want to store the pre-trained models downloaded from s3' , )
parser.add_argument(
'--data_subset' , type=A , default=-1 , help='If > 0: limit the data to a subset of data_subset instances.' )
parser.add_argument(
'--overwrite_output_dir' , action='store_true' , help='Whether to overwrite data in output directory' )
parser.add_argument(
'--overwrite_cache' , action='store_true' , help='Overwrite the cached training and evaluation sets' )
parser.add_argument(
'--dont_normalize_importance_by_layer' , action='store_true' , help='Don\'t normalize importance score by layers' )
parser.add_argument(
'--dont_normalize_global_importance' , action='store_true' , help='Don\'t normalize all importance scores between 0 and 1' , )
parser.add_argument(
'--try_masking' , action='store_true' , help='Whether to try to mask head until a threshold of accuracy.' )
parser.add_argument(
'--masking_threshold' , default=0.9 , type=A , help='masking threshold in term of metrics (stop masking when metric < threshold * original metric value).' , )
parser.add_argument(
'--masking_amount' , default=0.1 , type=A , help='Amount to heads to masking at each masking step.' )
parser.add_argument('--metric_name' , default='acc' , type=A , help='Metric to use for head masking.' )
parser.add_argument(
'--max_seq_length' , default=1_2_8 , type=A , help=(
'The maximum total input sequence length after WordPiece tokenization. \n'
'Sequences longer than this will be truncated, sequences shorter padded.'
) , )
parser.add_argument('--batch_size' , default=1 , type=A , help='Batch size.' )
parser.add_argument('--seed' , type=A , default=4_2 )
parser.add_argument('--local_rank' , type=A , default=-1 , help='local_rank for distributed training on gpus' )
parser.add_argument('--no_cuda' , action='store_true' , help='Whether not to use CUDA when available' )
parser.add_argument('--server_ip' , type=A , default='' , help='Can be used for distant debugging.' )
parser.add_argument('--server_port' , type=A , default='' , help='Can be used for distant debugging.' )
snake_case__ : Optional[int] = parser.parse_args()
if args.server_ip and args.server_port:
# Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script
import ptvsd
print('Waiting for debugger attach' )
ptvsd.enable_attach(address=(args.server_ip, args.server_port) , redirect_output=A )
ptvsd.wait_for_attach()
# Setup devices and distributed training
if args.local_rank == -1 or args.no_cuda:
snake_case__ : List[Any] = torch.device('cuda' if torch.cuda.is_available() and not args.no_cuda else 'cpu' )
snake_case__ : Optional[Any] = 0 if args.no_cuda else torch.cuda.device_count()
else:
torch.cuda.set_device(args.local_rank )
snake_case__ : int = torch.device('cuda' , args.local_rank )
snake_case__ : List[str] = 1
torch.distributed.init_process_group(backend='nccl' ) # Initializes the distributed backend
# Setup logging
logging.basicConfig(level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN )
logger.info('device: {} n_gpu: {}, distributed: {}'.format(args.device , args.n_gpu , bool(args.local_rank != -1 ) ) )
snake_case__ : Any = GPTaLMHeadModel.from_pretrained(args.model_name_or_path )
# Distributed and parallel training
model.to(args.device )
if args.local_rank != -1:
snake_case__ : List[str] = nn.parallel.DistributedDataParallel(
A , device_ids=[args.local_rank] , output_device=args.local_rank , find_unused_parameters=A )
elif args.n_gpu > 1:
snake_case__ : Optional[int] = nn.DataParallel(A )
# Print/save training arguments
os.makedirs(args.output_dir , exist_ok=A )
torch.save(A , os.path.join(args.output_dir , 'run_args.bin' ) )
logger.info('Training/evaluation parameters %s' , A )
# Prepare dataset
snake_case__ : Optional[Any] = np.concatenate(
[
np.loadtxt(args.data_dir , dtype=np.intaa ),
] )
snake_case__ : List[str] = (torch.from_numpy(A ),)
snake_case__ : int = TensorDataset(*A )
snake_case__ : Union[str, Any] = RandomSampler(A )
snake_case__ : Any = DataLoader(A , sampler=A , batch_size=args.batch_size )
# Compute head entropy and importance score
compute_heads_importance(A , A , A )
# Try head masking (set heads to zero until the score goes under a threshole)
# and head pruning (remove masked heads and see the effect on the network)
if args.try_masking and args.masking_threshold > 0.0 and args.masking_threshold < 1.0:
snake_case__ : Dict = mask_heads(A , A , A )
prune_heads(A , A , A , A )
if __name__ == "__main__":
main()
| 277 | 0 |
'''simple docstring'''
import baseaa
import io
import json
import os
from copy import deepcopy
from ..optimizer import AcceleratedOptimizer
from ..scheduler import AcceleratedScheduler
class A :
'''simple docstring'''
def __init__(self , _UpperCAmelCase ) -> Optional[int]:
if isinstance(_snake_case , _snake_case ):
# Don't modify user's data should they want to reuse it (e.g. in tests), because once we
# modified it, it will not be accepted here again, since `auto` values would have been overridden
__UpperCamelCase : Optional[Any] = deepcopy(_snake_case )
elif os.path.exists(_snake_case ):
with io.open(_snake_case , "r" , encoding="utf-8" ) as f:
__UpperCamelCase : Optional[int] = json.load(_snake_case )
else:
try:
__UpperCamelCase : Dict = baseaa.urlsafe_baadecode(_snake_case ).decode("utf-8" )
__UpperCamelCase : Optional[Any] = json.loads(_snake_case )
except (UnicodeDecodeError, AttributeError, ValueError):
raise ValueError(
f"Expected a string path to an existing deepspeed config, or a dictionary, or a base64 encoded string. Received: {config_file_or_dict}" )
__UpperCamelCase : List[Any] = config
self.set_stage_and_offload()
def a_ (self ) -> Optional[int]:
# zero stage - this is done as early as possible, before model is created, to allow
# ``is_deepspeed_zero3_enabled`` query and getting to the early deepspeed config object
# during ``zero.Init()`` which needs to know the dtype, and some other hparams.
__UpperCamelCase : Dict = self.get_value("zero_optimization.stage" , -1 )
# offload
__UpperCamelCase : Optional[int] = False
if self.is_zeroa() or self.is_zeroa():
__UpperCamelCase : int = set(["cpu", "nvme"] )
__UpperCamelCase : Any = set(
[
self.get_value("zero_optimization.offload_optimizer.device" ),
self.get_value("zero_optimization.offload_param.device" ),
] )
if len(offload_devices & offload_devices_valid ) > 0:
__UpperCamelCase : Tuple = True
def a_ (self , _UpperCAmelCase ) -> Optional[int]:
__UpperCamelCase : int = self.config
# find the config node of interest if it exists
__UpperCamelCase : Optional[Any] = ds_key_long.split("." )
__UpperCamelCase : Any = nodes.pop()
for node in nodes:
__UpperCamelCase : Optional[Any] = config.get(_snake_case )
if config is None:
return None, ds_key
return config, ds_key
def a_ (self , _UpperCAmelCase , _UpperCAmelCase=None ) -> Any:
__UpperCamelCase : Dict = self.find_config_node(_snake_case )
if config is None:
return default
return config.get(_snake_case , _snake_case )
def a_ (self , _UpperCAmelCase , _UpperCAmelCase=False ) -> Tuple:
__UpperCamelCase : Any = self.config
# find the config node of interest if it exists
__UpperCamelCase : List[str] = ds_key_long.split("." )
for node in nodes:
__UpperCamelCase : Any = config
__UpperCamelCase : Optional[Any] = config.get(_snake_case )
if config is None:
if must_exist:
raise ValueError(f"Can\'t find {ds_key_long} entry in the config: {self.config}" )
else:
return
# if found remove it
if parent_config is not None:
parent_config.pop(_snake_case )
def a_ (self , _UpperCAmelCase ) -> Tuple:
__UpperCamelCase : Any = self.get_value(_snake_case )
return False if value is None else bool(_snake_case )
def a_ (self , _UpperCAmelCase ) -> Union[str, Any]:
__UpperCamelCase : Any = self.get_value(_snake_case )
return False if value is None else not bool(_snake_case )
def a_ (self ) -> int:
return self._stage == 2
def a_ (self ) -> List[Any]:
return self._stage == 3
def a_ (self ) -> List[str]:
return self._offload
class A :
'''simple docstring'''
def __init__(self , _UpperCAmelCase ) -> Union[str, Any]:
__UpperCamelCase : Tuple = engine
def a_ (self , _UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
# runs backpropagation and handles mixed precision
self.engine.backward(_snake_case , **_snake_case )
# Deepspeed's `engine.step` performs the following operations:
# - gradient accumulation check
# - gradient clipping
# - optimizer step
# - zero grad
# - checking overflow
# - lr_scheduler step (only if engine.lr_scheduler is not None)
self.engine.step()
# and this plugin overrides the above calls with no-ops when Accelerate runs under
# Deepspeed, but allows normal functionality for non-Deepspeed cases thus enabling a simple
# training loop that works transparently under many training regimes.
class A ( lowerCAmelCase_ ):
'''simple docstring'''
def __init__(self , _UpperCAmelCase ) -> Union[str, Any]:
super().__init__(_snake_case , device_placement=_snake_case , scaler=_snake_case )
__UpperCamelCase : Optional[Any] = hasattr(self.optimizer , "overflow" )
def a_ (self , _UpperCAmelCase=None ) -> Optional[int]:
pass # `accelerator.backward(loss)` is doing that automatically. Therefore, its implementation is not needed
def a_ (self ) -> Dict:
pass # `accelerator.backward(loss)` is doing that automatically. Therefore, its implementation is not needed
@property
def a_ (self ) -> Any:
if self.__has_overflow__:
return self.optimizer.overflow
return False
class A ( lowerCAmelCase_ ):
'''simple docstring'''
def __init__(self , _UpperCAmelCase , _UpperCAmelCase ) -> List[Any]:
super().__init__(_snake_case , _snake_case )
def a_ (self ) -> Optional[int]:
pass # `accelerator.backward(loss)` is doing that automatically. Therefore, its implementation is not needed
class A :
'''simple docstring'''
def __init__(self , _UpperCAmelCase , _UpperCAmelCase=0.001 , _UpperCAmelCase=0 , **_UpperCAmelCase ) -> Union[str, Any]:
__UpperCamelCase : Optional[int] = params
__UpperCamelCase : Optional[Any] = lr
__UpperCamelCase : Optional[int] = weight_decay
__UpperCamelCase : int = kwargs
class A :
'''simple docstring'''
def __init__(self , _UpperCAmelCase , _UpperCAmelCase=None , _UpperCAmelCase=0 , **_UpperCAmelCase ) -> List[Any]:
__UpperCamelCase : Union[str, Any] = optimizer
__UpperCamelCase : Union[str, Any] = total_num_steps
__UpperCamelCase : Union[str, Any] = warmup_num_steps
__UpperCamelCase : Optional[int] = kwargs
| 298 |
import argparse
import json
from collections import OrderedDict
from pathlib import Path
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import (
SegformerConfig,
SegformerForImageClassification,
SegformerForSemanticSegmentation,
SegformerImageProcessor,
)
from transformers.utils import logging
logging.set_verbosity_info()
a_ :Dict = logging.get_logger(__name__)
def lowercase_ (A : Optional[Any] , A : Any=False ):
snake_case__ : List[Any] = OrderedDict()
for key, value in state_dict.items():
if encoder_only and not key.startswith('head' ):
snake_case__ : str = 'segformer.encoder.' + key
if key.startswith('backbone' ):
snake_case__ : str = key.replace('backbone' , 'segformer.encoder' )
if "patch_embed" in key:
# replace for example patch_embed1 by patch_embeddings.0
snake_case__ : Optional[int] = key[key.find('patch_embed' ) + len('patch_embed' )]
snake_case__ : int = key.replace(F'''patch_embed{idx}''' , F'''patch_embeddings.{int(A )-1}''' )
if "norm" in key:
snake_case__ : Optional[int] = key.replace('norm' , 'layer_norm' )
if "segformer.encoder.layer_norm" in key:
# replace for example layer_norm1 by layer_norm.0
snake_case__ : Tuple = key[key.find('segformer.encoder.layer_norm' ) + len('segformer.encoder.layer_norm' )]
snake_case__ : Union[str, Any] = key.replace(F'''layer_norm{idx}''' , F'''layer_norm.{int(A )-1}''' )
if "layer_norm1" in key:
snake_case__ : List[Any] = key.replace('layer_norm1' , 'layer_norm_1' )
if "layer_norm2" in key:
snake_case__ : List[Any] = key.replace('layer_norm2' , 'layer_norm_2' )
if "block" in key:
# replace for example block1 by block.0
snake_case__ : List[Any] = key[key.find('block' ) + len('block' )]
snake_case__ : List[Any] = key.replace(F'''block{idx}''' , F'''block.{int(A )-1}''' )
if "attn.q" in key:
snake_case__ : int = key.replace('attn.q' , 'attention.self.query' )
if "attn.proj" in key:
snake_case__ : str = key.replace('attn.proj' , 'attention.output.dense' )
if "attn" in key:
snake_case__ : Optional[int] = key.replace('attn' , 'attention.self' )
if "fc1" in key:
snake_case__ : str = key.replace('fc1' , 'dense1' )
if "fc2" in key:
snake_case__ : Dict = key.replace('fc2' , 'dense2' )
if "linear_pred" in key:
snake_case__ : Union[str, Any] = key.replace('linear_pred' , 'classifier' )
if "linear_fuse" in key:
snake_case__ : List[str] = key.replace('linear_fuse.conv' , 'linear_fuse' )
snake_case__ : List[Any] = key.replace('linear_fuse.bn' , 'batch_norm' )
if "linear_c" in key:
# replace for example linear_c4 by linear_c.3
snake_case__ : Optional[int] = key[key.find('linear_c' ) + len('linear_c' )]
snake_case__ : Tuple = key.replace(F'''linear_c{idx}''' , F'''linear_c.{int(A )-1}''' )
if key.startswith('head' ):
snake_case__ : Tuple = key.replace('head' , 'classifier' )
snake_case__ : Optional[int] = value
return new_state_dict
def lowercase_ (A : Tuple , A : Optional[int] ):
# for each of the encoder blocks:
for i in range(config.num_encoder_blocks ):
for j in range(config.depths[i] ):
# read in weights + bias of keys and values (which is a single matrix in the original implementation)
snake_case__ : List[str] = state_dict.pop(F'''segformer.encoder.block.{i}.{j}.attention.self.kv.weight''' )
snake_case__ : Optional[Any] = state_dict.pop(F'''segformer.encoder.block.{i}.{j}.attention.self.kv.bias''' )
# next, add keys and values (in that order) to the state dict
snake_case__ : str = kv_weight[
: config.hidden_sizes[i], :
]
snake_case__ : Dict = kv_bias[: config.hidden_sizes[i]]
snake_case__ : List[str] = kv_weight[
config.hidden_sizes[i] :, :
]
snake_case__ : List[Any] = kv_bias[
config.hidden_sizes[i] :
]
def lowercase_ ():
snake_case__ : Union[str, Any] = 'http://images.cocodataset.org/val2017/000000039769.jpg'
snake_case__ : Dict = Image.open(requests.get(A , stream=A ).raw )
return image
@torch.no_grad()
def lowercase_ (A : Any , A : Union[str, Any] , A : Optional[Any] ):
snake_case__ : List[str] = SegformerConfig()
snake_case__ : Dict = False
# set attributes based on model_name
snake_case__ : Optional[int] = 'huggingface/label-files'
if "segformer" in model_name:
snake_case__ : str = model_name[len('segformer.' ) : len('segformer.' ) + 2]
if "ade" in model_name:
snake_case__ : Optional[int] = 1_5_0
snake_case__ : int = 'ade20k-id2label.json'
snake_case__ : List[Any] = (1, 1_5_0, 1_2_8, 1_2_8)
elif "city" in model_name:
snake_case__ : str = 1_9
snake_case__ : List[str] = 'cityscapes-id2label.json'
snake_case__ : Optional[Any] = (1, 1_9, 1_2_8, 1_2_8)
else:
raise ValueError(F'''Model {model_name} not supported''' )
elif "mit" in model_name:
snake_case__ : str = True
snake_case__ : Union[str, Any] = model_name[4:6]
snake_case__ : Optional[Any] = 1_0_0_0
snake_case__ : Optional[int] = 'imagenet-1k-id2label.json'
snake_case__ : List[Any] = (1, 1_0_0_0)
else:
raise ValueError(F'''Model {model_name} not supported''' )
# set config attributes
snake_case__ : str = json.load(open(hf_hub_download(A , A , repo_type='dataset' ) , 'r' ) )
snake_case__ : List[Any] = {int(A ): v for k, v in idalabel.items()}
snake_case__ : Union[str, Any] = idalabel
snake_case__ : Tuple = {v: k for k, v in idalabel.items()}
if size == "b0":
pass
elif size == "b1":
snake_case__ : List[Any] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : Tuple = 2_5_6
elif size == "b2":
snake_case__ : List[str] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : int = 7_6_8
snake_case__ : List[Any] = [3, 4, 6, 3]
elif size == "b3":
snake_case__ : Optional[Any] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : int = 7_6_8
snake_case__ : Optional[Any] = [3, 4, 1_8, 3]
elif size == "b4":
snake_case__ : str = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : Optional[Any] = 7_6_8
snake_case__ : Union[str, Any] = [3, 8, 2_7, 3]
elif size == "b5":
snake_case__ : List[str] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : Optional[Any] = 7_6_8
snake_case__ : Any = [3, 6, 4_0, 3]
else:
raise ValueError(F'''Size {size} not supported''' )
# load image processor (only resize + normalize)
snake_case__ : Dict = SegformerImageProcessor(
image_scale=(5_1_2, 5_1_2) , keep_ratio=A , align=A , do_random_crop=A )
# prepare image
snake_case__ : List[str] = prepare_img()
snake_case__ : Dict = image_processor(images=A , return_tensors='pt' ).pixel_values
logger.info(F'''Converting model {model_name}...''' )
# load original state dict
if encoder_only:
snake_case__ : Tuple = torch.load(A , map_location=torch.device('cpu' ) )
else:
snake_case__ : int = torch.load(A , map_location=torch.device('cpu' ) )['state_dict']
# rename keys
snake_case__ : List[Any] = rename_keys(A , encoder_only=A )
if not encoder_only:
del state_dict["decode_head.conv_seg.weight"]
del state_dict["decode_head.conv_seg.bias"]
# key and value matrices need special treatment
read_in_k_v(A , A )
# create HuggingFace model and load state dict
if encoder_only:
snake_case__ : str = False
snake_case__ : List[Any] = SegformerForImageClassification(A )
else:
snake_case__ : Dict = SegformerForSemanticSegmentation(A )
model.load_state_dict(A )
model.eval()
# forward pass
snake_case__ : int = model(A )
snake_case__ : Any = outputs.logits
# set expected_slice based on model name
# ADE20k checkpoints
if model_name == "segformer.b0.512x512.ade.160k":
snake_case__ : Dict = torch.tensor(
[
[[-4.6310, -5.5232, -6.2356], [-5.1921, -6.1444, -6.5996], [-5.4424, -6.2790, -6.7574]],
[[-12.1391, -13.3122, -13.9554], [-12.8732, -13.9352, -14.3563], [-12.9438, -13.8226, -14.2513]],
[[-12.5134, -13.4686, -14.4915], [-12.8669, -14.4343, -14.7758], [-13.2523, -14.5819, -15.0694]],
] )
elif model_name == "segformer.b1.512x512.ade.160k":
snake_case__ : Optional[int] = torch.tensor(
[
[[-7.5820, -8.7231, -8.3215], [-8.0600, -10.3529, -10.0304], [-7.5208, -9.4103, -9.6239]],
[[-12.6918, -13.8994, -13.7137], [-13.3196, -15.7523, -15.4789], [-12.9343, -14.8757, -14.9689]],
[[-11.1911, -11.9421, -11.3243], [-11.3342, -13.6839, -13.3581], [-10.3909, -12.1832, -12.4858]],
] )
elif model_name == "segformer.b2.512x512.ade.160k":
snake_case__ : List[Any] = torch.tensor(
[
[[-11.8173, -14.3850, -16.3128], [-14.5648, -16.5804, -18.6568], [-14.7223, -15.7387, -18.4218]],
[[-15.7290, -17.9171, -19.4423], [-18.3105, -19.9448, -21.4661], [-17.9296, -18.6497, -20.7910]],
[[-15.0783, -17.0336, -18.2789], [-16.8771, -18.6870, -20.1612], [-16.2454, -17.1426, -19.5055]],
] )
elif model_name == "segformer.b3.512x512.ade.160k":
snake_case__ : Union[str, Any] = torch.tensor(
[
[[-9.0878, -10.2081, -10.1891], [-9.3144, -10.7941, -10.9843], [-9.2294, -10.3855, -10.5704]],
[[-12.2316, -13.9068, -13.6102], [-12.9161, -14.3702, -14.3235], [-12.5233, -13.7174, -13.7932]],
[[-14.6275, -15.2490, -14.9727], [-14.3400, -15.9687, -16.2827], [-14.1484, -15.4033, -15.8937]],
] )
elif model_name == "segformer.b4.512x512.ade.160k":
snake_case__ : Dict = torch.tensor(
[
[[-12.3144, -13.2447, -14.0802], [-13.3614, -14.5816, -15.6117], [-13.3340, -14.4433, -16.2219]],
[[-19.2781, -20.4128, -20.7506], [-20.6153, -21.6566, -22.0998], [-19.9800, -21.0430, -22.1494]],
[[-18.8739, -19.7804, -21.1834], [-20.1233, -21.6765, -23.2944], [-20.0315, -21.2641, -23.6944]],
] )
elif model_name == "segformer.b5.640x640.ade.160k":
snake_case__ : List[Any] = torch.tensor(
[
[[-9.5524, -12.0835, -11.7348], [-10.5229, -13.6446, -14.5662], [-9.5842, -12.8851, -13.9414]],
[[-15.3432, -17.5323, -17.0818], [-16.3330, -18.9255, -19.2101], [-15.1340, -17.7848, -18.3971]],
[[-12.6072, -14.9486, -14.6631], [-13.7629, -17.0907, -17.7745], [-12.7899, -16.1695, -17.1671]],
] )
# Cityscapes checkpoints
elif model_name == "segformer.b0.1024x1024.city.160k":
snake_case__ : str = torch.tensor(
[
[[-11.9295, -13.4057, -14.8106], [-13.3431, -14.8179, -15.3781], [-14.2836, -15.5942, -16.1588]],
[[-11.4906, -12.8067, -13.6564], [-13.1189, -14.0500, -14.1543], [-13.8748, -14.5136, -14.8789]],
[[0.5374, 0.1067, -0.4742], [0.1141, -0.2255, -0.7099], [-0.3000, -0.5924, -1.3105]],
] )
elif model_name == "segformer.b0.512x1024.city.160k":
snake_case__ : Tuple = torch.tensor(
[
[[-7.8217, -9.8767, -10.1717], [-9.4438, -10.9058, -11.4047], [-9.7939, -12.3495, -12.1079]],
[[-7.1514, -9.5336, -10.0860], [-9.7776, -11.6822, -11.8439], [-10.1411, -12.7655, -12.8972]],
[[0.3021, 0.0805, -0.2310], [-0.0328, -0.1605, -0.2714], [-0.1408, -0.5477, -0.6976]],
] )
elif model_name == "segformer.b0.640x1280.city.160k":
snake_case__ : Any = torch.tensor(
[
[
[-1.1_372e01, -1.2_787e01, -1.3_477e01],
[-1.2_536e01, -1.4_194e01, -1.4_409e01],
[-1.3_217e01, -1.4_888e01, -1.5_327e01],
],
[
[-1.4_791e01, -1.7_122e01, -1.8_277e01],
[-1.7_163e01, -1.9_192e01, -1.9_533e01],
[-1.7_897e01, -1.9_991e01, -2.0_315e01],
],
[
[7.6_723e-01, 4.1_921e-01, -7.7_878e-02],
[4.7_772e-01, 9.5_557e-03, -2.8_082e-01],
[3.6_032e-01, -2.4_826e-01, -5.1_168e-01],
],
] )
elif model_name == "segformer.b0.768x768.city.160k":
snake_case__ : Optional[int] = torch.tensor(
[
[[-9.4959, -11.3087, -11.7479], [-11.0025, -12.6540, -12.3319], [-11.4064, -13.0487, -12.9905]],
[[-9.8905, -11.3084, -12.0854], [-11.1726, -12.7698, -12.9583], [-11.5985, -13.3278, -14.1774]],
[[0.2213, 0.0192, -0.2466], [-0.1731, -0.4213, -0.4874], [-0.3126, -0.6541, -1.1389]],
] )
elif model_name == "segformer.b1.1024x1024.city.160k":
snake_case__ : Union[str, Any] = torch.tensor(
[
[[-13.5748, -13.9111, -12.6500], [-14.3500, -15.3683, -14.2328], [-14.7532, -16.0424, -15.6087]],
[[-17.1651, -15.8725, -12.9653], [-17.2580, -17.3718, -14.8223], [-16.6058, -16.8783, -16.7452]],
[[-3.6456, -3.0209, -1.4203], [-3.0797, -3.1959, -2.0000], [-1.8757, -1.9217, -1.6997]],
] )
elif model_name == "segformer.b2.1024x1024.city.160k":
snake_case__ : List[str] = torch.tensor(
[
[[-16.0976, -16.4856, -17.3962], [-16.6234, -19.0342, -19.7685], [-16.0900, -18.0661, -19.1180]],
[[-18.4750, -18.8488, -19.5074], [-19.4030, -22.1570, -22.5977], [-19.1191, -20.8486, -22.3783]],
[[-4.5178, -5.5037, -6.5109], [-5.0884, -7.2174, -8.0334], [-4.4156, -5.8117, -7.2970]],
] )
elif model_name == "segformer.b3.1024x1024.city.160k":
snake_case__ : List[Any] = torch.tensor(
[
[[-14.2081, -14.4732, -14.1977], [-14.5867, -16.4423, -16.6356], [-13.4441, -14.9685, -16.8696]],
[[-14.4576, -14.7073, -15.0451], [-15.0816, -17.6237, -17.9873], [-14.4213, -16.0199, -18.5992]],
[[-4.7349, -4.9588, -5.0966], [-4.3210, -6.9325, -7.2591], [-3.4312, -4.7484, -7.1917]],
] )
elif model_name == "segformer.b4.1024x1024.city.160k":
snake_case__ : str = torch.tensor(
[
[[-11.7737, -11.9526, -11.3273], [-13.6692, -14.4574, -13.8878], [-13.8937, -14.6924, -15.9345]],
[[-14.6706, -14.5330, -14.1306], [-16.1502, -16.8180, -16.4269], [-16.8338, -17.8939, -20.1746]],
[[1.0491, 0.8289, 1.0310], [1.1044, 0.5219, 0.8055], [1.0899, 0.6926, 0.5590]],
] )
elif model_name == "segformer.b5.1024x1024.city.160k":
snake_case__ : List[str] = torch.tensor(
[
[[-12.5641, -13.4777, -13.0684], [-13.9587, -15.8983, -16.6557], [-13.3109, -15.7350, -16.3141]],
[[-14.7074, -15.4352, -14.5944], [-16.6353, -18.1663, -18.6120], [-15.1702, -18.0329, -18.1547]],
[[-1.7990, -2.0951, -1.7784], [-2.6397, -3.8245, -3.9686], [-1.5264, -2.8126, -2.9316]],
] )
else:
snake_case__ : Tuple = logits.argmax(-1 ).item()
print('Predicted class:' , model.config.idalabel[predicted_class_idx] )
# verify logits
if not encoder_only:
assert logits.shape == expected_shape
assert torch.allclose(logits[0, :3, :3, :3] , A , atol=1e-2 )
# finally, save model and image processor
logger.info(F'''Saving PyTorch model and image processor to {pytorch_dump_folder_path}...''' )
Path(A ).mkdir(exist_ok=A )
model.save_pretrained(A )
image_processor.save_pretrained(A )
if __name__ == "__main__":
a_ :Optional[int] = argparse.ArgumentParser()
parser.add_argument(
"--model_name",
default="segformer.b0.512x512.ade.160k",
type=str,
help="Name of the model you'd like to convert.",
)
parser.add_argument(
"--checkpoint_path", default=None, type=str, help="Path to the original PyTorch checkpoint (.pth file)."
)
parser.add_argument(
"--pytorch_dump_folder_path", default=None, type=str, help="Path to the folder to output PyTorch model."
)
a_ :Union[str, Any] = parser.parse_args()
convert_segformer_checkpoint(args.model_name, args.checkpoint_path, args.pytorch_dump_folder_path)
| 277 | 0 |
import argparse
import json
from collections import OrderedDict
from pathlib import Path
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import PoolFormerConfig, PoolFormerForImageClassification, PoolFormerImageProcessor
from transformers.utils import logging
logging.set_verbosity_info()
_snake_case = logging.get_logger(__name__)
def lowerCAmelCase_ ( snake_case_,snake_case_,snake_case_,snake_case_ ):
_A : str = original_name.split(""".""" )[0]
_A : List[str] = key.split(""".""" )
_A : Tuple = int(key_list[key_list.index(snake_case_ ) - 2] )
_A : Dict = int(key_list[key_list.index(snake_case_ ) - 1] )
_A : Any = orig_block_num - offset
_A : Dict = key.replace(f'''{orig_block_num}.{layer_num}.{original_name}''',f'''block.{new_block_num}.{layer_num}.{new_name}''' )
return key
def lowerCAmelCase_ ( snake_case_ ):
_A : Dict = OrderedDict()
_A : int = 0, 0
for key, value in state_dict.items():
if key.startswith("""network""" ):
_A : str = key.replace("""network""","""poolformer.encoder""" )
if "proj" in key:
# Works for the first embedding as well as the internal embedding layers
if key.endswith("""bias""" ) and "patch_embed" not in key:
patch_emb_offset += 1
_A : Dict = key[: key.find("""proj""" )]
_A : int = key.replace(snake_case_,f'''patch_embeddings.{total_embed_found}.''' )
_A : Dict = key.replace("""proj""","""projection""" )
if key.endswith("""bias""" ):
total_embed_found += 1
if "patch_embeddings" in key:
_A : Dict = 'poolformer.encoder.' + key
if "mlp.fc1" in key:
_A : Dict = replace_key_with_offset(snake_case_,snake_case_,"""mlp.fc1""","""output.conv1""" )
if "mlp.fc2" in key:
_A : List[Any] = replace_key_with_offset(snake_case_,snake_case_,"""mlp.fc2""","""output.conv2""" )
if "norm1" in key:
_A : List[str] = replace_key_with_offset(snake_case_,snake_case_,"""norm1""","""before_norm""" )
if "norm2" in key:
_A : Dict = replace_key_with_offset(snake_case_,snake_case_,"""norm2""","""after_norm""" )
if "layer_scale_1" in key:
_A : Optional[Any] = replace_key_with_offset(snake_case_,snake_case_,"""layer_scale_1""","""layer_scale_1""" )
if "layer_scale_2" in key:
_A : Tuple = replace_key_with_offset(snake_case_,snake_case_,"""layer_scale_2""","""layer_scale_2""" )
if "head" in key:
_A : Union[str, Any] = key.replace("""head""","""classifier""" )
_A : int = value
return new_state_dict
def lowerCAmelCase_ ( ):
_A : Dict = 'http://images.cocodataset.org/val2017/000000039769.jpg'
_A : List[str] = Image.open(requests.get(snake_case_,stream=snake_case_ ).raw )
return image
@torch.no_grad()
def lowerCAmelCase_ ( snake_case_,snake_case_,snake_case_ ):
_A : Any = PoolFormerConfig()
# set attributes based on model_name
_A : str = 'huggingface/label-files'
_A : List[Any] = model_name[-3:]
_A : Optional[int] = 1000
_A : Tuple = 'imagenet-1k-id2label.json'
_A : Union[str, Any] = (1, 1000)
# set config attributes
_A : Optional[int] = json.load(open(hf_hub_download(snake_case_,snake_case_,repo_type="""dataset""" ),"""r""" ) )
_A : Any = {int(snake_case_ ): v for k, v in idalabel.items()}
_A : int = idalabel
_A : List[str] = {v: k for k, v in idalabel.items()}
if size == "s12":
_A : str = [2, 2, 6, 2]
_A : Optional[int] = [64, 128, 320, 512]
_A : Union[str, Any] = 4.0
_A : str = 0.9
elif size == "s24":
_A : List[str] = [4, 4, 12, 4]
_A : Any = [64, 128, 320, 512]
_A : List[str] = 4.0
_A : Any = 0.9
elif size == "s36":
_A : int = [6, 6, 18, 6]
_A : int = [64, 128, 320, 512]
_A : Dict = 4.0
_A : Dict = 1e-6
_A : Any = 0.9
elif size == "m36":
_A : int = [6, 6, 18, 6]
_A : Dict = [96, 192, 384, 768]
_A : Tuple = 4.0
_A : Union[str, Any] = 1e-6
_A : str = 0.95
elif size == "m48":
_A : Union[str, Any] = [8, 8, 24, 8]
_A : Optional[int] = [96, 192, 384, 768]
_A : Optional[Any] = 4.0
_A : int = 1e-6
_A : Optional[Any] = 0.95
else:
raise ValueError(f'''Size {size} not supported''' )
# load image processor
_A : List[Any] = PoolFormerImageProcessor(crop_pct=snake_case_ )
# Prepare image
_A : Optional[Any] = prepare_img()
_A : Optional[Any] = image_processor(images=snake_case_,return_tensors="""pt""" ).pixel_values
logger.info(f'''Converting model {model_name}...''' )
# load original state dict
_A : Optional[Any] = torch.load(snake_case_,map_location=torch.device("""cpu""" ) )
# rename keys
_A : Union[str, Any] = rename_keys(snake_case_ )
# create HuggingFace model and load state dict
_A : str = PoolFormerForImageClassification(snake_case_ )
model.load_state_dict(snake_case_ )
model.eval()
# Define image processor
_A : Union[str, Any] = PoolFormerImageProcessor(crop_pct=snake_case_ )
_A : Any = image_processor(images=prepare_img(),return_tensors="""pt""" ).pixel_values
# forward pass
_A : List[Any] = model(snake_case_ )
_A : List[Any] = outputs.logits
# define expected logit slices for different models
if size == "s12":
_A : List[Any] = torch.tensor([-0.30_45, -0.67_58, -0.48_69] )
elif size == "s24":
_A : Optional[int] = torch.tensor([0.44_02, -0.13_74, -0.80_45] )
elif size == "s36":
_A : Optional[Any] = torch.tensor([-0.60_80, -0.51_33, -0.58_98] )
elif size == "m36":
_A : Any = torch.tensor([0.39_52, 0.22_63, -1.26_68] )
elif size == "m48":
_A : Optional[int] = torch.tensor([0.11_67, -0.06_56, -0.34_23] )
else:
raise ValueError(f'''Size {size} not supported''' )
# verify logits
assert logits.shape == expected_shape
assert torch.allclose(logits[0, :3],snake_case_,atol=1e-2 )
# finally, save model and image processor
logger.info(f'''Saving PyTorch model and image processor to {pytorch_dump_folder_path}...''' )
Path(snake_case_ ).mkdir(exist_ok=snake_case_ )
model.save_pretrained(snake_case_ )
print(f'''Saving image processor to {pytorch_dump_folder_path}''' )
image_processor.save_pretrained(snake_case_ )
if __name__ == "__main__":
_snake_case = argparse.ArgumentParser()
parser.add_argument(
"--model_name",
default="poolformer_s12",
type=str,
help="Name of the model you'd like to convert.",
)
parser.add_argument(
"--checkpoint_path", default=None, type=str, help="Path to the original PyTorch checkpoint (.pth file)."
)
parser.add_argument(
"--pytorch_dump_folder_path", default=None, type=str, help="Path to the folder to output PyTorch model."
)
_snake_case = parser.parse_args()
convert_poolformer_checkpoint(args.model_name, args.checkpoint_path, args.pytorch_dump_folder_path)
| 26 |
import argparse
import json
import os
import fairseq
import torch
from fairseq.data import Dictionary
from transformers import (
WavaVecaConfig,
WavaVecaCTCTokenizer,
WavaVecaFeatureExtractor,
WavaVecaForCTC,
WavaVecaForPreTraining,
WavaVecaProcessor,
logging,
)
from transformers.models.wavaveca.modeling_wavaveca import WavaVecaForSequenceClassification
logging.set_verbosity_info()
a_ :List[Any] = logging.get_logger(__name__)
a_ :List[Any] = {
"post_extract_proj": "feature_projection.projection",
"encoder.pos_conv.0": "encoder.pos_conv_embed.conv",
"self_attn.k_proj": "encoder.layers.*.attention.k_proj",
"self_attn.v_proj": "encoder.layers.*.attention.v_proj",
"self_attn.q_proj": "encoder.layers.*.attention.q_proj",
"self_attn.out_proj": "encoder.layers.*.attention.out_proj",
"self_attn_layer_norm": "encoder.layers.*.layer_norm",
"fc1": "encoder.layers.*.feed_forward.intermediate_dense",
"fc2": "encoder.layers.*.feed_forward.output_dense",
"final_layer_norm": "encoder.layers.*.final_layer_norm",
"encoder.layer_norm": "encoder.layer_norm",
"adapter_layer": "encoder.layers.*.adapter_layer",
"w2v_model.layer_norm": "feature_projection.layer_norm",
"quantizer.weight_proj": "quantizer.weight_proj",
"quantizer.vars": "quantizer.codevectors",
"project_q": "project_q",
"final_proj": "project_hid",
"w2v_encoder.proj": "lm_head",
"mask_emb": "masked_spec_embed",
"pooling_layer.linear": "projector",
"pooling_layer.projection": "classifier",
}
a_ :List[Any] = [
"lm_head",
"quantizer.weight_proj",
"quantizer.codevectors",
"project_q",
"project_hid",
"projector",
"classifier",
]
def lowercase_ (A : Dict ):
snake_case__ : Optional[Any] = {}
with open(A , 'r' ) as file:
for line_number, line in enumerate(A ):
snake_case__ : Dict = line.strip()
if line:
snake_case__ : int = line.split()
snake_case__ : List[str] = line_number
snake_case__ : Dict = words[0]
snake_case__ : Optional[Any] = value
return result
def lowercase_ (A : int , A : int , A : Optional[int] , A : Optional[Any] , A : Tuple ):
for attribute in key.split('.' ):
snake_case__ : Optional[int] = getattr(A , A )
snake_case__ : Union[str, Any] = None
for param_key in PARAM_MAPPING.keys():
if full_name.endswith(A ):
snake_case__ : List[str] = PARAM_MAPPING[full_name.split('.' )[-1]]
snake_case__ : Dict = 'param'
if weight_type is not None and weight_type != "param":
snake_case__ : Union[str, Any] = getattr(A , A ).shape
elif weight_type is not None and weight_type == "param":
snake_case__ : Optional[int] = hf_pointer
for attribute in hf_param_name.split('.' ):
snake_case__ : Optional[Any] = getattr(A , A )
snake_case__ : Dict = shape_pointer.shape
# let's reduce dimension
snake_case__ : List[Any] = value[0]
else:
snake_case__ : Union[str, Any] = hf_pointer.shape
if hf_shape != value.shape:
raise ValueError(
F'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
F''' {value.shape} for {full_name}''' )
if weight_type == "weight":
snake_case__ : Any = value
elif weight_type == "weight_g":
snake_case__ : List[Any] = value
elif weight_type == "weight_v":
snake_case__ : Any = value
elif weight_type == "bias":
snake_case__ : List[Any] = value
elif weight_type == "param":
for attribute in hf_param_name.split('.' ):
snake_case__ : int = getattr(A , A )
snake_case__ : Optional[int] = value
else:
snake_case__ : Optional[Any] = value
logger.info(F'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def lowercase_ (A : Tuple , A : List[Any] , A : int , A : str , A : Tuple ):
snake_case__ : Optional[int] = None
for param_key in PARAM_MAPPING.keys():
if full_name.endswith(A ):
snake_case__ : List[str] = PARAM_MAPPING[full_name.split('.' )[-1]]
snake_case__ : str = 'param'
if weight_type is not None and weight_type != "param":
snake_case__ : int = '.'.join([key, weight_type] )
elif weight_type is not None and weight_type == "param":
snake_case__ : Any = '.'.join([key, hf_param_name] )
else:
snake_case__ : Dict = key
snake_case__ : List[str] = value if 'lm_head' in full_key else value[0]
a_ :List[str] = {
"W_a": "linear_1.weight",
"W_b": "linear_2.weight",
"b_a": "linear_1.bias",
"b_b": "linear_2.bias",
"ln_W": "norm.weight",
"ln_b": "norm.bias",
}
def lowercase_ (A : str , A : Optional[Any] , A : Optional[Any]=None , A : List[str]=None ):
snake_case__ : Optional[int] = False
for key, mapped_key in MAPPING.items():
snake_case__ : Tuple = 'wav2vec2.' + mapped_key if mapped_key not in TOP_LEVEL_KEYS else mapped_key
if key in name or key.split('w2v_model.' )[-1] == name.split('.' )[0]:
snake_case__ : Optional[int] = True
if "*" in mapped_key:
snake_case__ : List[Any] = name.split(A )[0].split('.' )[-2]
snake_case__ : Union[str, Any] = mapped_key.replace('*' , A )
if "weight_g" in name:
snake_case__ : Tuple = 'weight_g'
elif "weight_v" in name:
snake_case__ : List[str] = 'weight_v'
elif "bias" in name:
snake_case__ : Dict = 'bias'
elif "weight" in name:
# TODO: don't match quantizer.weight_proj
snake_case__ : Optional[int] = 'weight'
else:
snake_case__ : str = None
if hf_dict is not None:
rename_dict(A , A , A , A , A )
else:
set_recursively(A , A , A , A , A )
return is_used
return is_used
def lowercase_ (A : Optional[Any] , A : Dict , A : Optional[int] ):
snake_case__ : Dict = []
snake_case__ : Tuple = fairseq_model.state_dict()
snake_case__ : str = hf_model.wavaveca.feature_extractor
for name, value in fairseq_dict.items():
snake_case__ : str = False
if "conv_layers" in name:
load_conv_layer(
A , A , A , A , hf_model.config.feat_extract_norm == 'group' , )
snake_case__ : Any = True
else:
snake_case__ : Dict = load_wavaveca_layer(A , A , A )
if not is_used:
unused_weights.append(A )
logger.warning(F'''Unused weights: {unused_weights}''' )
def lowercase_ (A : Dict , A : Optional[Any] , A : Tuple , A : str , A : List[str] ):
snake_case__ : List[Any] = full_name.split('conv_layers.' )[-1]
snake_case__ : List[str] = name.split('.' )
snake_case__ : List[Any] = int(items[0] )
snake_case__ : str = int(items[1] )
if type_id == 0:
if "bias" in name:
if value.shape != feature_extractor.conv_layers[layer_id].conv.bias.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.''' )
snake_case__ : Any = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
if value.shape != feature_extractor.conv_layers[layer_id].conv.weight.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.''' )
snake_case__ : str = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
if value.shape != feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape} was found.''' )
snake_case__ : str = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
if value.shape != feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape} was found.''' )
snake_case__ : int = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(A )
@torch.no_grad()
def lowercase_ (A : Union[str, Any] , A : str , A : Tuple=None , A : List[str]=None , A : Any=True , A : Optional[int]=False ):
if config_path is not None:
snake_case__ : List[Any] = WavaVecaConfig.from_pretrained(A )
else:
snake_case__ : List[Any] = WavaVecaConfig()
if is_seq_class:
snake_case__ : Dict = read_txt_into_dict(A )
snake_case__ : Any = idalabel
snake_case__ : Union[str, Any] = WavaVecaForSequenceClassification(A )
snake_case__ : Any = WavaVecaFeatureExtractor(
feature_size=1 , sampling_rate=1_6_0_0_0 , padding_value=0 , do_normalize=A , return_attention_mask=A , )
feature_extractor.save_pretrained(A )
elif is_finetuned:
if dict_path:
snake_case__ : str = Dictionary.load(A )
# important change bos & pad token id since CTC symbol is <pad> and
# not <s> as in fairseq
snake_case__ : List[str] = target_dict.pad_index
snake_case__ : Optional[int] = target_dict.bos_index
snake_case__ : Optional[int] = target_dict.eos_index
snake_case__ : List[Any] = len(target_dict.symbols )
snake_case__ : str = os.path.join(A , 'vocab.json' )
if not os.path.isdir(A ):
logger.error('--pytorch_dump_folder_path ({}) should be a directory'.format(A ) )
return
os.makedirs(A , exist_ok=A )
snake_case__ : Optional[Any] = target_dict.indices
# fairseq has the <pad> and <s> switched
snake_case__ : Optional[Any] = 0
snake_case__ : Union[str, Any] = 1
with open(A , 'w' , encoding='utf-8' ) as vocab_handle:
json.dump(A , A )
snake_case__ : List[Any] = WavaVecaCTCTokenizer(
A , unk_token=target_dict.unk_word , pad_token=target_dict.pad_word , bos_token=target_dict.bos_word , eos_token=target_dict.eos_word , word_delimiter_token='|' , do_lower_case=A , )
snake_case__ : str = True if config.feat_extract_norm == 'layer' else False
snake_case__ : Optional[Any] = WavaVecaFeatureExtractor(
feature_size=1 , sampling_rate=1_6_0_0_0 , padding_value=0 , do_normalize=A , return_attention_mask=A , )
snake_case__ : Union[str, Any] = WavaVecaProcessor(feature_extractor=A , tokenizer=A )
processor.save_pretrained(A )
snake_case__ : str = WavaVecaForCTC(A )
else:
snake_case__ : int = WavaVecaForPreTraining(A )
if is_finetuned or is_seq_class:
snake_case__ , snake_case__ , snake_case__ : str = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] , arg_overrides={'data': '/'.join(dict_path.split('/' )[:-1] )} )
else:
snake_case__ : Tuple = argparse.Namespace(task='audio_pretraining' )
snake_case__ : str = fairseq.tasks.setup_task(A )
snake_case__ , snake_case__ , snake_case__ : Any = fairseq.checkpoint_utils.load_model_ensemble_and_task([checkpoint_path] , task=A )
snake_case__ : List[Any] = model[0].eval()
recursively_load_weights(A , A , not is_finetuned )
hf_wavavec.save_pretrained(A )
if __name__ == "__main__":
a_ :List[Any] = argparse.ArgumentParser()
parser.add_argument("--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model.")
parser.add_argument("--checkpoint_path", default=None, type=str, help="Path to fairseq checkpoint")
parser.add_argument("--dict_path", default=None, type=str, help="Path to dict of fine-tuned model")
parser.add_argument("--config_path", default=None, type=str, help="Path to hf config.json of model to convert")
parser.add_argument(
"--not_finetuned", action="store_true", help="Whether the model to convert is a fine-tuned model or not"
)
parser.add_argument(
"--is_seq_class",
action="store_true",
help="Whether the model to convert is a fine-tuned sequence classification model or not",
)
a_ :str = parser.parse_args()
a_ :Tuple = not args.not_finetuned and not args.is_seq_class
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.config_path,
args.dict_path,
is_finetuned,
args.is_seq_class,
)
| 277 | 0 |
import datasets
UpperCAmelCase__ = "\\n@InProceedings{conneau2018xnli,\n author = \"Conneau, Alexis\n and Rinott, Ruty\n and Lample, Guillaume\n and Williams, Adina\n and Bowman, Samuel R.\n and Schwenk, Holger\n and Stoyanov, Veselin\",\n title = \"XNLI: Evaluating Cross-lingual Sentence Representations\",\n booktitle = \"Proceedings of the 2018 Conference on Empirical Methods\n in Natural Language Processing\",\n year = \"2018\",\n publisher = \"Association for Computational Linguistics\",\n location = \"Brussels, Belgium\",\n}\n"
UpperCAmelCase__ = "\\nXNLI is a subset of a few thousand examples from MNLI which has been translated\ninto a 14 different languages (some low-ish resource). As with MNLI, the goal is\nto predict textual entailment (does sentence A imply/contradict/neither sentence\nB) and is a classification task (given two sentences, predict one of three\nlabels).\n"
UpperCAmelCase__ = "\nComputes XNLI score which is just simple accuracy.\nArgs:\n predictions: Predicted labels.\n references: Ground truth labels.\nReturns:\n 'accuracy': accuracy\nExamples:\n\n >>> predictions = [0, 1]\n >>> references = [0, 1]\n >>> xnli_metric = datasets.load_metric(\"xnli\")\n >>> results = xnli_metric.compute(predictions=predictions, references=references)\n >>> print(results)\n {'accuracy': 1.0}\n"
def A ( _UpperCAmelCase : Any , _UpperCAmelCase : Union[str, Any] ) -> Any:
'''simple docstring'''
return (preds == labels).mean()
@datasets.utils.file_utils.add_start_docstrings(_DESCRIPTION , _KWARGS_DESCRIPTION )
class __lowerCAmelCase ( datasets.Metric ):
def _lowerCamelCase ( self : List[str]) -> int:
"""simple docstring"""
return datasets.MetricInfo(
description=_DESCRIPTION , citation=_CITATION , inputs_description=_KWARGS_DESCRIPTION , features=datasets.Features(
{
'predictions': datasets.Value('int64' if self.config_name != 'sts-b' else 'float32'),
'references': datasets.Value('int64' if self.config_name != 'sts-b' else 'float32'),
}) , codebase_urls=[] , reference_urls=[] , format='numpy' , )
def _lowerCamelCase ( self : List[str] , A : Dict , A : Optional[Any]) -> List[str]:
"""simple docstring"""
return {"accuracy": simple_accuracy(_snake_case , _snake_case)}
| 339 |
from typing import Dict, List
from nltk.translate import gleu_score
import datasets
from datasets import MetricInfo
a_ :Any = "\\n@misc{wu2016googles,\n title={Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation},\n author={Yonghui Wu and Mike Schuster and Zhifeng Chen and Quoc V. Le and Mohammad Norouzi and Wolfgang Macherey\n and Maxim Krikun and Yuan Cao and Qin Gao and Klaus Macherey and Jeff Klingner and Apurva Shah and Melvin\n Johnson and Xiaobing Liu and Łukasz Kaiser and Stephan Gouws and Yoshikiyo Kato and Taku Kudo and Hideto\n Kazawa and Keith Stevens and George Kurian and Nishant Patil and Wei Wang and Cliff Young and\n Jason Smith and Jason Riesa and Alex Rudnick and Oriol Vinyals and Greg Corrado and Macduff Hughes\n and Jeffrey Dean},\n year={2016},\n eprint={1609.08144},\n archivePrefix={arXiv},\n primaryClass={cs.CL}\n}\n"
a_ :List[str] = "\\nThe BLEU score has some undesirable properties when used for single\nsentences, as it was designed to be a corpus measure. We therefore\nuse a slightly different score for our RL experiments which we call\nthe 'GLEU score'. For the GLEU score, we record all sub-sequences of\n1, 2, 3 or 4 tokens in output and target sequence (n-grams). We then\ncompute a recall, which is the ratio of the number of matching n-grams\nto the number of total n-grams in the target (ground truth) sequence,\nand a precision, which is the ratio of the number of matching n-grams\nto the number of total n-grams in the generated output sequence. Then\nGLEU score is simply the minimum of recall and precision. This GLEU\nscore's range is always between 0 (no matches) and 1 (all match) and\nit is symmetrical when switching output and target. According to\nour experiments, GLEU score correlates quite well with the BLEU\nmetric on a corpus level but does not have its drawbacks for our per\nsentence reward objective.\n"
a_ :List[str] = "\\nComputes corpus-level Google BLEU (GLEU) score of translated segments against one or more references.\nInstead of averaging the sentence level GLEU scores (i.e. macro-average precision), Wu et al. (2016) sum up the matching\ntokens and the max of hypothesis and reference tokens for each sentence, then compute using the aggregate values.\n\nArgs:\n predictions (list of str): list of translations to score.\n Each translation should be tokenized into a list of tokens.\n references (list of list of str): list of lists of references for each translation.\n Each reference should be tokenized into a list of tokens.\n min_len (int): The minimum order of n-gram this function should extract. Defaults to 1.\n max_len (int): The maximum order of n-gram this function should extract. Defaults to 4.\n\nReturns:\n 'google_bleu': google_bleu score\n\nExamples:\n Example 1:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses, references=list_of_references)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.44\n\n Example 2:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n >>> ref1b = ['It', 'is', 'a', 'guide', 'to', 'action', 'that',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'will', 'never',\n ... 'heed', 'the', 'cat', 'commands']\n >>> ref1c = ['It', 'is', 'the', 'practical', 'guide', 'for', 'the',\n ... 'rubber', 'duck', 'army', 'never', 'to', 'heed', 'the', 'directions',\n ... 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a, ref1b, ref1c], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses, references=list_of_references)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.61\n\n Example 3:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n >>> ref1b = ['It', 'is', 'a', 'guide', 'to', 'action', 'that',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'will', 'never',\n ... 'heed', 'the', 'cat', 'commands']\n >>> ref1c = ['It', 'is', 'the', 'practical', 'guide', 'for', 'the',\n ... 'rubber', 'duck', 'army', 'never', 'to', 'heed', 'the', 'directions',\n ... 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a, ref1b, ref1c], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses, references=list_of_references, min_len=2)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.53\n\n Example 4:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n >>> ref1b = ['It', 'is', 'a', 'guide', 'to', 'action', 'that',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'will', 'never',\n ... 'heed', 'the', 'cat', 'commands']\n >>> ref1c = ['It', 'is', 'the', 'practical', 'guide', 'for', 'the',\n ... 'rubber', 'duck', 'army', 'never', 'to', 'heed', 'the', 'directions',\n ... 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a, ref1b, ref1c], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses,references=list_of_references, min_len=2, max_len=6)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.4\n"
@datasets.utils.file_utils.add_start_docstrings(_DESCRIPTION , _KWARGS_DESCRIPTION )
class snake_case__ ( datasets.Metric ):
"""simple docstring"""
def lowercase_ ( self : str ) ->MetricInfo:
return datasets.MetricInfo(
description=_DESCRIPTION, citation=_CITATION, inputs_description=_KWARGS_DESCRIPTION, features=datasets.Features(
{
'predictions': datasets.Sequence(datasets.Value('string', id='token' ), id='sequence' ),
'references': datasets.Sequence(
datasets.Sequence(datasets.Value('string', id='token' ), id='sequence' ), id='references' ),
} ), )
def lowercase_ ( self : str, _snake_case : List[List[List[str]]], _snake_case : List[List[str]], _snake_case : int = 1, _snake_case : int = 4, ) ->Dict[str, float]:
return {
"google_bleu": gleu_score.corpus_gleu(
list_of_references=_snake_case, hypotheses=_snake_case, min_len=_snake_case, max_len=_snake_case )
}
| 277 | 0 |
from collections import OrderedDict
from typing import Mapping
from ...configuration_utils import PretrainedConfig
from ...onnx import OnnxConfig
from ...utils import logging
lowercase = logging.get_logger(__name__)
lowercase = {
"facebook/data2vec-text-base": "https://huggingface.co/data2vec/resolve/main/config.json",
}
class UpperCamelCase_ ( lowerCAmelCase_ ):
'''simple docstring'''
lowerCAmelCase = '''data2vec-text'''
def __init__( self , a=3_05_22 , a=7_68 , a=12 , a=12 , a=30_72 , a="gelu" , a=0.1 , a=0.1 , a=5_12 , a=2 , a=0.02 , a=1E-12 , a=1 , a=0 , a=2 , a="absolute" , a=True , a=None , **a , ) -> int:
super().__init__(pad_token_id=_snake_case , bos_token_id=_snake_case , eos_token_id=_snake_case , **_snake_case )
snake_case_ = vocab_size
snake_case_ = hidden_size
snake_case_ = num_hidden_layers
snake_case_ = num_attention_heads
snake_case_ = hidden_act
snake_case_ = intermediate_size
snake_case_ = hidden_dropout_prob
snake_case_ = attention_probs_dropout_prob
snake_case_ = max_position_embeddings
snake_case_ = type_vocab_size
snake_case_ = initializer_range
snake_case_ = layer_norm_eps
snake_case_ = position_embedding_type
snake_case_ = use_cache
snake_case_ = classifier_dropout
class UpperCamelCase_ ( lowerCAmelCase_ ):
'''simple docstring'''
@property
def _UpperCamelCase ( self ) -> Mapping[str, Mapping[int, str]]:
if self.task == "multiple-choice":
snake_case_ = {0: 'batch', 1: 'choice', 2: 'sequence'}
else:
snake_case_ = {0: 'batch', 1: 'sequence'}
return OrderedDict(
[
('input_ids', dynamic_axis),
('attention_mask', dynamic_axis),
] )
| 178 |
from math import factorial
def lowercase_ (A : int , A : int , A : float ):
if successes > trials:
raise ValueError('successes must be lower or equal to trials' )
if trials < 0 or successes < 0:
raise ValueError('the function is defined for non-negative integers' )
if not isinstance(A , A ) or not isinstance(A , A ):
raise ValueError('the function is defined for non-negative integers' )
if not 0 < prob < 1:
raise ValueError('prob has to be in range of 1 - 0' )
snake_case__ : List[Any] = (prob**successes) * ((1 - prob) ** (trials - successes))
# Calculate the binomial coefficient: n! / k!(n-k)!
snake_case__ : List[str] = float(factorial(A ) )
coefficient /= factorial(A ) * factorial(trials - successes )
return probability * coefficient
if __name__ == "__main__":
from doctest import testmod
testmod()
print("Probability of 2 successes out of 4 trails")
print("with probability of 0.75 is:", end=" ")
print(binomial_distribution(2, 4, 0.75))
| 277 | 0 |
"""simple docstring"""
def lowercase (SCREAMING_SNAKE_CASE_ : int = 2_00_00_00 ) -> List[Any]:
SCREAMING_SNAKE_CASE = [0 for i in range(n + 1 )]
SCREAMING_SNAKE_CASE = 1
SCREAMING_SNAKE_CASE = 1
for i in range(2 , int(n**0.5 ) + 1 ):
if primality_list[i] == 0:
for j in range(i * i , n + 1 , SCREAMING_SNAKE_CASE_ ):
SCREAMING_SNAKE_CASE = 1
SCREAMING_SNAKE_CASE = 0
for i in range(SCREAMING_SNAKE_CASE_ ):
if primality_list[i] == 0:
sum_of_primes += i
return sum_of_primes
if __name__ == "__main__":
print(f'''{solution() = }''')
| 113 |
from collections import UserDict
from typing import Union
import numpy as np
import requests
from ..utils import (
add_end_docstrings,
logging,
)
from .audio_classification import ffmpeg_read
from .base import PIPELINE_INIT_ARGS, Pipeline
a_ :List[Any] = logging.get_logger(__name__)
@add_end_docstrings(lowerCAmelCase_ )
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
def __init__( self : Optional[Any], **_snake_case : str ) ->Dict:
super().__init__(**_snake_case )
if self.framework != "pt":
raise ValueError(F'''The {self.__class__} is only available in PyTorch.''' )
# No specific FOR_XXX available yet
def __call__( self : Union[str, Any], _snake_case : Union[np.ndarray, bytes, str], **_snake_case : Tuple ) ->Dict:
return super().__call__(_snake_case, **_snake_case )
def lowercase_ ( self : Tuple, **_snake_case : Any ) ->Union[str, Any]:
snake_case__ : str = {}
if "candidate_labels" in kwargs:
snake_case__ : str = kwargs['candidate_labels']
if "hypothesis_template" in kwargs:
snake_case__ : str = kwargs['hypothesis_template']
return preprocess_params, {}, {}
def lowercase_ ( self : Dict, _snake_case : str, _snake_case : Optional[int]=None, _snake_case : List[str]="This is a sound of {}." ) ->int:
if isinstance(_snake_case, _snake_case ):
if audio.startswith('http://' ) or audio.startswith('https://' ):
# We need to actually check for a real protocol, otherwise it's impossible to use a local file
# like http_huggingface_co.png
snake_case__ : List[Any] = requests.get(_snake_case ).content
else:
with open(_snake_case, 'rb' ) as f:
snake_case__ : Union[str, Any] = f.read()
if isinstance(_snake_case, _snake_case ):
snake_case__ : List[Any] = ffmpeg_read(_snake_case, self.feature_extractor.sampling_rate )
if not isinstance(_snake_case, np.ndarray ):
raise ValueError('We expect a numpy ndarray as input' )
if len(audio.shape ) != 1:
raise ValueError('We expect a single channel audio input for ZeroShotAudioClassificationPipeline' )
snake_case__ : Tuple = self.feature_extractor(
[audio], sampling_rate=self.feature_extractor.sampling_rate, return_tensors='pt' )
snake_case__ : int = candidate_labels
snake_case__ : int = [hypothesis_template.format(_snake_case ) for x in candidate_labels]
snake_case__ : Optional[int] = self.tokenizer(_snake_case, return_tensors=self.framework, padding=_snake_case )
snake_case__ : List[Any] = [text_inputs]
return inputs
def lowercase_ ( self : Optional[int], _snake_case : Optional[Any] ) ->int:
snake_case__ : Optional[int] = model_inputs.pop('candidate_labels' )
snake_case__ : str = model_inputs.pop('text_inputs' )
if isinstance(text_inputs[0], _snake_case ):
snake_case__ : Optional[Any] = text_inputs[0]
else:
# Batching case.
snake_case__ : int = text_inputs[0][0]
snake_case__ : Any = self.model(**_snake_case, **_snake_case )
snake_case__ : List[Any] = {
'candidate_labels': candidate_labels,
'logits': outputs.logits_per_audio,
}
return model_outputs
def lowercase_ ( self : Union[str, Any], _snake_case : str ) ->List[str]:
snake_case__ : int = model_outputs.pop('candidate_labels' )
snake_case__ : List[Any] = model_outputs['logits'][0]
if self.framework == "pt":
snake_case__ : Tuple = logits.softmax(dim=0 )
snake_case__ : Union[str, Any] = probs.tolist()
else:
raise ValueError('`tf` framework not supported.' )
snake_case__ : Union[str, Any] = [
{'score': score, 'label': candidate_label}
for score, candidate_label in sorted(zip(_snake_case, _snake_case ), key=lambda _snake_case : -x[0] )
]
return result
| 277 | 0 |
'''simple docstring'''
import cva
import numpy as np
class _lowercase :
'''simple docstring'''
def __init__( self : Tuple , SCREAMING_SNAKE_CASE__ : float , SCREAMING_SNAKE_CASE__ : int ) -> List[Any]:
if k in (0.0_4, 0.0_6):
__lowerCAmelCase = k
__lowerCAmelCase = window_size
else:
raise ValueError("""invalid k value""" )
def __str__( self : Optional[int] ) -> str:
return str(self.k )
def a ( self : Optional[int] , SCREAMING_SNAKE_CASE__ : str ) -> tuple[cva.Mat, list[list[int]]]:
__lowerCAmelCase = cva.imread(_snake_case , 0 )
__lowerCAmelCase = img.shape
__lowerCAmelCase = []
__lowerCAmelCase = img.copy()
__lowerCAmelCase = cva.cvtColor(_snake_case , cva.COLOR_GRAY2RGB )
__lowerCAmelCase = np.gradient(_snake_case )
__lowerCAmelCase = dx**2
__lowerCAmelCase = dy**2
__lowerCAmelCase = dx * dy
__lowerCAmelCase = 0.0_4
__lowerCAmelCase = self.window_size // 2
for y in range(_snake_case , h - offset ):
for x in range(_snake_case , w - offset ):
__lowerCAmelCase = ixx[
y - offset : y + offset + 1, x - offset : x + offset + 1
].sum()
__lowerCAmelCase = iyy[
y - offset : y + offset + 1, x - offset : x + offset + 1
].sum()
__lowerCAmelCase = ixy[
y - offset : y + offset + 1, x - offset : x + offset + 1
].sum()
__lowerCAmelCase = (wxx * wyy) - (wxy**2)
__lowerCAmelCase = wxx + wyy
__lowerCAmelCase = det - k * (trace**2)
# Can change the value
if r > 0.5:
corner_list.append([x, y, r] )
color_img.itemset((y, x, 0) , 0 )
color_img.itemset((y, x, 1) , 0 )
color_img.itemset((y, x, 2) , 2_55 )
return color_img, corner_list
if __name__ == "__main__":
_A : Dict = HarrisCorner(0.0_4, 3)
_A : List[Any] = edge_detect.detect('''path_to_image''')
cva.imwrite('''detect.png''', color_img)
| 229 |
import collections
import inspect
import unittest
from transformers import SwinvaConfig
from transformers.testing_utils import require_torch, require_vision, slow, torch_device
from transformers.utils import cached_property, is_torch_available, is_vision_available
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, _config_zero_init, floats_tensor, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from torch import nn
from transformers import SwinvaForImageClassification, SwinvaForMaskedImageModeling, SwinvaModel
from transformers.models.swinva.modeling_swinva import SWINV2_PRETRAINED_MODEL_ARCHIVE_LIST
if is_vision_available():
from PIL import Image
from transformers import AutoImageProcessor
class snake_case__ :
"""simple docstring"""
def __init__( self : Tuple, _snake_case : Any, _snake_case : int=1_3, _snake_case : Optional[int]=3_2, _snake_case : Tuple=2, _snake_case : Any=3, _snake_case : Tuple=1_6, _snake_case : Tuple=[1, 2, 1], _snake_case : Dict=[2, 2, 4], _snake_case : str=2, _snake_case : Union[str, Any]=2.0, _snake_case : Dict=True, _snake_case : Dict=0.0, _snake_case : str=0.0, _snake_case : str=0.1, _snake_case : List[str]="gelu", _snake_case : int=False, _snake_case : Optional[Any]=True, _snake_case : List[Any]=0.0_2, _snake_case : Union[str, Any]=1e-5, _snake_case : Union[str, Any]=True, _snake_case : List[Any]=None, _snake_case : Any=True, _snake_case : List[Any]=1_0, _snake_case : str=8, ) ->Union[str, Any]:
snake_case__ : Any = parent
snake_case__ : Tuple = batch_size
snake_case__ : Tuple = image_size
snake_case__ : Any = patch_size
snake_case__ : Optional[int] = num_channels
snake_case__ : Tuple = embed_dim
snake_case__ : Any = depths
snake_case__ : Any = num_heads
snake_case__ : List[str] = window_size
snake_case__ : Dict = mlp_ratio
snake_case__ : Optional[int] = qkv_bias
snake_case__ : Optional[Any] = hidden_dropout_prob
snake_case__ : List[str] = attention_probs_dropout_prob
snake_case__ : Union[str, Any] = drop_path_rate
snake_case__ : str = hidden_act
snake_case__ : Union[str, Any] = use_absolute_embeddings
snake_case__ : Union[str, Any] = patch_norm
snake_case__ : Any = layer_norm_eps
snake_case__ : Tuple = initializer_range
snake_case__ : Dict = is_training
snake_case__ : Any = scope
snake_case__ : Optional[Any] = use_labels
snake_case__ : str = type_sequence_label_size
snake_case__ : List[Any] = encoder_stride
def lowercase_ ( self : Tuple ) ->str:
snake_case__ : Tuple = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size] )
snake_case__ : List[Any] = None
if self.use_labels:
snake_case__ : Optional[Any] = ids_tensor([self.batch_size], self.type_sequence_label_size )
snake_case__ : Any = self.get_config()
return config, pixel_values, labels
def lowercase_ ( self : Optional[int] ) ->Optional[int]:
return SwinvaConfig(
image_size=self.image_size, patch_size=self.patch_size, num_channels=self.num_channels, embed_dim=self.embed_dim, depths=self.depths, num_heads=self.num_heads, window_size=self.window_size, mlp_ratio=self.mlp_ratio, qkv_bias=self.qkv_bias, hidden_dropout_prob=self.hidden_dropout_prob, attention_probs_dropout_prob=self.attention_probs_dropout_prob, drop_path_rate=self.drop_path_rate, hidden_act=self.hidden_act, use_absolute_embeddings=self.use_absolute_embeddings, path_norm=self.patch_norm, layer_norm_eps=self.layer_norm_eps, initializer_range=self.initializer_range, encoder_stride=self.encoder_stride, )
def lowercase_ ( self : Optional[int], _snake_case : str, _snake_case : List[str], _snake_case : int ) ->Dict:
snake_case__ : List[Any] = SwinvaModel(config=_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Optional[int] = model(_snake_case )
snake_case__ : List[Any] = ((config.image_size // config.patch_size) ** 2) // (4 ** (len(config.depths ) - 1))
snake_case__ : List[Any] = int(config.embed_dim * 2 ** (len(config.depths ) - 1) )
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, expected_seq_len, expected_dim) )
def lowercase_ ( self : Optional[Any], _snake_case : Any, _snake_case : List[str], _snake_case : Dict ) ->List[Any]:
snake_case__ : List[str] = SwinvaForMaskedImageModeling(config=_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Union[str, Any] = model(_snake_case )
self.parent.assertEqual(
result.logits.shape, (self.batch_size, self.num_channels, self.image_size, self.image_size) )
# test greyscale images
snake_case__ : Optional[Any] = 1
snake_case__ : Optional[int] = SwinvaForMaskedImageModeling(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Tuple = floats_tensor([self.batch_size, 1, self.image_size, self.image_size] )
snake_case__ : Any = model(_snake_case )
self.parent.assertEqual(result.logits.shape, (self.batch_size, 1, self.image_size, self.image_size) )
def lowercase_ ( self : List[str], _snake_case : int, _snake_case : List[Any], _snake_case : Optional[int] ) ->Any:
snake_case__ : Tuple = self.type_sequence_label_size
snake_case__ : int = SwinvaForImageClassification(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Tuple = model(_snake_case, labels=_snake_case )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.type_sequence_label_size) )
def lowercase_ ( self : Any ) ->Dict:
snake_case__ : str = self.prepare_config_and_inputs()
snake_case__ , snake_case__ , snake_case__ : List[str] = config_and_inputs
snake_case__ : Union[str, Any] = {'pixel_values': pixel_values}
return config, inputs_dict
@require_torch
class snake_case__ ( lowerCAmelCase_ , lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = (
(SwinvaModel, SwinvaForImageClassification, SwinvaForMaskedImageModeling) if is_torch_available() else ()
)
_SCREAMING_SNAKE_CASE = (
{"""feature-extraction""": SwinvaModel, """image-classification""": SwinvaForImageClassification}
if is_torch_available()
else {}
)
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
def lowercase_ ( self : Union[str, Any] ) ->Dict:
snake_case__ : Optional[int] = SwinvaModelTester(self )
snake_case__ : int = ConfigTester(self, config_class=_snake_case, embed_dim=3_7 )
def lowercase_ ( self : Tuple ) ->int:
self.config_tester.create_and_test_config_to_json_string()
self.config_tester.create_and_test_config_to_json_file()
self.config_tester.create_and_test_config_from_and_save_pretrained()
self.config_tester.create_and_test_config_with_num_labels()
self.config_tester.check_config_can_be_init_without_params()
self.config_tester.check_config_arguments_init()
def lowercase_ ( self : Any ) ->str:
snake_case__ : int = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*_snake_case )
@unittest.skip(reason='Got `CUDA error: misaligned address` with PyTorch 2.0.0.' )
def lowercase_ ( self : Any ) ->Union[str, Any]:
pass
@unittest.skip(reason='Swinv2 does not use inputs_embeds' )
def lowercase_ ( self : str ) ->Union[str, Any]:
pass
def lowercase_ ( self : Optional[Any] ) ->Union[str, Any]:
snake_case__ , snake_case__ : Optional[int] = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
snake_case__ : Union[str, Any] = model_class(_snake_case )
self.assertIsInstance(model.get_input_embeddings(), (nn.Module) )
snake_case__ : Union[str, Any] = model.get_output_embeddings()
self.assertTrue(x is None or isinstance(_snake_case, nn.Linear ) )
def lowercase_ ( self : List[str] ) ->Optional[int]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
snake_case__ : Any = model_class(_snake_case )
snake_case__ : Dict = inspect.signature(model.forward )
# signature.parameters is an OrderedDict => so arg_names order is deterministic
snake_case__ : Optional[Any] = [*signature.parameters.keys()]
snake_case__ : List[Any] = ['pixel_values']
self.assertListEqual(arg_names[:1], _snake_case )
def lowercase_ ( self : str ) ->Union[str, Any]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : int = True
for model_class in self.all_model_classes:
snake_case__ : str = True
snake_case__ : Union[str, Any] = False
snake_case__ : Tuple = True
snake_case__ : int = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : Optional[int] = model(**self._prepare_for_class(_snake_case, _snake_case ) )
snake_case__ : List[str] = outputs.attentions
snake_case__ : List[Any] = len(self.model_tester.depths )
self.assertEqual(len(_snake_case ), _snake_case )
# check that output_attentions also work using config
del inputs_dict["output_attentions"]
snake_case__ : str = True
snake_case__ : Tuple = config.window_size**2
snake_case__ : Optional[int] = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : str = model(**self._prepare_for_class(_snake_case, _snake_case ) )
snake_case__ : Tuple = outputs.attentions
self.assertEqual(len(_snake_case ), _snake_case )
self.assertListEqual(
list(attentions[0].shape[-3:] ), [self.model_tester.num_heads[0], window_size_squared, window_size_squared], )
snake_case__ : Optional[Any] = len(_snake_case )
# Check attention is always last and order is fine
snake_case__ : Optional[int] = True
snake_case__ : Dict = True
snake_case__ : List[Any] = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : Optional[int] = model(**self._prepare_for_class(_snake_case, _snake_case ) )
if hasattr(self.model_tester, 'num_hidden_states_types' ):
snake_case__ : str = self.model_tester.num_hidden_states_types
else:
# also another +1 for reshaped_hidden_states
snake_case__ : Dict = 2
self.assertEqual(out_len + added_hidden_states, len(_snake_case ) )
snake_case__ : Any = outputs.attentions
self.assertEqual(len(_snake_case ), _snake_case )
self.assertListEqual(
list(self_attentions[0].shape[-3:] ), [self.model_tester.num_heads[0], window_size_squared, window_size_squared], )
def lowercase_ ( self : Dict, _snake_case : Tuple, _snake_case : Any, _snake_case : int, _snake_case : Optional[int] ) ->str:
snake_case__ : Dict = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : List[Any] = model(**self._prepare_for_class(_snake_case, _snake_case ) )
snake_case__ : Dict = outputs.hidden_states
snake_case__ : int = getattr(
self.model_tester, 'expected_num_hidden_layers', len(self.model_tester.depths ) + 1 )
self.assertEqual(len(_snake_case ), _snake_case )
# Swinv2 has a different seq_length
snake_case__ : int = (
config.patch_size
if isinstance(config.patch_size, collections.abc.Iterable )
else (config.patch_size, config.patch_size)
)
snake_case__ : Optional[Any] = (image_size[1] // patch_size[1]) * (image_size[0] // patch_size[0])
self.assertListEqual(
list(hidden_states[0].shape[-2:] ), [num_patches, self.model_tester.embed_dim], )
snake_case__ : Union[str, Any] = outputs.reshaped_hidden_states
self.assertEqual(len(_snake_case ), _snake_case )
snake_case__ , snake_case__ , snake_case__ , snake_case__ : str = reshaped_hidden_states[0].shape
snake_case__ : Any = (
reshaped_hidden_states[0].view(_snake_case, _snake_case, height * width ).permute(0, 2, 1 )
)
self.assertListEqual(
list(reshaped_hidden_states.shape[-2:] ), [num_patches, self.model_tester.embed_dim], )
def lowercase_ ( self : str ) ->List[Any]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : List[Any] = (
self.model_tester.image_size
if isinstance(self.model_tester.image_size, collections.abc.Iterable )
else (self.model_tester.image_size, self.model_tester.image_size)
)
for model_class in self.all_model_classes:
snake_case__ : Optional[int] = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, _snake_case )
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
snake_case__ : Dict = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, _snake_case )
def lowercase_ ( self : List[str] ) ->str:
snake_case__ , snake_case__ : Optional[Any] = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : List[str] = 3
snake_case__ : Union[str, Any] = (
self.model_tester.image_size
if isinstance(self.model_tester.image_size, collections.abc.Iterable )
else (self.model_tester.image_size, self.model_tester.image_size)
)
snake_case__ : str = (
config.patch_size
if isinstance(config.patch_size, collections.abc.Iterable )
else (config.patch_size, config.patch_size)
)
snake_case__ : Tuple = image_size[0] + patch_size[0] - (image_size[0] % patch_size[0])
snake_case__ : Optional[Any] = image_size[1] + patch_size[1] - (image_size[1] % patch_size[1])
for model_class in self.all_model_classes:
snake_case__ : int = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, (padded_height, padded_width) )
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
snake_case__ : List[str] = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, (padded_height, padded_width) )
def lowercase_ ( self : List[str] ) ->Optional[int]:
snake_case__ : Any = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_masked_image_modeling(*_snake_case )
def lowercase_ ( self : List[Any] ) ->str:
snake_case__ : Any = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_image_classification(*_snake_case )
@slow
def lowercase_ ( self : str ) ->Union[str, Any]:
for model_name in SWINV2_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
snake_case__ : Dict = SwinvaModel.from_pretrained(_snake_case )
self.assertIsNotNone(_snake_case )
def lowercase_ ( self : Optional[int] ) ->List[str]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : List[Any] = _config_zero_init(_snake_case )
for model_class in self.all_model_classes:
snake_case__ : List[str] = model_class(config=_snake_case )
for name, param in model.named_parameters():
if "embeddings" not in name and "logit_scale" not in name and param.requires_grad:
self.assertIn(
((param.data.mean() * 1e9).round() / 1e9).item(), [0.0, 1.0], msg=F'''Parameter {name} of model {model_class} seems not properly initialized''', )
@require_vision
@require_torch
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
@cached_property
def lowercase_ ( self : Union[str, Any] ) ->List[str]:
return (
AutoImageProcessor.from_pretrained('microsoft/swinv2-tiny-patch4-window8-256' )
if is_vision_available()
else None
)
@slow
def lowercase_ ( self : int ) ->List[Any]:
snake_case__ : Any = SwinvaForImageClassification.from_pretrained('microsoft/swinv2-tiny-patch4-window8-256' ).to(
_snake_case )
snake_case__ : int = self.default_image_processor
snake_case__ : Union[str, Any] = Image.open('./tests/fixtures/tests_samples/COCO/000000039769.png' )
snake_case__ : Optional[Any] = image_processor(images=_snake_case, return_tensors='pt' ).to(_snake_case )
# forward pass
with torch.no_grad():
snake_case__ : List[str] = model(**_snake_case )
# verify the logits
snake_case__ : int = torch.Size((1, 1_0_0_0) )
self.assertEqual(outputs.logits.shape, _snake_case )
snake_case__ : Optional[int] = torch.tensor([-0.3_9_4_7, -0.4_3_0_6, 0.0_0_2_6] ).to(_snake_case )
self.assertTrue(torch.allclose(outputs.logits[0, :3], _snake_case, atol=1e-4 ) )
| 277 | 0 |
"""simple docstring"""
import argparse
import json
from pathlib import Path
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import (
BertTokenizer,
ViltConfig,
ViltForImageAndTextRetrieval,
ViltForImagesAndTextClassification,
ViltForMaskedLM,
ViltForQuestionAnswering,
ViltImageProcessor,
ViltProcessor,
)
from transformers.utils import logging
logging.set_verbosity_info()
UpperCAmelCase = logging.get_logger(__name__)
def lowercase ( a__ : Dict , a__ : int=False , a__ : int=False , a__ : int=False ) -> Tuple:
_UpperCamelCase = []
for i in range(config.num_hidden_layers ):
# encoder layers: output projection, 2 feedforward neural networks and 2 layernorms
rename_keys.append((F'''transformer.blocks.{i}.norm1.weight''', F'''vilt.encoder.layer.{i}.layernorm_before.weight''') )
rename_keys.append((F'''transformer.blocks.{i}.norm1.bias''', F'''vilt.encoder.layer.{i}.layernorm_before.bias''') )
rename_keys.append(
(F'''transformer.blocks.{i}.attn.proj.weight''', F'''vilt.encoder.layer.{i}.attention.output.dense.weight''') )
rename_keys.append(
(F'''transformer.blocks.{i}.attn.proj.bias''', F'''vilt.encoder.layer.{i}.attention.output.dense.bias''') )
rename_keys.append((F'''transformer.blocks.{i}.norm2.weight''', F'''vilt.encoder.layer.{i}.layernorm_after.weight''') )
rename_keys.append((F'''transformer.blocks.{i}.norm2.bias''', F'''vilt.encoder.layer.{i}.layernorm_after.bias''') )
rename_keys.append(
(F'''transformer.blocks.{i}.mlp.fc1.weight''', F'''vilt.encoder.layer.{i}.intermediate.dense.weight''') )
rename_keys.append((F'''transformer.blocks.{i}.mlp.fc1.bias''', F'''vilt.encoder.layer.{i}.intermediate.dense.bias''') )
rename_keys.append((F'''transformer.blocks.{i}.mlp.fc2.weight''', F'''vilt.encoder.layer.{i}.output.dense.weight''') )
rename_keys.append((F'''transformer.blocks.{i}.mlp.fc2.bias''', F'''vilt.encoder.layer.{i}.output.dense.bias''') )
# embeddings
rename_keys.extend(
[
# text embeddings
('''text_embeddings.word_embeddings.weight''', '''vilt.embeddings.text_embeddings.word_embeddings.weight'''),
(
'''text_embeddings.position_embeddings.weight''',
'''vilt.embeddings.text_embeddings.position_embeddings.weight''',
),
('''text_embeddings.position_ids''', '''vilt.embeddings.text_embeddings.position_ids'''),
(
'''text_embeddings.token_type_embeddings.weight''',
'''vilt.embeddings.text_embeddings.token_type_embeddings.weight''',
),
('''text_embeddings.LayerNorm.weight''', '''vilt.embeddings.text_embeddings.LayerNorm.weight'''),
('''text_embeddings.LayerNorm.bias''', '''vilt.embeddings.text_embeddings.LayerNorm.bias'''),
# patch embeddings
('''transformer.cls_token''', '''vilt.embeddings.cls_token'''),
('''transformer.patch_embed.proj.weight''', '''vilt.embeddings.patch_embeddings.projection.weight'''),
('''transformer.patch_embed.proj.bias''', '''vilt.embeddings.patch_embeddings.projection.bias'''),
('''transformer.pos_embed''', '''vilt.embeddings.position_embeddings'''),
# token type embeddings
('''token_type_embeddings.weight''', '''vilt.embeddings.token_type_embeddings.weight'''),
] )
# final layernorm + pooler
rename_keys.extend(
[
('''transformer.norm.weight''', '''vilt.layernorm.weight'''),
('''transformer.norm.bias''', '''vilt.layernorm.bias'''),
('''pooler.dense.weight''', '''vilt.pooler.dense.weight'''),
('''pooler.dense.bias''', '''vilt.pooler.dense.bias'''),
] )
# classifier head(s)
if vqa_model:
# classification head
rename_keys.extend(
[
('''vqa_classifier.0.weight''', '''classifier.0.weight'''),
('''vqa_classifier.0.bias''', '''classifier.0.bias'''),
('''vqa_classifier.1.weight''', '''classifier.1.weight'''),
('''vqa_classifier.1.bias''', '''classifier.1.bias'''),
('''vqa_classifier.3.weight''', '''classifier.3.weight'''),
('''vqa_classifier.3.bias''', '''classifier.3.bias'''),
] )
elif nlvr_model:
# classification head
rename_keys.extend(
[
('''nlvr2_classifier.0.weight''', '''classifier.0.weight'''),
('''nlvr2_classifier.0.bias''', '''classifier.0.bias'''),
('''nlvr2_classifier.1.weight''', '''classifier.1.weight'''),
('''nlvr2_classifier.1.bias''', '''classifier.1.bias'''),
('''nlvr2_classifier.3.weight''', '''classifier.3.weight'''),
('''nlvr2_classifier.3.bias''', '''classifier.3.bias'''),
] )
else:
pass
return rename_keys
def lowercase ( a__ : List[str] , a__ : List[Any] ) -> List[str]:
for i in range(config.num_hidden_layers ):
_UpperCamelCase = 'vilt.'
# read in weights + bias of input projection layer (in timm, this is a single matrix + bias)
_UpperCamelCase = state_dict.pop(F'''transformer.blocks.{i}.attn.qkv.weight''' )
_UpperCamelCase = state_dict.pop(F'''transformer.blocks.{i}.attn.qkv.bias''' )
# next, add query, keys and values (in that order) to the state dict
_UpperCamelCase = in_proj_weight[
: config.hidden_size, :
]
_UpperCamelCase = in_proj_bias[: config.hidden_size]
_UpperCamelCase = in_proj_weight[
config.hidden_size : config.hidden_size * 2, :
]
_UpperCamelCase = in_proj_bias[
config.hidden_size : config.hidden_size * 2
]
_UpperCamelCase = in_proj_weight[
-config.hidden_size :, :
]
_UpperCamelCase = in_proj_bias[-config.hidden_size :]
def lowercase ( a__ : int ) -> Any:
_UpperCamelCase = ['head.weight', 'head.bias']
for k in ignore_keys:
state_dict.pop(a__ , a__ )
def lowercase ( a__ : int , a__ : List[Any] , a__ : List[str] ) -> Dict:
_UpperCamelCase = dct.pop(a__ )
_UpperCamelCase = val
@torch.no_grad()
def lowercase ( a__ : int , a__ : Union[str, Any] ) -> Any:
_UpperCamelCase = ViltConfig(image_size=384 , patch_size=32 , tie_word_embeddings=a__ )
_UpperCamelCase = False
_UpperCamelCase = False
_UpperCamelCase = False
_UpperCamelCase = False
if "vqa" in checkpoint_url:
_UpperCamelCase = True
_UpperCamelCase = 3129
_UpperCamelCase = 'huggingface/label-files'
_UpperCamelCase = 'vqa2-id2label.json'
_UpperCamelCase = json.load(open(hf_hub_download(a__ , a__ , repo_type='''dataset''' ) , '''r''' ) )
_UpperCamelCase = {int(a__ ): v for k, v in idalabel.items()}
_UpperCamelCase = idalabel
_UpperCamelCase = {v: k for k, v in idalabel.items()}
_UpperCamelCase = ViltForQuestionAnswering(a__ )
elif "nlvr" in checkpoint_url:
_UpperCamelCase = True
_UpperCamelCase = 2
_UpperCamelCase = {0: 'False', 1: 'True'}
_UpperCamelCase = {v: k for k, v in config.idalabel.items()}
_UpperCamelCase = 3
_UpperCamelCase = ViltForImagesAndTextClassification(a__ )
elif "irtr" in checkpoint_url:
_UpperCamelCase = True
_UpperCamelCase = ViltForImageAndTextRetrieval(a__ )
elif "mlm_itm" in checkpoint_url:
_UpperCamelCase = True
_UpperCamelCase = ViltForMaskedLM(a__ )
else:
raise ValueError('''Unknown model type''' )
# load state_dict of original model, remove and rename some keys
_UpperCamelCase = torch.hub.load_state_dict_from_url(a__ , map_location='''cpu''' )['state_dict']
_UpperCamelCase = create_rename_keys(a__ , a__ , a__ , a__ )
for src, dest in rename_keys:
rename_key(a__ , a__ , a__ )
read_in_q_k_v(a__ , a__ )
if mlm_model or irtr_model:
_UpperCamelCase = ['itm_score.fc.weight', 'itm_score.fc.bias']
for k in ignore_keys:
state_dict.pop(a__ , a__ )
# load state dict into HuggingFace model
model.eval()
if mlm_model:
_UpperCamelCase = model.load_state_dict(a__ , strict=a__ )
assert missing_keys == ["mlm_score.decoder.bias"]
else:
model.load_state_dict(a__ )
# Define processor
_UpperCamelCase = ViltImageProcessor(size=384 )
_UpperCamelCase = BertTokenizer.from_pretrained('''bert-base-uncased''' )
_UpperCamelCase = ViltProcessor(a__ , a__ )
# Forward pass on example inputs (image + text)
if nlvr_model:
_UpperCamelCase = Image.open(requests.get('''https://lil.nlp.cornell.edu/nlvr/exs/ex0_0.jpg''' , stream=a__ ).raw )
_UpperCamelCase = Image.open(requests.get('''https://lil.nlp.cornell.edu/nlvr/exs/ex0_0.jpg''' , stream=a__ ).raw )
_UpperCamelCase = (
'The left image contains twice the number of dogs as the right image, and at least two dogs in total are'
' standing.'
)
_UpperCamelCase = processor(a__ , a__ , return_tensors='''pt''' )
_UpperCamelCase = processor(a__ , a__ , return_tensors='''pt''' )
_UpperCamelCase = model(
input_ids=encoding_a.input_ids , pixel_values=encoding_a.pixel_values , pixel_values_a=encoding_a.pixel_values , )
else:
_UpperCamelCase = Image.open(requests.get('''http://images.cocodataset.org/val2017/000000039769.jpg''' , stream=a__ ).raw )
if mlm_model:
_UpperCamelCase = 'a bunch of [MASK] laying on a [MASK].'
else:
_UpperCamelCase = 'How many cats are there?'
_UpperCamelCase = processor(a__ , a__ , return_tensors='''pt''' )
_UpperCamelCase = model(**a__ )
# Verify outputs
if mlm_model:
_UpperCamelCase = torch.Size([1, 11, 30522] )
_UpperCamelCase = torch.tensor([-12.5061, -12.5123, -12.5174] )
assert outputs.logits.shape == expected_shape
assert torch.allclose(outputs.logits[0, 0, :3] , a__ , atol=1e-4 )
# verify masked token prediction equals "cats"
_UpperCamelCase = outputs.logits[0, 4, :].argmax(-1 ).item()
assert tokenizer.decode([predicted_id] ) == "cats"
elif vqa_model:
_UpperCamelCase = torch.Size([1, 3129] )
_UpperCamelCase = torch.tensor([-15.9495, -18.1472, -10.3041] )
assert torch.allclose(outputs.logits[0, :3] , a__ , atol=1e-4 )
assert outputs.logits.shape == expected_shape
assert torch.allclose(outputs.logits[0, 0, :3] , a__ , atol=1e-4 )
# verify vqa prediction equals "2"
_UpperCamelCase = outputs.logits.argmax(-1 ).item()
assert model.config.idalabel[predicted_idx] == "2"
elif nlvr_model:
_UpperCamelCase = torch.Size([1, 2] )
_UpperCamelCase = torch.tensor([-2.8721, 2.1291] )
assert torch.allclose(outputs.logits[0, :3] , a__ , atol=1e-4 )
assert outputs.logits.shape == expected_shape
Path(a__ ).mkdir(exist_ok=a__ )
print(F'''Saving model and processor to {pytorch_dump_folder_path}''' )
model.save_pretrained(a__ )
processor.save_pretrained(a__ )
if __name__ == "__main__":
UpperCAmelCase = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"""--checkpoint_url""",
default="""https://github.com/dandelin/ViLT/releases/download/200k/vilt_200k_mlm_itm.ckpt""",
type=str,
help="""URL of the checkpoint you'd like to convert.""",
)
parser.add_argument(
"""--pytorch_dump_folder_path""", default=None, type=str, help="""Path to the output PyTorch model directory."""
)
UpperCAmelCase = parser.parse_args()
convert_vilt_checkpoint(args.checkpoint_url, args.pytorch_dump_folder_path)
| 256 |
import unittest
import numpy as np
from transformers.testing_utils import require_torch, require_vision
from transformers.utils import is_torch_available, is_vision_available
from ...test_image_processing_common import ImageProcessingSavingTestMixin, prepare_image_inputs
if is_torch_available():
import torch
if is_vision_available():
from PIL import Image
from transformers import PoolFormerImageProcessor
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
def __init__( self : Optional[int], _snake_case : List[Any], _snake_case : str=7, _snake_case : Tuple=3, _snake_case : List[str]=3_0, _snake_case : Tuple=4_0_0, _snake_case : Any=True, _snake_case : List[Any]=None, _snake_case : int=0.9, _snake_case : Optional[Any]=None, _snake_case : str=True, _snake_case : Union[str, Any]=[0.5, 0.5, 0.5], _snake_case : Union[str, Any]=[0.5, 0.5, 0.5], ) ->List[Any]:
snake_case__ : int = size if size is not None else {'shortest_edge': 3_0}
snake_case__ : Tuple = crop_size if crop_size is not None else {'height': 3_0, 'width': 3_0}
snake_case__ : Union[str, Any] = parent
snake_case__ : Dict = batch_size
snake_case__ : int = num_channels
snake_case__ : Tuple = min_resolution
snake_case__ : Any = max_resolution
snake_case__ : List[Any] = do_resize_and_center_crop
snake_case__ : str = size
snake_case__ : str = crop_pct
snake_case__ : List[str] = crop_size
snake_case__ : Optional[int] = do_normalize
snake_case__ : Tuple = image_mean
snake_case__ : Tuple = image_std
def lowercase_ ( self : Optional[int] ) ->int:
return {
"size": self.size,
"do_resize_and_center_crop": self.do_resize_and_center_crop,
"crop_pct": self.crop_pct,
"crop_size": self.crop_size,
"do_normalize": self.do_normalize,
"image_mean": self.image_mean,
"image_std": self.image_std,
}
@require_torch
@require_vision
class snake_case__ ( lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = PoolFormerImageProcessor if is_vision_available() else None
def lowercase_ ( self : Union[str, Any] ) ->Dict:
snake_case__ : Union[str, Any] = PoolFormerImageProcessingTester(self )
@property
def lowercase_ ( self : int ) ->Dict:
return self.image_processor_tester.prepare_image_processor_dict()
def lowercase_ ( self : Union[str, Any] ) ->Optional[int]:
snake_case__ : List[str] = self.image_processing_class(**self.image_processor_dict )
self.assertTrue(hasattr(_snake_case, 'do_resize_and_center_crop' ) )
self.assertTrue(hasattr(_snake_case, 'size' ) )
self.assertTrue(hasattr(_snake_case, 'crop_pct' ) )
self.assertTrue(hasattr(_snake_case, 'do_normalize' ) )
self.assertTrue(hasattr(_snake_case, 'image_mean' ) )
self.assertTrue(hasattr(_snake_case, 'image_std' ) )
def lowercase_ ( self : List[str] ) ->List[str]:
snake_case__ : Union[str, Any] = self.image_processing_class.from_dict(self.image_processor_dict )
self.assertEqual(image_processor.size, {'shortest_edge': 3_0} )
self.assertEqual(image_processor.crop_size, {'height': 3_0, 'width': 3_0} )
snake_case__ : int = self.image_processing_class.from_dict(self.image_processor_dict, size=4_2, crop_size=8_4 )
self.assertEqual(image_processor.size, {'shortest_edge': 4_2} )
self.assertEqual(image_processor.crop_size, {'height': 8_4, 'width': 8_4} )
def lowercase_ ( self : List[Any] ) ->List[Any]:
pass
def lowercase_ ( self : List[str] ) ->str:
# Initialize image_processing
snake_case__ : Union[str, Any] = self.image_processing_class(**self.image_processor_dict )
# create random PIL images
snake_case__ : List[str] = prepare_image_inputs(self.image_processor_tester, equal_resolution=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case, Image.Image )
# Test not batched input
snake_case__ : Optional[int] = image_processing(image_inputs[0], return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
# Test batched
snake_case__ : str = image_processing(_snake_case, return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
def lowercase_ ( self : int ) ->List[Any]:
# Initialize image_processing
snake_case__ : Optional[int] = self.image_processing_class(**self.image_processor_dict )
# create random numpy tensors
snake_case__ : Dict = prepare_image_inputs(self.image_processor_tester, equal_resolution=_snake_case, numpify=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case, np.ndarray )
# Test not batched input
snake_case__ : Dict = image_processing(image_inputs[0], return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
# Test batched
snake_case__ : List[Any] = image_processing(_snake_case, return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
def lowercase_ ( self : List[str] ) ->List[str]:
# Initialize image_processing
snake_case__ : Tuple = self.image_processing_class(**self.image_processor_dict )
# create random PyTorch tensors
snake_case__ : List[str] = prepare_image_inputs(self.image_processor_tester, equal_resolution=_snake_case, torchify=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case, torch.Tensor )
# Test not batched input
snake_case__ : Tuple = image_processing(image_inputs[0], return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
# Test batched
snake_case__ : Optional[Any] = image_processing(_snake_case, return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
| 277 | 0 |
import itertools
import random
import unittest
import numpy as np
from transformers import BatchFeature, SpeechTaFeatureExtractor
from transformers.testing_utils import require_torch
from transformers.utils.import_utils import is_torch_available
from ...test_sequence_feature_extraction_common import SequenceFeatureExtractionTestMixin
if is_torch_available():
import torch
_a = random.Random()
def _a ( SCREAMING_SNAKE_CASE : int , SCREAMING_SNAKE_CASE : Union[str, Any]=1.0 , SCREAMING_SNAKE_CASE : List[str]=None , SCREAMING_SNAKE_CASE : Any=None ) -> Any:
"""simple docstring"""
if rng is None:
__lowerCAmelCase: List[str] = global_rng
__lowerCAmelCase: int = []
for batch_idx in range(shape[0] ):
values.append([] )
for _ in range(shape[1] ):
values[-1].append(rng.random() * scale )
return values
@require_torch
class A_ ( unittest.TestCase ):
def __init__( self : Optional[Any] , UpperCAmelCase : List[str] , UpperCAmelCase : Tuple=7 , UpperCAmelCase : Union[str, Any]=4_0_0 , UpperCAmelCase : Any=2_0_0_0 , UpperCAmelCase : Dict=1 , UpperCAmelCase : Optional[Any]=0.0 , UpperCAmelCase : List[Any]=1_6_0_0_0 , UpperCAmelCase : List[Any]=True , UpperCAmelCase : List[Any]=8_0 , UpperCAmelCase : Dict=1_6 , UpperCAmelCase : str=6_4 , UpperCAmelCase : Tuple="hann_window" , UpperCAmelCase : Union[str, Any]=8_0 , UpperCAmelCase : Optional[Any]=7_6_0_0 , UpperCAmelCase : str=1E-10 , UpperCAmelCase : Any=True , ) -> Union[str, Any]:
__lowerCAmelCase: Optional[int] = parent
__lowerCAmelCase: Optional[Any] = batch_size
__lowerCAmelCase: List[Any] = min_seq_length
__lowerCAmelCase: List[Any] = max_seq_length
__lowerCAmelCase: Any = (self.max_seq_length - self.min_seq_length) // (self.batch_size - 1)
__lowerCAmelCase: Tuple = feature_size
__lowerCAmelCase: List[Any] = padding_value
__lowerCAmelCase: Any = sampling_rate
__lowerCAmelCase: Dict = do_normalize
__lowerCAmelCase: Union[str, Any] = num_mel_bins
__lowerCAmelCase: Any = hop_length
__lowerCAmelCase: Any = win_length
__lowerCAmelCase: Any = win_function
__lowerCAmelCase: Optional[int] = fmin
__lowerCAmelCase: int = fmax
__lowerCAmelCase: Union[str, Any] = mel_floor
__lowerCAmelCase: Union[str, Any] = return_attention_mask
def UpperCAmelCase ( self : Optional[int] ) -> List[str]:
return {
"feature_size": self.feature_size,
"padding_value": self.padding_value,
"sampling_rate": self.sampling_rate,
"do_normalize": self.do_normalize,
"num_mel_bins": self.num_mel_bins,
"hop_length": self.hop_length,
"win_length": self.win_length,
"win_function": self.win_function,
"fmin": self.fmin,
"fmax": self.fmax,
"mel_floor": self.mel_floor,
"return_attention_mask": self.return_attention_mask,
}
def UpperCAmelCase ( self : Any , UpperCAmelCase : Optional[Any]=False , UpperCAmelCase : List[str]=False ) -> Union[str, Any]:
def _flatten(UpperCAmelCase : List[str] ):
return list(itertools.chain(*_snake_case ) )
if equal_length:
__lowerCAmelCase: Any = floats_list((self.batch_size, self.max_seq_length) )
else:
# make sure that inputs increase in size
__lowerCAmelCase: int = [
_flatten(floats_list((x, self.feature_size) ) )
for x in range(self.min_seq_length , self.max_seq_length , self.seq_length_diff )
]
if numpify:
__lowerCAmelCase: Any = [np.asarray(_snake_case ) for x in speech_inputs]
return speech_inputs
def UpperCAmelCase ( self : Union[str, Any] , UpperCAmelCase : str=False , UpperCAmelCase : Dict=False ) -> List[str]:
if equal_length:
__lowerCAmelCase: Optional[Any] = [floats_list((self.max_seq_length, self.num_mel_bins) ) for _ in range(self.batch_size )]
else:
# make sure that inputs increase in size
__lowerCAmelCase: List[str] = [
floats_list((x, self.num_mel_bins) )
for x in range(self.min_seq_length , self.max_seq_length , self.seq_length_diff )
]
if numpify:
__lowerCAmelCase: int = [np.asarray(_snake_case ) for x in speech_inputs]
return speech_inputs
@require_torch
class A_ ( lowerCAmelCase_ , unittest.TestCase ):
_lowercase : int = SpeechTaFeatureExtractor
def UpperCAmelCase ( self : int ) -> Union[str, Any]:
__lowerCAmelCase: List[str] = SpeechTaFeatureExtractionTester(self )
def UpperCAmelCase ( self : Any , UpperCAmelCase : Dict ) -> Any:
self.assertTrue(np.all(np.mean(_snake_case , axis=0 ) < 1E-3 ) )
self.assertTrue(np.all(np.abs(np.var(_snake_case , axis=0 ) - 1 ) < 1E-3 ) )
def UpperCAmelCase ( self : List[Any] ) -> Union[str, Any]:
# Tests that all call wrap to encode_plus and batch_encode_plus
__lowerCAmelCase: Optional[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
# create three inputs of length 800, 1000, and 1200
__lowerCAmelCase: int = [floats_list((1, x) )[0] for x in range(8_0_0 , 1_4_0_0 , 2_0_0 )]
__lowerCAmelCase: Tuple = [np.asarray(_snake_case ) for speech_input in speech_inputs]
# Test not batched input
__lowerCAmelCase: str = feat_extract(speech_inputs[0] , return_tensors='np' ).input_values
__lowerCAmelCase: List[str] = feat_extract(np_speech_inputs[0] , return_tensors='np' ).input_values
self.assertTrue(np.allclose(_snake_case , _snake_case , atol=1E-3 ) )
# Test batched
__lowerCAmelCase: Any = feat_extract(_snake_case , return_tensors='np' ).input_values
__lowerCAmelCase: Union[str, Any] = feat_extract(_snake_case , return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case , _snake_case ):
self.assertTrue(np.allclose(_snake_case , _snake_case , atol=1E-3 ) )
def UpperCAmelCase ( self : int ) -> Optional[int]:
__lowerCAmelCase: List[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
__lowerCAmelCase: Tuple = [floats_list((1, x) )[0] for x in range(8_0_0 , 1_4_0_0 , 2_0_0 )]
__lowerCAmelCase: int = ['longest', 'max_length', 'do_not_pad']
__lowerCAmelCase: List[str] = [None, 1_6_0_0, None]
for max_length, padding in zip(_snake_case , _snake_case ):
__lowerCAmelCase: Optional[int] = feat_extract(_snake_case , padding=_snake_case , max_length=_snake_case , return_tensors='np' )
__lowerCAmelCase: Optional[int] = processed.input_values
self._check_zero_mean_unit_variance(input_values[0][:8_0_0] )
self.assertTrue(input_values[0][8_0_0:].sum() < 1E-6 )
self._check_zero_mean_unit_variance(input_values[1][:1_0_0_0] )
self.assertTrue(input_values[0][1_0_0_0:].sum() < 1E-6 )
self._check_zero_mean_unit_variance(input_values[2][:1_2_0_0] )
def UpperCAmelCase ( self : Union[str, Any] ) -> Optional[Any]:
__lowerCAmelCase: Optional[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
__lowerCAmelCase: Tuple = range(8_0_0 , 1_4_0_0 , 2_0_0 )
__lowerCAmelCase: Optional[Any] = [floats_list((1, x) )[0] for x in lengths]
__lowerCAmelCase: Union[str, Any] = ['longest', 'max_length', 'do_not_pad']
__lowerCAmelCase: str = [None, 1_6_0_0, None]
for max_length, padding in zip(_snake_case , _snake_case ):
__lowerCAmelCase: List[str] = feat_extract(_snake_case , max_length=_snake_case , padding=_snake_case )
__lowerCAmelCase: Tuple = processed.input_values
self._check_zero_mean_unit_variance(input_values[0][:8_0_0] )
self._check_zero_mean_unit_variance(input_values[1][:1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2][:1_2_0_0] )
def UpperCAmelCase ( self : List[Any] ) -> Optional[Any]:
__lowerCAmelCase: Any = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
__lowerCAmelCase: str = [floats_list((1, x) )[0] for x in range(8_0_0 , 1_4_0_0 , 2_0_0 )]
__lowerCAmelCase: Optional[Any] = feat_extract(
_snake_case , truncation=_snake_case , max_length=1_0_0_0 , padding='max_length' , return_tensors='np' )
__lowerCAmelCase: int = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1] )
self._check_zero_mean_unit_variance(input_values[2] )
def UpperCAmelCase ( self : int ) -> Union[str, Any]:
__lowerCAmelCase: str = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
__lowerCAmelCase: Dict = [floats_list((1, x) )[0] for x in range(8_0_0 , 1_4_0_0 , 2_0_0 )]
__lowerCAmelCase: str = feat_extract(
_snake_case , truncation=_snake_case , max_length=1_0_0_0 , padding='longest' , return_tensors='np' )
__lowerCAmelCase: Dict = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1, :1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2] )
# make sure that if max_length < longest -> then pad to max_length
self.assertTrue(input_values.shape == (3, 1_0_0_0) )
__lowerCAmelCase: Tuple = [floats_list((1, x) )[0] for x in range(8_0_0 , 1_4_0_0 , 2_0_0 )]
__lowerCAmelCase: List[str] = feat_extract(
_snake_case , truncation=_snake_case , max_length=2_0_0_0 , padding='longest' , return_tensors='np' )
__lowerCAmelCase: Optional[Any] = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1, :1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2] )
# make sure that if max_length > longest -> then pad to longest
self.assertTrue(input_values.shape == (3, 1_2_0_0) )
def UpperCAmelCase ( self : List[str] ) -> Dict:
__lowerCAmelCase: Dict = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
__lowerCAmelCase: List[Any] = np.random.rand(1_0_0 ).astype(np.floataa )
__lowerCAmelCase: int = np_speech_inputs.tolist()
for inputs in [py_speech_inputs, np_speech_inputs]:
__lowerCAmelCase: int = feature_extractor.pad([{'input_values': inputs}] , return_tensors='np' )
self.assertTrue(np_processed.input_values.dtype == np.floataa )
__lowerCAmelCase: Optional[int] = feature_extractor.pad([{'input_values': inputs}] , return_tensors='pt' )
self.assertTrue(pt_processed.input_values.dtype == torch.floataa )
def UpperCAmelCase ( self : Optional[int] ) -> Optional[Any]:
# Tests that all call wrap to encode_plus and batch_encode_plus
__lowerCAmelCase: str = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
# create three inputs of length 800, 1000, and 1200
__lowerCAmelCase: List[Any] = [floats_list((1, x) )[0] for x in range(8_0_0 , 1_4_0_0 , 2_0_0 )]
__lowerCAmelCase: Dict = [np.asarray(_snake_case ) for speech_input in speech_inputs]
# Test feature size
__lowerCAmelCase: Optional[int] = feature_extractor(audio_target=_snake_case , padding=_snake_case , return_tensors='np' ).input_values
self.assertTrue(input_values.ndim == 3 )
self.assertTrue(input_values.shape[-1] == feature_extractor.num_mel_bins )
# Test not batched input
__lowerCAmelCase: Dict = feature_extractor(speech_inputs[0] , return_tensors='np' ).input_values
__lowerCAmelCase: Any = feature_extractor(np_speech_inputs[0] , return_tensors='np' ).input_values
self.assertTrue(np.allclose(_snake_case , _snake_case , atol=1E-3 ) )
# Test batched
__lowerCAmelCase: Dict = feature_extractor(_snake_case , return_tensors='np' ).input_values
__lowerCAmelCase: Dict = feature_extractor(_snake_case , return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case , _snake_case ):
self.assertTrue(np.allclose(_snake_case , _snake_case , atol=1E-3 ) )
# Test 2-D numpy arrays are batched.
__lowerCAmelCase: Optional[Any] = [floats_list((1, x) )[0] for x in (8_0_0, 8_0_0, 8_0_0)]
__lowerCAmelCase: int = np.asarray(_snake_case )
__lowerCAmelCase: Union[str, Any] = feature_extractor(_snake_case , return_tensors='np' ).input_values
__lowerCAmelCase: Union[str, Any] = feature_extractor(_snake_case , return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case , _snake_case ):
self.assertTrue(np.allclose(_snake_case , _snake_case , atol=1E-3 ) )
def UpperCAmelCase ( self : Union[str, Any] ) -> str:
__lowerCAmelCase: int = self.feat_extract_tester.prepare_inputs_for_target()
__lowerCAmelCase: List[str] = self.feature_extraction_class(**self.feat_extract_dict )
__lowerCAmelCase: Optional[Any] = feat_extract.model_input_names[0]
__lowerCAmelCase: Tuple = BatchFeature({input_name: speech_inputs} )
self.assertTrue(all(len(_snake_case ) == len(_snake_case ) for x, y in zip(_snake_case , processed_features[input_name] ) ) )
__lowerCAmelCase: int = self.feat_extract_tester.prepare_inputs_for_target(equal_length=_snake_case )
__lowerCAmelCase: Union[str, Any] = BatchFeature({input_name: speech_inputs} , tensor_type='np' )
__lowerCAmelCase: Dict = processed_features[input_name]
if len(batch_features_input.shape ) < 3:
__lowerCAmelCase: List[str] = batch_features_input[:, :, None]
self.assertTrue(
batch_features_input.shape
== (self.feat_extract_tester.batch_size, len(speech_inputs[0] ), self.feat_extract_tester.num_mel_bins) )
@require_torch
def UpperCAmelCase ( self : List[str] ) -> Any:
__lowerCAmelCase: int = self.feat_extract_tester.prepare_inputs_for_target(equal_length=_snake_case )
__lowerCAmelCase: Optional[Any] = self.feature_extraction_class(**self.feat_extract_dict )
__lowerCAmelCase: Tuple = feat_extract.model_input_names[0]
__lowerCAmelCase: List[Any] = BatchFeature({input_name: speech_inputs} , tensor_type='pt' )
__lowerCAmelCase: Tuple = processed_features[input_name]
if len(batch_features_input.shape ) < 3:
__lowerCAmelCase: Any = batch_features_input[:, :, None]
self.assertTrue(
batch_features_input.shape
== (self.feat_extract_tester.batch_size, len(speech_inputs[0] ), self.feat_extract_tester.num_mel_bins) )
@require_torch
def UpperCAmelCase ( self : Optional[int] ) -> Tuple:
__lowerCAmelCase: Dict = self.feature_extraction_class(**self.feat_extract_dict )
__lowerCAmelCase: Union[str, Any] = self.feat_extract_tester.prepare_inputs_for_target()
__lowerCAmelCase: Optional[Any] = feat_extract.model_input_names[0]
__lowerCAmelCase: List[str] = BatchFeature({input_name: speech_inputs} )
__lowerCAmelCase: int = feat_extract.num_mel_bins # hack!
__lowerCAmelCase: Tuple = feat_extract.pad(_snake_case , padding='longest' , return_tensors='np' )[input_name]
__lowerCAmelCase: Union[str, Any] = feat_extract.pad(_snake_case , padding='longest' , return_tensors='pt' )[input_name]
self.assertTrue(abs(input_np.astype(np.floataa ).sum() - input_pt.numpy().astype(np.floataa ).sum() ) < 1E-2 )
def UpperCAmelCase ( self : int ) -> Any:
__lowerCAmelCase: Any = self.feat_extract_dict
__lowerCAmelCase: List[Any] = True
__lowerCAmelCase: Union[str, Any] = self.feature_extraction_class(**_snake_case )
__lowerCAmelCase: Any = self.feat_extract_tester.prepare_inputs_for_target()
__lowerCAmelCase: List[Any] = [len(_snake_case ) for x in speech_inputs]
__lowerCAmelCase: Union[str, Any] = feat_extract.model_input_names[0]
__lowerCAmelCase: Optional[int] = BatchFeature({input_name: speech_inputs} )
__lowerCAmelCase: List[str] = feat_extract.num_mel_bins # hack!
__lowerCAmelCase: str = feat_extract.pad(_snake_case , padding='longest' , return_tensors='np' )
self.assertIn('attention_mask' , _snake_case )
self.assertListEqual(list(processed.attention_mask.shape ) , list(processed[input_name].shape[:2] ) )
self.assertListEqual(processed.attention_mask.sum(-1 ).tolist() , _snake_case )
def UpperCAmelCase ( self : Optional[int] ) -> str:
__lowerCAmelCase: int = self.feat_extract_dict
__lowerCAmelCase: List[str] = True
__lowerCAmelCase: Tuple = self.feature_extraction_class(**_snake_case )
__lowerCAmelCase: List[str] = self.feat_extract_tester.prepare_inputs_for_target()
__lowerCAmelCase: str = [len(_snake_case ) for x in speech_inputs]
__lowerCAmelCase: Optional[Any] = feat_extract.model_input_names[0]
__lowerCAmelCase: Optional[int] = BatchFeature({input_name: speech_inputs} )
__lowerCAmelCase: Optional[Any] = min(_snake_case )
__lowerCAmelCase: Union[str, Any] = feat_extract.num_mel_bins # hack!
__lowerCAmelCase: Tuple = feat_extract.pad(
_snake_case , padding='max_length' , max_length=_snake_case , truncation=_snake_case , return_tensors='np' )
self.assertIn('attention_mask' , _snake_case )
self.assertListEqual(
list(processed_pad.attention_mask.shape ) , [processed_pad[input_name].shape[0], max_length] )
self.assertListEqual(
processed_pad.attention_mask[:, :max_length].sum(-1 ).tolist() , [max_length for x in speech_inputs] )
def UpperCAmelCase ( self : List[Any] , UpperCAmelCase : Optional[int] ) -> Optional[Any]:
from datasets import load_dataset
__lowerCAmelCase: str = load_dataset('hf-internal-testing/librispeech_asr_dummy' , 'clean' , split='validation' )
# automatic decoding with librispeech
__lowerCAmelCase: Dict = ds.sort('id' ).select(range(_snake_case ) )[:num_samples]['audio']
return [x["array"] for x in speech_samples]
def UpperCAmelCase ( self : str ) -> str:
# fmt: off
__lowerCAmelCase: List[Any] = torch.tensor(
[2.38_04E-03, 2.07_52E-03, 1.98_36E-03, 2.10_57E-03, 1.61_74E-03,
3.05_18E-04, 9.15_53E-05, 3.35_69E-04, 9.76_56E-04, 1.83_11E-03,
2.01_42E-03, 2.10_57E-03, 1.73_95E-03, 4.57_76E-04, -3.96_73E-04,
4.57_76E-04, 1.00_71E-03, 9.15_53E-05, 4.88_28E-04, 1.15_97E-03,
7.32_42E-04, 9.46_04E-04, 1.80_05E-03, 1.83_11E-03, 8.85_01E-04,
4.27_25E-04, 4.88_28E-04, 7.32_42E-04, 1.09_86E-03, 2.10_57E-03] )
# fmt: on
__lowerCAmelCase: Union[str, Any] = self._load_datasamples(1 )
__lowerCAmelCase: Optional[int] = SpeechTaFeatureExtractor()
__lowerCAmelCase: List[Any] = feature_extractor(_snake_case , return_tensors='pt' ).input_values
self.assertEquals(input_values.shape , (1, 9_3_6_8_0) )
self.assertTrue(torch.allclose(input_values[0, :3_0] , _snake_case , atol=1E-6 ) )
def UpperCAmelCase ( self : Any ) -> str:
# fmt: off
__lowerCAmelCase: Optional[Any] = torch.tensor(
[-2.6870, -3.0104, -3.1356, -3.5352, -3.0044, -3.0353, -3.4719, -3.6777,
-3.1520, -2.9435, -2.6553, -2.8795, -2.9944, -2.5921, -3.0279, -3.0386,
-3.0864, -3.1291, -3.2353, -2.7444, -2.6831, -2.7287, -3.1761, -3.1571,
-3.2726, -3.0582, -3.1007, -3.4533, -3.4695, -3.0998] )
# fmt: on
__lowerCAmelCase: List[str] = self._load_datasamples(1 )
__lowerCAmelCase: str = SpeechTaFeatureExtractor()
__lowerCAmelCase: Optional[Any] = feature_extractor(audio_target=_snake_case , return_tensors='pt' ).input_values
self.assertEquals(input_values.shape , (1, 3_6_6, 8_0) )
self.assertTrue(torch.allclose(input_values[0, 0, :3_0] , _snake_case , atol=1E-4 ) )
| 322 |
from collections import deque
from .hash_table import HashTable
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
def __init__( self : Optional[Any], *_snake_case : Optional[Any], **_snake_case : List[Any] ) ->Optional[int]:
super().__init__(*_snake_case, **_snake_case )
def lowercase_ ( self : Optional[Any], _snake_case : Tuple, _snake_case : Dict ) ->Dict:
snake_case__ : int = deque([] ) if self.values[key] is None else self.values[key]
self.values[key].appendleft(_snake_case )
snake_case__ : Dict = self.values[key]
def lowercase_ ( self : Any ) ->Optional[Any]:
return (
sum(self.charge_factor - len(_snake_case ) for slot in self.values )
/ self.size_table
* self.charge_factor
)
def lowercase_ ( self : Union[str, Any], _snake_case : str, _snake_case : Optional[int]=None ) ->Optional[Any]:
if not (
len(self.values[key] ) == self.charge_factor and self.values.count(_snake_case ) == 0
):
return key
return super()._collision_resolution(_snake_case, _snake_case )
| 277 | 0 |
import sys
from typing import Tuple
import numpy as np
import torch
from PIL import Image
from torch import nn
from transformers.image_utils import PILImageResampling
from utils import img_tensorize
class __lowerCAmelCase :
def __init__( self , lowerCAmelCase__ , lowerCAmelCase__=sys.maxsize ) -> Any:
'''simple docstring'''
a__ : Any ='bilinear'
a__ : Optional[int] =max_size
a__ : Union[str, Any] =short_edge_length
def __call__( self , lowerCAmelCase__ ) -> List[Any]:
'''simple docstring'''
a__ : List[str] =[]
for img in imgs:
a__ : Union[str, Any] =img.shape[:2]
# later: provide list and randomly choose index for resize
a__ : Dict =np.random.randint(self.short_edge_length[0] , self.short_edge_length[1] + 1 )
if size == 0:
return img
a__ : Any =size * 1.0 / min(_snake_case , _snake_case )
if h < w:
a__ : str =size, scale * w
else:
a__ : Optional[int] =scale * h, size
if max(_snake_case , _snake_case ) > self.max_size:
a__ : Union[str, Any] =self.max_size * 1.0 / max(_snake_case , _snake_case )
a__ : Optional[int] =newh * scale
a__ : Dict =neww * scale
a__ : List[str] =int(neww + 0.5 )
a__ : Dict =int(newh + 0.5 )
if img.dtype == np.uinta:
a__ : str =Image.fromarray(_snake_case )
a__ : List[Any] =pil_image.resize((neww, newh) , PILImageResampling.BILINEAR )
a__ : List[Any] =np.asarray(_snake_case )
else:
a__ : Dict =img.permute(2 , 0 , 1 ).unsqueeze(0 ) # 3, 0, 1) # hw(c) -> nchw
a__ : int =nn.functional.interpolate(
_snake_case , (newh, neww) , mode=self.interp_method , align_corners=_snake_case ).squeeze(0 )
img_augs.append(_snake_case )
return img_augs
class __lowerCAmelCase :
def __init__( self , lowerCAmelCase__ ) -> Union[str, Any]:
'''simple docstring'''
a__ : Optional[int] =ResizeShortestEdge([cfg.INPUT.MIN_SIZE_TEST, cfg.INPUT.MIN_SIZE_TEST] , cfg.INPUT.MAX_SIZE_TEST )
a__ : Optional[Any] =cfg.INPUT.FORMAT
a__ : Optional[Any] =cfg.SIZE_DIVISIBILITY
a__ : int =cfg.PAD_VALUE
a__ : Optional[int] =cfg.INPUT.MAX_SIZE_TEST
a__ : Optional[int] =cfg.MODEL.DEVICE
a__ : List[str] =torch.tensor(cfg.MODEL.PIXEL_STD ).to(self.device ).view(len(cfg.MODEL.PIXEL_STD ) , 1 , 1 )
a__ : Dict =torch.tensor(cfg.MODEL.PIXEL_MEAN ).to(self.device ).view(len(cfg.MODEL.PIXEL_STD ) , 1 , 1 )
a__ : Any =lambda lowerCAmelCase__ : (x - self.pixel_mean) / self.pixel_std
def _lowercase ( self , lowerCAmelCase__ ) -> List[Any]:
'''simple docstring'''
a__ : str =tuple(max(_snake_case ) for s in zip(*[img.shape for img in images] ) )
a__ : Any =[im.shape[-2:] for im in images]
a__ : Any =[
nn.functional.pad(
_snake_case , [0, max_size[-1] - size[1], 0, max_size[-2] - size[0]] , value=self.pad_value , )
for size, im in zip(_snake_case , _snake_case )
]
return torch.stack(_snake_case ), torch.tensor(_snake_case )
def __call__( self , lowerCAmelCase__ , lowerCAmelCase__=False ) -> int:
'''simple docstring'''
with torch.no_grad():
if not isinstance(_snake_case , _snake_case ):
a__ : Optional[Any] =[images]
if single_image:
assert len(_snake_case ) == 1
for i in range(len(_snake_case ) ):
if isinstance(images[i] , torch.Tensor ):
images.insert(_snake_case , images.pop(_snake_case ).to(self.device ).float() )
elif not isinstance(images[i] , torch.Tensor ):
images.insert(
_snake_case , torch.as_tensor(img_tensorize(images.pop(_snake_case ) , input_format=self.input_format ) )
.to(self.device )
.float() , )
# resize smallest edge
a__ : Union[str, Any] =torch.tensor([im.shape[:2] for im in images] )
a__ : str =self.aug(_snake_case )
# transpose images and convert to torch tensors
# images = [torch.as_tensor(i.astype("float32")).permute(2, 0, 1).to(self.device) for i in images]
# now normalize before pad to avoid useless arithmetic
a__ : str =[self.normalizer(_snake_case ) for x in images]
# now pad them to do the following operations
a__ : List[str] =self.pad(_snake_case )
# Normalize
if self.size_divisibility > 0:
raise NotImplementedError()
# pad
a__ : Tuple =torch.true_divide(_snake_case , _snake_case )
if single_image:
return images[0], sizes[0], scales_yx[0]
else:
return images, sizes, scales_yx
def _A ( SCREAMING_SNAKE_CASE : List[str] , SCREAMING_SNAKE_CASE : Any ):
"""simple docstring"""
boxes[:, 0::2] *= scale_yx[:, 1]
boxes[:, 1::2] *= scale_yx[:, 0]
return boxes
def _A ( SCREAMING_SNAKE_CASE : List[str] , SCREAMING_SNAKE_CASE : Tuple[int, int] ):
"""simple docstring"""
assert torch.isfinite(SCREAMING_SNAKE_CASE ).all(), "Box tensor contains infinite or NaN!"
a__ : Dict =box_size
tensor[:, 0].clamp_(min=0 , max=SCREAMING_SNAKE_CASE )
tensor[:, 1].clamp_(min=0 , max=SCREAMING_SNAKE_CASE )
tensor[:, 2].clamp_(min=0 , max=SCREAMING_SNAKE_CASE )
tensor[:, 3].clamp_(min=0 , max=SCREAMING_SNAKE_CASE )
| 95 |
def lowercase_ (A : Union[str, Any] , A : List[str] , A : int , A : Optional[int] ):
global f # a global dp table for knapsack
if f[i][j] < 0:
if j < wt[i - 1]:
snake_case__ : Union[str, Any] = mf_knapsack(i - 1 , A , A , A )
else:
snake_case__ : Any = max(
mf_knapsack(i - 1 , A , A , A ) , mf_knapsack(i - 1 , A , A , j - wt[i - 1] ) + val[i - 1] , )
snake_case__ : Optional[int] = val
return f[i][j]
def lowercase_ (A : Optional[int] , A : Union[str, Any] , A : str , A : Dict ):
snake_case__ : int = [[0] * (w + 1) for _ in range(n + 1 )]
for i in range(1 , n + 1 ):
for w_ in range(1 , w + 1 ):
if wt[i - 1] <= w_:
snake_case__ : Union[str, Any] = max(val[i - 1] + dp[i - 1][w_ - wt[i - 1]] , dp[i - 1][w_] )
else:
snake_case__ : str = dp[i - 1][w_]
return dp[n][w_], dp
def lowercase_ (A : int , A : list , A : list ):
if not (isinstance(A , (list, tuple) ) and isinstance(A , (list, tuple) )):
raise ValueError(
'Both the weights and values vectors must be either lists or tuples' )
snake_case__ : Dict = len(A )
if num_items != len(A ):
snake_case__ : str = (
'The number of weights must be the same as the number of values.\n'
F'''But got {num_items} weights and {len(A )} values'''
)
raise ValueError(A )
for i in range(A ):
if not isinstance(wt[i] , A ):
snake_case__ : Optional[int] = (
'All weights must be integers but got weight of '
F'''type {type(wt[i] )} at index {i}'''
)
raise TypeError(A )
snake_case__ , snake_case__ : Optional[int] = knapsack(A , A , A , A )
snake_case__ : set = set()
_construct_solution(A , A , A , A , A )
return optimal_val, example_optional_set
def lowercase_ (A : list , A : list , A : int , A : int , A : set ):
# for the current item i at a maximum weight j to be part of an optimal subset,
# the optimal value at (i, j) must be greater than the optimal value at (i-1, j).
# where i - 1 means considering only the previous items at the given maximum weight
if i > 0 and j > 0:
if dp[i - 1][j] == dp[i][j]:
_construct_solution(A , A , i - 1 , A , A )
else:
optimal_set.add(A )
_construct_solution(A , A , i - 1 , j - wt[i - 1] , A )
if __name__ == "__main__":
a_ :Any = [3, 2, 4, 4]
a_ :List[Any] = [4, 3, 2, 3]
a_ :Union[str, Any] = 4
a_ :List[str] = 6
a_ :Union[str, Any] = [[0] * (w + 1)] + [[0] + [-1] * (w + 1) for _ in range(n + 1)]
a_ , a_ :List[Any] = knapsack(w, wt, val, n)
print(optimal_solution)
print(mf_knapsack(n, wt, val, w)) # switched the n and w
# testing the dynamic programming problem with example
# the optimal subset for the above example are items 3 and 4
a_ , a_ :Any = knapsack_with_example_solution(w, wt, val)
assert optimal_solution == 8
assert optimal_subset == {3, 4}
print("optimal_value = ", optimal_solution)
print("An optimal subset corresponding to the optimal value", optimal_subset)
| 277 | 0 |
from __future__ import annotations
def a ( snake_case__: list[float] ):
'''simple docstring'''
if len(snake_case__ ) < 2:
raise ValueError('''Monogons and Digons are not polygons in the Euclidean space''' )
if any(i <= 0 for i in nums ):
raise ValueError('''All values must be greater than 0''' )
lowercase_ = nums.copy()
copy_nums.sort()
return copy_nums[-1] < sum(copy_nums[:-1] )
if __name__ == "__main__":
import doctest
doctest.testmod()
| 30 |
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_flax_available, is_torch_available
a_ :int = {
"configuration_longt5": ["LONGT5_PRETRAINED_CONFIG_ARCHIVE_MAP", "LongT5Config", "LongT5OnnxConfig"],
}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :List[str] = [
"LONGT5_PRETRAINED_MODEL_ARCHIVE_LIST",
"LongT5EncoderModel",
"LongT5ForConditionalGeneration",
"LongT5Model",
"LongT5PreTrainedModel",
]
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :int = [
"FlaxLongT5ForConditionalGeneration",
"FlaxLongT5Model",
"FlaxLongT5PreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_longta import LONGT5_PRETRAINED_CONFIG_ARCHIVE_MAP, LongTaConfig, LongTaOnnxConfig
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_longta import (
LONGT5_PRETRAINED_MODEL_ARCHIVE_LIST,
LongTaEncoderModel,
LongTaForConditionalGeneration,
LongTaModel,
LongTaPreTrainedModel,
)
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_flax_longta import (
FlaxLongTaForConditionalGeneration,
FlaxLongTaModel,
FlaxLongTaPreTrainedModel,
)
else:
import sys
a_ :Optional[int] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)
| 277 | 0 |
"""simple docstring"""
import argparse
import json
import os
import fairseq
import torch
from fairseq.data import Dictionary
from transformers import (
WavaVecaConfig,
WavaVecaCTCTokenizer,
WavaVecaFeatureExtractor,
WavaVecaForCTC,
WavaVecaForPreTraining,
WavaVecaProcessor,
logging,
)
from transformers.models.wavaveca.modeling_wavaveca import WavaVecaForSequenceClassification
logging.set_verbosity_info()
lowercase_ = logging.get_logger(__name__)
lowercase_ = {
"post_extract_proj": "feature_projection.projection",
"encoder.pos_conv.0": "encoder.pos_conv_embed.conv",
"self_attn.k_proj": "encoder.layers.*.attention.k_proj",
"self_attn.v_proj": "encoder.layers.*.attention.v_proj",
"self_attn.q_proj": "encoder.layers.*.attention.q_proj",
"self_attn.out_proj": "encoder.layers.*.attention.out_proj",
"self_attn_layer_norm": "encoder.layers.*.layer_norm",
"fc1": "encoder.layers.*.feed_forward.intermediate_dense",
"fc2": "encoder.layers.*.feed_forward.output_dense",
"final_layer_norm": "encoder.layers.*.final_layer_norm",
"encoder.layer_norm": "encoder.layer_norm",
"adapter_layer": "encoder.layers.*.adapter_layer",
"w2v_model.layer_norm": "feature_projection.layer_norm",
"quantizer.weight_proj": "quantizer.weight_proj",
"quantizer.vars": "quantizer.codevectors",
"project_q": "project_q",
"final_proj": "project_hid",
"w2v_encoder.proj": "lm_head",
"mask_emb": "masked_spec_embed",
"pooling_layer.linear": "projector",
"pooling_layer.projection": "classifier",
}
lowercase_ = [
"lm_head",
"quantizer.weight_proj",
"quantizer.codevectors",
"project_q",
"project_hid",
"projector",
"classifier",
]
def lowercase ( lowerCAmelCase__ : Dict ) -> Optional[Any]:
__a = {}
with open(lowerCAmelCase__ , '''r''' ) as file:
for line_number, line in enumerate(lowerCAmelCase__ ):
__a = line.strip()
if line:
__a = line.split()
__a = line_number
__a = words[0]
__a = value
return result
def lowercase ( lowerCAmelCase__ : int , lowerCAmelCase__ : int , lowerCAmelCase__ : Optional[int] , lowerCAmelCase__ : Optional[Any] , lowerCAmelCase__ : Tuple ) -> Tuple:
for attribute in key.split('''.''' ):
__a = getattr(lowerCAmelCase__ , lowerCAmelCase__ )
__a = None
for param_key in PARAM_MAPPING.keys():
if full_name.endswith(lowerCAmelCase__ ):
__a = PARAM_MAPPING[full_name.split('''.''' )[-1]]
__a = 'param'
if weight_type is not None and weight_type != "param":
__a = getattr(lowerCAmelCase__ , lowerCAmelCase__ ).shape
elif weight_type is not None and weight_type == "param":
__a = hf_pointer
for attribute in hf_param_name.split('''.''' ):
__a = getattr(lowerCAmelCase__ , lowerCAmelCase__ )
__a = shape_pointer.shape
# let's reduce dimension
__a = value[0]
else:
__a = hf_pointer.shape
if hf_shape != value.shape:
raise ValueError(
f'''Shape of hf {key + '.' + weight_type if weight_type is not None else ''} is {hf_shape}, but should be'''
f''' {value.shape} for {full_name}''' )
if weight_type == "weight":
__a = value
elif weight_type == "weight_g":
__a = value
elif weight_type == "weight_v":
__a = value
elif weight_type == "bias":
__a = value
elif weight_type == "param":
for attribute in hf_param_name.split('''.''' ):
__a = getattr(lowerCAmelCase__ , lowerCAmelCase__ )
__a = value
else:
__a = value
logger.info(f'''{key + '.' + weight_type if weight_type is not None else ''} was initialized from {full_name}.''' )
def lowercase ( lowerCAmelCase__ : Tuple , lowerCAmelCase__ : List[Any] , lowerCAmelCase__ : int , lowerCAmelCase__ : str , lowerCAmelCase__ : Tuple ) -> int:
__a = None
for param_key in PARAM_MAPPING.keys():
if full_name.endswith(lowerCAmelCase__ ):
__a = PARAM_MAPPING[full_name.split('''.''' )[-1]]
__a = 'param'
if weight_type is not None and weight_type != "param":
__a = '.'.join([key, weight_type] )
elif weight_type is not None and weight_type == "param":
__a = '.'.join([key, hf_param_name] )
else:
__a = key
__a = value if 'lm_head' in full_key else value[0]
lowercase_ = {
"W_a": "linear_1.weight",
"W_b": "linear_2.weight",
"b_a": "linear_1.bias",
"b_b": "linear_2.bias",
"ln_W": "norm.weight",
"ln_b": "norm.bias",
}
def lowercase ( lowerCAmelCase__ : str , lowerCAmelCase__ : Optional[Any] , lowerCAmelCase__ : Optional[Any]=None , lowerCAmelCase__ : List[str]=None ) -> Any:
__a = False
for key, mapped_key in MAPPING.items():
__a = 'wav2vec2.' + mapped_key if mapped_key not in TOP_LEVEL_KEYS else mapped_key
if key in name or key.split('''w2v_model.''' )[-1] == name.split('''.''' )[0]:
__a = True
if "*" in mapped_key:
__a = name.split(lowerCAmelCase__ )[0].split('''.''' )[-2]
__a = mapped_key.replace('''*''' , lowerCAmelCase__ )
if "weight_g" in name:
__a = 'weight_g'
elif "weight_v" in name:
__a = 'weight_v'
elif "bias" in name:
__a = 'bias'
elif "weight" in name:
# TODO: don't match quantizer.weight_proj
__a = 'weight'
else:
__a = None
if hf_dict is not None:
rename_dict(lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ )
else:
set_recursively(lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ )
return is_used
return is_used
def lowercase ( lowerCAmelCase__ : Optional[Any] , lowerCAmelCase__ : Dict , lowerCAmelCase__ : Optional[int] ) -> str:
__a = []
__a = fairseq_model.state_dict()
__a = hf_model.wavaveca.feature_extractor
for name, value in fairseq_dict.items():
__a = False
if "conv_layers" in name:
load_conv_layer(
lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , hf_model.config.feat_extract_norm == '''group''' , )
__a = True
else:
__a = load_wavaveca_layer(lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ )
if not is_used:
unused_weights.append(lowerCAmelCase__ )
logger.warning(f'''Unused weights: {unused_weights}''' )
def lowercase ( lowerCAmelCase__ : Dict , lowerCAmelCase__ : Optional[Any] , lowerCAmelCase__ : Tuple , lowerCAmelCase__ : str , lowerCAmelCase__ : List[str] ) -> List[str]:
__a = full_name.split('''conv_layers.''' )[-1]
__a = name.split('''.''' )
__a = int(items[0] )
__a = int(items[1] )
if type_id == 0:
if "bias" in name:
if value.shape != feature_extractor.conv_layers[layer_id].conv.bias.data.shape:
raise ValueError(
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.''' )
__a = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
if value.shape != feature_extractor.conv_layers[layer_id].conv.weight.data.shape:
raise ValueError(
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.''' )
__a = value
logger.info(f'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
if value.shape != feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape:
raise ValueError(
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape} was found.''' )
__a = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
if value.shape != feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape:
raise ValueError(
f'''{full_name} has size {value.shape}, but'''
f''' {feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape} was found.''' )
__a = value
logger.info(f'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(lowerCAmelCase__ )
@torch.no_grad()
def lowercase ( lowerCAmelCase__ : Union[str, Any] , lowerCAmelCase__ : str , lowerCAmelCase__ : Tuple=None , lowerCAmelCase__ : List[str]=None , lowerCAmelCase__ : Any=True , lowerCAmelCase__ : Optional[int]=False ) -> Dict:
if config_path is not None:
__a = WavaVecaConfig.from_pretrained(lowerCAmelCase__ )
else:
__a = WavaVecaConfig()
if is_seq_class:
__a = read_txt_into_dict(lowerCAmelCase__ )
__a = idalabel
__a = WavaVecaForSequenceClassification(lowerCAmelCase__ )
__a = WavaVecaFeatureExtractor(
feature_size=1 , sampling_rate=16000 , padding_value=0 , do_normalize=lowerCAmelCase__ , return_attention_mask=lowerCAmelCase__ , )
feature_extractor.save_pretrained(lowerCAmelCase__ )
elif is_finetuned:
if dict_path:
__a = Dictionary.load(lowerCAmelCase__ )
# important change bos & pad token id since CTC symbol is <pad> and
# not <s> as in fairseq
__a = target_dict.pad_index
__a = target_dict.bos_index
__a = target_dict.eos_index
__a = len(target_dict.symbols )
__a = os.path.join(lowerCAmelCase__ , '''vocab.json''' )
if not os.path.isdir(lowerCAmelCase__ ):
logger.error('''--pytorch_dump_folder_path ({}) should be a directory'''.format(lowerCAmelCase__ ) )
return
os.makedirs(lowerCAmelCase__ , exist_ok=lowerCAmelCase__ )
__a = target_dict.indices
# fairseq has the <pad> and <s> switched
__a = 0
__a = 1
with open(lowerCAmelCase__ , '''w''' , encoding='''utf-8''' ) as vocab_handle:
json.dump(lowerCAmelCase__ , lowerCAmelCase__ )
__a = WavaVecaCTCTokenizer(
lowerCAmelCase__ , unk_token=target_dict.unk_word , pad_token=target_dict.pad_word , bos_token=target_dict.bos_word , eos_token=target_dict.eos_word , word_delimiter_token='''|''' , do_lower_case=lowerCAmelCase__ , )
__a = True if config.feat_extract_norm == 'layer' else False
__a = WavaVecaFeatureExtractor(
feature_size=1 , sampling_rate=16000 , padding_value=0 , do_normalize=lowerCAmelCase__ , return_attention_mask=lowerCAmelCase__ , )
__a = WavaVecaProcessor(feature_extractor=lowerCAmelCase__ , tokenizer=lowerCAmelCase__ )
processor.save_pretrained(lowerCAmelCase__ )
__a = WavaVecaForCTC(lowerCAmelCase__ )
else:
__a = WavaVecaForPreTraining(lowerCAmelCase__ )
if is_finetuned or is_seq_class:
__a = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] , arg_overrides={'''data''': '''/'''.join(dict_path.split('''/''' )[:-1] )} )
else:
__a = argparse.Namespace(task='''audio_pretraining''' )
__a = fairseq.tasks.setup_task(lowerCAmelCase__ )
__a = fairseq.checkpoint_utils.load_model_ensemble_and_task([checkpoint_path] , task=lowerCAmelCase__ )
__a = model[0].eval()
recursively_load_weights(lowerCAmelCase__ , lowerCAmelCase__ , not is_finetuned )
hf_wavavec.save_pretrained(lowerCAmelCase__ )
if __name__ == "__main__":
lowercase_ = argparse.ArgumentParser()
parser.add_argument("--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model.")
parser.add_argument("--checkpoint_path", default=None, type=str, help="Path to fairseq checkpoint")
parser.add_argument("--dict_path", default=None, type=str, help="Path to dict of fine-tuned model")
parser.add_argument("--config_path", default=None, type=str, help="Path to hf config.json of model to convert")
parser.add_argument(
"--not_finetuned", action="store_true", help="Whether the model to convert is a fine-tuned model or not"
)
parser.add_argument(
"--is_seq_class",
action="store_true",
help="Whether the model to convert is a fine-tuned sequence classification model or not",
)
lowercase_ = parser.parse_args()
lowercase_ = not args.not_finetuned and not args.is_seq_class
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.config_path,
args.dict_path,
is_finetuned,
args.is_seq_class,
)
| 45 |
import argparse
import json
import os
import torch
from torch import nn
from transformers import NllbMoeConfig, NllbMoeModel
from transformers.modeling_utils import dtype_byte_size
from transformers.utils import WEIGHTS_INDEX_NAME, WEIGHTS_NAME
def lowercase_ (A : List[str] ):
snake_case__ : Tuple = [
'encoder.version',
'decoder.version',
'model.encoder.version',
'model.decoder.version',
'decoder.output_projection.weight',
'_float_tensor',
'encoder.embed_positions._float_tensor',
'decoder.embed_positions._float_tensor',
]
for k in ignore_keys:
state_dict.pop(A , A )
def lowercase_ (A : str ):
snake_case__ , snake_case__ : Union[str, Any] = emb.weight.shape
snake_case__ : str = nn.Linear(A , A , bias=A )
snake_case__ : str = emb.weight.data
return lin_layer
def lowercase_ (A : Optional[int] , A : Union[str, Any]=None ):
snake_case__ : Any = {}
for old_key in state_dict.keys():
snake_case__ : Tuple = old_key
if "moe_layer.experts." in key:
if expert_idx is not None:
snake_case__ : int = key.replace('moe_layer.experts.0' , F'''ffn.experts.expert_{expert_idx}''' )
else:
snake_case__ : Any = key.replace('moe_layer.experts.' , 'ffn.experts.expert_' )
if "gate" in key:
snake_case__ : Dict = key.replace('.moe_layer.gate.wg' , '.ffn.router.classifier' )
if "fc2" and "experts" not in key:
snake_case__ : str = key.replace('.fc2.' , '.ffn.fc2.' )
if "fc1" and "experts" not in key:
snake_case__ : str = key.replace('.fc1.' , '.ffn.fc1.' )
if ".encoder_attn." in key:
snake_case__ : Tuple = key.replace('.encoder_attn.' , '.cross_attention.' )
if "encoder_attn_layer_norm" in key:
snake_case__ : Tuple = key.replace('encoder_attn_layer_norm' , 'cross_attention_layer_norm' )
if "final_layer_norm" in key:
snake_case__ : Optional[int] = key.replace('final_layer_norm' , 'ff_layer_norm' )
snake_case__ : Dict = state_dict[old_key]
return new_dict
def lowercase_ (A : List[Any] , A : Tuple , A : List[Any] , A : List[str] , A : str = WEIGHTS_NAME ):
snake_case__ : Dict = []
snake_case__ : str = 0
os.makedirs(A , exist_ok=A )
for expert in range(A ):
snake_case__ : Tuple = switch_checkpoint_path + F'''-rank-{expert}.pt'''
if os.path.isfile(A ):
snake_case__ : Optional[Any] = torch.load(A )['model']
remove_ignore_keys_(A )
snake_case__ : Optional[Any] = rename_fairseq_keys(A , A )
snake_case__ : Dict = os.path.join(
A , weights_name.replace('.bin' , F'''-{len(A )+1:05d}-of-???.bin''' ) )
torch.save(A , A )
sharded_state_dicts.append(expert_state.keys() )
total_size += sum([value.numel() for key, value in expert_state.items()] ) * dtype_byte_size(
expert_state[list(A )[0]].dtype )
# Add the last block
snake_case__ : Tuple = os.path.join(A , weights_name.replace('.bin' , F'''-{len(A )+1:05d}-of-???.bin''' ) )
snake_case__ : Union[str, Any] = torch.load(switch_checkpoint_path + '-shared.pt' )['model']
remove_ignore_keys_(A )
snake_case__ : str = rename_fairseq_keys(A , A )
snake_case__ : Any = shared_weights['decoder.embed_tokens.weight']
sharded_state_dicts.append(shared_weights.keys() )
# If we only have the shared weights (dummy model/experts saved on the same file)
if len(A ) == 1:
snake_case__ : Any = os.path.join(A , A )
torch.save(A , A )
return {weights_name: sharded_state_dicts[0]}, None
else:
torch.save(A , A )
# Otherwise, let's build the index
snake_case__ : Tuple = {}
for idx, shard in enumerate(A ):
snake_case__ : Optional[int] = weights_name.replace('.bin' , F'''-{idx+1:05d}-of-{len(A ):05d}.bin''' )
snake_case__ : List[Any] = os.path.join(A , weights_name.replace('.bin' , F'''-{idx+1:05d}-of-???.bin''' ) )
os.rename(A , os.path.join(A , A ) )
for key in shard:
snake_case__ : Any = shard_file
# Add the metadata
snake_case__ : int = {'total_size': total_size}
snake_case__ : Dict = {'metadata': metadata, 'weight_map': weight_map}
with open(os.path.join(A , A ) , 'w' , encoding='utf-8' ) as f:
snake_case__ : Any = json.dumps(A , indent=2 , sort_keys=A ) + '\n'
f.write(A )
return metadata, index
if __name__ == "__main__":
a_ :int = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--nllb_moe_checkpoint_path",
default="/home/arthur_huggingface_co/fairseq/weights/checkpoints/model_moe_54b/checkpoint_2_300000",
type=str,
required=False,
help="Path to a directory containing a folder per layer. Follows the original Google format.",
)
parser.add_argument("--dtype", default="float32", type=str, required=False, help="dtype of the saved model")
parser.add_argument(
"--pytorch_dump_folder_path",
default="/home/arthur_huggingface_co/fairseq/weights/checkpoints/hf-converted-moe-54b",
type=str,
required=False,
help="Path to the output pytorch model.",
)
a_ :Optional[Any] = parser.parse_args()
a_ , a_ :Optional[Any] = shard_on_the_fly(
args.nllb_moe_checkpoint_path,
args.pytorch_dump_folder_path,
128,
args.dtype,
)
a_ :List[str] = NllbMoeConfig.from_pretrained(
"facebook/nllb-200-3.3B", encoder_sparse_step=4, decoder_sparse_step=4, num_experts=128
)
config.save_pretrained(args.pytorch_dump_folder_path)
a_ :int = NllbMoeModel.from_pretrained(args.pytorch_dump_folder_path)
print("Done")
model.save_pretrained(args.pytorch_dump_folder_path)
| 277 | 0 |
'''simple docstring'''
import json
from typing import List, Optional, Tuple
from tokenizers import pre_tokenizers, processors
from ...tokenization_utils_base import AddedToken, BatchEncoding
from ...tokenization_utils_fast import PreTrainedTokenizerFast
from ...utils import logging
from .tokenization_bart import BartTokenizer
_lowerCAmelCase = logging.get_logger(__name__)
_lowerCAmelCase = {"vocab_file": "vocab.json", "merges_file": "merges.txt", "tokenizer_file": "tokenizer.json"}
# See all BART models at https://huggingface.co/models?filter=bart
_lowerCAmelCase = {
"vocab_file": {
"facebook/bart-base": "https://huggingface.co/facebook/bart-base/resolve/main/vocab.json",
"facebook/bart-large": "https://huggingface.co/facebook/bart-large/resolve/main/vocab.json",
"facebook/bart-large-mnli": "https://huggingface.co/facebook/bart-large-mnli/resolve/main/vocab.json",
"facebook/bart-large-cnn": "https://huggingface.co/facebook/bart-large-cnn/resolve/main/vocab.json",
"facebook/bart-large-xsum": "https://huggingface.co/facebook/bart-large-xsum/resolve/main/vocab.json",
"yjernite/bart_eli5": "https://huggingface.co/yjernite/bart_eli5/resolve/main/vocab.json",
},
"merges_file": {
"facebook/bart-base": "https://huggingface.co/facebook/bart-base/resolve/main/merges.txt",
"facebook/bart-large": "https://huggingface.co/facebook/bart-large/resolve/main/merges.txt",
"facebook/bart-large-mnli": "https://huggingface.co/facebook/bart-large-mnli/resolve/main/merges.txt",
"facebook/bart-large-cnn": "https://huggingface.co/facebook/bart-large-cnn/resolve/main/merges.txt",
"facebook/bart-large-xsum": "https://huggingface.co/facebook/bart-large-xsum/resolve/main/merges.txt",
"yjernite/bart_eli5": "https://huggingface.co/yjernite/bart_eli5/resolve/main/merges.txt",
},
"tokenizer_file": {
"facebook/bart-base": "https://huggingface.co/facebook/bart-base/resolve/main/tokenizer.json",
"facebook/bart-large": "https://huggingface.co/facebook/bart-large/resolve/main/tokenizer.json",
"facebook/bart-large-mnli": "https://huggingface.co/facebook/bart-large-mnli/resolve/main/tokenizer.json",
"facebook/bart-large-cnn": "https://huggingface.co/facebook/bart-large-cnn/resolve/main/tokenizer.json",
"facebook/bart-large-xsum": "https://huggingface.co/facebook/bart-large-xsum/resolve/main/tokenizer.json",
"yjernite/bart_eli5": "https://huggingface.co/yjernite/bart_eli5/resolve/main/tokenizer.json",
},
}
_lowerCAmelCase = {
"facebook/bart-base": 1024,
"facebook/bart-large": 1024,
"facebook/bart-large-mnli": 1024,
"facebook/bart-large-cnn": 1024,
"facebook/bart-large-xsum": 1024,
"yjernite/bart_eli5": 1024,
}
class A ( lowerCAmelCase_ ):
'''simple docstring'''
A = VOCAB_FILES_NAMES
A = PRETRAINED_VOCAB_FILES_MAP
A = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
A = ["input_ids", "attention_mask"]
A = BartTokenizer
def __init__(self , _UpperCAmelCase=None , _UpperCAmelCase=None , _UpperCAmelCase=None , _UpperCAmelCase="replace" , _UpperCAmelCase="<s>" , _UpperCAmelCase="</s>" , _UpperCAmelCase="</s>" , _UpperCAmelCase="<s>" , _UpperCAmelCase="<unk>" , _UpperCAmelCase="<pad>" , _UpperCAmelCase="<mask>" , _UpperCAmelCase=False , _UpperCAmelCase=True , **_UpperCAmelCase , ) -> int:
super().__init__(
_snake_case , _snake_case , tokenizer_file=_snake_case , errors=_snake_case , bos_token=_snake_case , eos_token=_snake_case , sep_token=_snake_case , cls_token=_snake_case , unk_token=_snake_case , pad_token=_snake_case , mask_token=_snake_case , add_prefix_space=_snake_case , trim_offsets=_snake_case , **_snake_case , )
__UpperCamelCase : str = json.loads(self.backend_tokenizer.pre_tokenizer.__getstate__() )
if pre_tok_state.get("add_prefix_space" , _snake_case ) != add_prefix_space:
__UpperCamelCase : Tuple = getattr(_snake_case , pre_tok_state.pop("type" ) )
__UpperCamelCase : List[Any] = add_prefix_space
__UpperCamelCase : str = pre_tok_class(**_snake_case )
__UpperCamelCase : int = add_prefix_space
# the pre_tokenizer is already updated in the GPT2TokenizerFast `__init__`
__UpperCamelCase : Any = 'post_processor'
__UpperCamelCase : List[Any] = getattr(self.backend_tokenizer , _snake_case , _snake_case )
if tokenizer_component_instance:
__UpperCamelCase : Union[str, Any] = json.loads(tokenizer_component_instance.__getstate__() )
# The lists 'sep' and 'cls' must be cased in tuples for the object `post_processor_class`
if "sep" in state:
__UpperCamelCase : Tuple = tuple(state["sep"] )
if "cls" in state:
__UpperCamelCase : Optional[Any] = tuple(state["cls"] )
__UpperCamelCase : int = False
if state.get("add_prefix_space" , _snake_case ) != add_prefix_space:
__UpperCamelCase : int = add_prefix_space
__UpperCamelCase : Tuple = True
if state.get("trim_offsets" , _snake_case ) != trim_offsets:
__UpperCamelCase : Optional[Any] = trim_offsets
__UpperCamelCase : Tuple = True
if changes_to_apply:
__UpperCamelCase : Union[str, Any] = getattr(_snake_case , state.pop("type" ) )
__UpperCamelCase : List[str] = component_class(**_snake_case )
setattr(self.backend_tokenizer , _snake_case , _snake_case )
@property
def a_ (self ) -> str:
if self._mask_token is None:
if self.verbose:
logger.error("Using mask_token, but it is not set yet." )
return None
return str(self._mask_token )
@mask_token.setter
def a_ (self , _UpperCAmelCase ) -> Dict:
__UpperCamelCase : List[str] = AddedToken(_snake_case , lstrip=_snake_case , rstrip=_snake_case ) if isinstance(_snake_case , _snake_case ) else value
__UpperCamelCase : Union[str, Any] = value
def a_ (self , *_UpperCAmelCase , **_UpperCAmelCase ) -> BatchEncoding:
__UpperCamelCase : str = kwargs.get("is_split_into_words" , _snake_case )
if is_split_into_words and not self.add_prefix_space:
raise ValueError(
f"You need to instantiate {self.__class__.__name__} with add_prefix_space=True "
"to use it with pretokenized inputs." )
return super()._batch_encode_plus(*_snake_case , **_snake_case )
def a_ (self , *_UpperCAmelCase , **_UpperCAmelCase ) -> BatchEncoding:
__UpperCamelCase : List[str] = kwargs.get("is_split_into_words" , _snake_case )
if is_split_into_words and not self.add_prefix_space:
raise ValueError(
f"You need to instantiate {self.__class__.__name__} with add_prefix_space=True "
"to use it with pretokenized inputs." )
return super()._encode_plus(*_snake_case , **_snake_case )
def a_ (self , _UpperCAmelCase , _UpperCAmelCase = None ) -> Tuple[str]:
__UpperCamelCase : Dict = self._tokenizer.model.save(_snake_case , name=_snake_case )
return tuple(_snake_case )
def a_ (self , _UpperCAmelCase , _UpperCAmelCase=None ) -> Dict:
__UpperCamelCase : Tuple = [self.bos_token_id] + token_ids_a + [self.eos_token_id]
if token_ids_a is None:
return output
return output + [self.eos_token_id] + token_ids_a + [self.eos_token_id]
def a_ (self , _UpperCAmelCase , _UpperCAmelCase = None ) -> List[int]:
__UpperCamelCase : Any = [self.sep_token_id]
__UpperCamelCase : Optional[Any] = [self.cls_token_id]
if token_ids_a is None:
return len(cls + token_ids_a + sep ) * [0]
return len(cls + token_ids_a + sep + sep + token_ids_a + sep ) * [0]
| 298 |
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_sentencepiece_available,
is_tokenizers_available,
is_torch_available,
)
a_ :Optional[Any] = {"configuration_reformer": ["REFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP", "ReformerConfig"]}
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :str = ["ReformerTokenizer"]
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :int = ["ReformerTokenizerFast"]
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :List[str] = [
"REFORMER_PRETRAINED_MODEL_ARCHIVE_LIST",
"ReformerAttention",
"ReformerForMaskedLM",
"ReformerForQuestionAnswering",
"ReformerForSequenceClassification",
"ReformerLayer",
"ReformerModel",
"ReformerModelWithLMHead",
"ReformerPreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_reformer import REFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, ReformerConfig
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_reformer import ReformerTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_reformer_fast import ReformerTokenizerFast
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_reformer import (
REFORMER_PRETRAINED_MODEL_ARCHIVE_LIST,
ReformerAttention,
ReformerForMaskedLM,
ReformerForQuestionAnswering,
ReformerForSequenceClassification,
ReformerLayer,
ReformerModel,
ReformerModelWithLMHead,
ReformerPreTrainedModel,
)
else:
import sys
a_ :Optional[Any] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)
| 277 | 0 |
import shutil
import tempfile
import unittest
import numpy as np
import pytest
from transformers.testing_utils import require_vision
from transformers.utils import is_vision_available
if is_vision_available():
from PIL import Image
from transformers import AutoProcessor, BlipaProcessor, BlipImageProcessor, GPTaTokenizer, PreTrainedTokenizerFast
@require_vision
class lowercase ( unittest.TestCase ):
def a__ ( self ) -> Tuple:
_A : List[Any] = tempfile.mkdtemp()
_A : str = BlipImageProcessor()
_A : Any = GPTaTokenizer.from_pretrained("""hf-internal-testing/tiny-random-GPT2Model""" )
_A : int = BlipaProcessor(_snake_case , _snake_case )
processor.save_pretrained(self.tmpdirname )
def a__ ( self , **_a ) -> Dict:
return AutoProcessor.from_pretrained(self.tmpdirname , **_snake_case ).tokenizer
def a__ ( self , **_a ) -> str:
return AutoProcessor.from_pretrained(self.tmpdirname , **_snake_case ).image_processor
def a__ ( self ) -> List[str]:
shutil.rmtree(self.tmpdirname )
def a__ ( self ) -> Any:
_A : str = [np.random.randint(255 , size=(3, 30, 400) , dtype=np.uinta )]
_A : Optional[int] = [Image.fromarray(np.moveaxis(_snake_case , 0 , -1 ) ) for x in image_inputs]
return image_inputs
def a__ ( self ) -> Optional[Any]:
_A : Union[str, Any] = BlipaProcessor(tokenizer=self.get_tokenizer() , image_processor=self.get_image_processor() )
processor.save_pretrained(self.tmpdirname )
_A : Optional[Any] = self.get_tokenizer(bos_token="""(BOS)""" , eos_token="""(EOS)""" )
_A : int = self.get_image_processor(do_normalize=_snake_case , padding_value=1.0 )
_A : Optional[int] = BlipaProcessor.from_pretrained(
self.tmpdirname , bos_token="""(BOS)""" , eos_token="""(EOS)""" , do_normalize=_snake_case , padding_value=1.0 )
self.assertEqual(processor.tokenizer.get_vocab() , tokenizer_add_kwargs.get_vocab() )
self.assertIsInstance(processor.tokenizer , _snake_case )
self.assertEqual(processor.image_processor.to_json_string() , image_processor_add_kwargs.to_json_string() )
self.assertIsInstance(processor.image_processor , _snake_case )
def a__ ( self ) -> List[Any]:
_A : Optional[Any] = self.get_image_processor()
_A : List[Any] = self.get_tokenizer()
_A : Union[str, Any] = BlipaProcessor(tokenizer=_snake_case , image_processor=_snake_case )
_A : Tuple = self.prepare_image_inputs()
_A : Optional[Any] = image_processor(_snake_case , return_tensors="""np""" )
_A : Tuple = processor(images=_snake_case , return_tensors="""np""" )
for key in input_feat_extract.keys():
self.assertAlmostEqual(input_feat_extract[key].sum() , input_processor[key].sum() , delta=1e-2 )
def a__ ( self ) -> Union[str, Any]:
_A : str = self.get_image_processor()
_A : List[Any] = self.get_tokenizer()
_A : int = BlipaProcessor(tokenizer=_snake_case , image_processor=_snake_case )
_A : List[Any] = 'lower newer'
_A : str = processor(text=_snake_case )
_A : Dict = tokenizer(_snake_case , return_token_type_ids=_snake_case )
for key in encoded_tok.keys():
self.assertListEqual(encoded_tok[key] , encoded_processor[key] )
def a__ ( self ) -> List[Any]:
_A : Tuple = self.get_image_processor()
_A : Optional[int] = self.get_tokenizer()
_A : Optional[Any] = BlipaProcessor(tokenizer=_snake_case , image_processor=_snake_case )
_A : Dict = 'lower newer'
_A : Dict = self.prepare_image_inputs()
_A : Union[str, Any] = processor(text=_snake_case , images=_snake_case )
self.assertListEqual(list(inputs.keys() ) , ["""pixel_values""", """input_ids""", """attention_mask"""] )
# test if it raises when no input is passed
with pytest.raises(_snake_case ):
processor()
def a__ ( self ) -> Optional[int]:
_A : Optional[int] = self.get_image_processor()
_A : int = self.get_tokenizer()
_A : List[Any] = BlipaProcessor(tokenizer=_snake_case , image_processor=_snake_case )
_A : Optional[Any] = [[1, 4, 5, 8, 1, 0, 8], [3, 4, 3, 1, 1, 8, 9]]
_A : Union[str, Any] = processor.batch_decode(_snake_case )
_A : List[str] = tokenizer.batch_decode(_snake_case )
self.assertListEqual(_snake_case , _snake_case )
def a__ ( self ) -> Optional[int]:
_A : Dict = self.get_image_processor()
_A : List[Any] = self.get_tokenizer()
_A : Optional[int] = BlipaProcessor(tokenizer=_snake_case , image_processor=_snake_case )
_A : Tuple = 'lower newer'
_A : Optional[int] = self.prepare_image_inputs()
_A : str = processor(text=_snake_case , images=_snake_case )
# For now the processor supports only ['pixel_values', 'input_ids', 'attention_mask']
self.assertListEqual(list(inputs.keys() ) , ["""pixel_values""", """input_ids""", """attention_mask"""] )
| 26 |
import itertools
import random
import unittest
import numpy as np
from transformers import BatchFeature, SpeechTaFeatureExtractor
from transformers.testing_utils import require_torch
from transformers.utils.import_utils import is_torch_available
from ...test_sequence_feature_extraction_common import SequenceFeatureExtractionTestMixin
if is_torch_available():
import torch
a_ :Any = random.Random()
def lowercase_ (A : int , A : Union[str, Any]=1.0 , A : List[str]=None , A : Any=None ):
if rng is None:
snake_case__ : List[str] = global_rng
snake_case__ : int = []
for batch_idx in range(shape[0] ):
values.append([] )
for _ in range(shape[1] ):
values[-1].append(rng.random() * scale )
return values
@require_torch
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
def __init__( self : Optional[Any], _snake_case : List[str], _snake_case : Tuple=7, _snake_case : Union[str, Any]=4_0_0, _snake_case : Any=2_0_0_0, _snake_case : Dict=1, _snake_case : Optional[Any]=0.0, _snake_case : List[Any]=1_6_0_0_0, _snake_case : List[Any]=True, _snake_case : List[Any]=8_0, _snake_case : Dict=1_6, _snake_case : str=6_4, _snake_case : Tuple="hann_window", _snake_case : Union[str, Any]=8_0, _snake_case : Optional[Any]=7_6_0_0, _snake_case : str=1e-10, _snake_case : Any=True, ) ->Union[str, Any]:
snake_case__ : Optional[int] = parent
snake_case__ : Optional[Any] = batch_size
snake_case__ : List[Any] = min_seq_length
snake_case__ : List[Any] = max_seq_length
snake_case__ : Any = (self.max_seq_length - self.min_seq_length) // (self.batch_size - 1)
snake_case__ : Tuple = feature_size
snake_case__ : List[Any] = padding_value
snake_case__ : Any = sampling_rate
snake_case__ : Dict = do_normalize
snake_case__ : Union[str, Any] = num_mel_bins
snake_case__ : Any = hop_length
snake_case__ : Any = win_length
snake_case__ : Any = win_function
snake_case__ : Optional[int] = fmin
snake_case__ : int = fmax
snake_case__ : Union[str, Any] = mel_floor
snake_case__ : Union[str, Any] = return_attention_mask
def lowercase_ ( self : Optional[int] ) ->List[str]:
return {
"feature_size": self.feature_size,
"padding_value": self.padding_value,
"sampling_rate": self.sampling_rate,
"do_normalize": self.do_normalize,
"num_mel_bins": self.num_mel_bins,
"hop_length": self.hop_length,
"win_length": self.win_length,
"win_function": self.win_function,
"fmin": self.fmin,
"fmax": self.fmax,
"mel_floor": self.mel_floor,
"return_attention_mask": self.return_attention_mask,
}
def lowercase_ ( self : Any, _snake_case : Optional[Any]=False, _snake_case : List[str]=False ) ->Union[str, Any]:
def _flatten(_snake_case : List[str] ):
return list(itertools.chain(*_snake_case ) )
if equal_length:
snake_case__ : Any = floats_list((self.batch_size, self.max_seq_length) )
else:
# make sure that inputs increase in size
snake_case__ : int = [
_flatten(floats_list((x, self.feature_size) ) )
for x in range(self.min_seq_length, self.max_seq_length, self.seq_length_diff )
]
if numpify:
snake_case__ : Any = [np.asarray(_snake_case ) for x in speech_inputs]
return speech_inputs
def lowercase_ ( self : Union[str, Any], _snake_case : str=False, _snake_case : Dict=False ) ->List[str]:
if equal_length:
snake_case__ : Optional[Any] = [floats_list((self.max_seq_length, self.num_mel_bins) ) for _ in range(self.batch_size )]
else:
# make sure that inputs increase in size
snake_case__ : List[str] = [
floats_list((x, self.num_mel_bins) )
for x in range(self.min_seq_length, self.max_seq_length, self.seq_length_diff )
]
if numpify:
snake_case__ : int = [np.asarray(_snake_case ) for x in speech_inputs]
return speech_inputs
@require_torch
class snake_case__ ( lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = SpeechTaFeatureExtractor
def lowercase_ ( self : int ) ->Union[str, Any]:
snake_case__ : List[str] = SpeechTaFeatureExtractionTester(self )
def lowercase_ ( self : Any, _snake_case : Dict ) ->Any:
self.assertTrue(np.all(np.mean(_snake_case, axis=0 ) < 1e-3 ) )
self.assertTrue(np.all(np.abs(np.var(_snake_case, axis=0 ) - 1 ) < 1e-3 ) )
def lowercase_ ( self : List[Any] ) ->Union[str, Any]:
# Tests that all call wrap to encode_plus and batch_encode_plus
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
# create three inputs of length 800, 1000, and 1200
snake_case__ : int = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Tuple = [np.asarray(_snake_case ) for speech_input in speech_inputs]
# Test not batched input
snake_case__ : str = feat_extract(speech_inputs[0], return_tensors='np' ).input_values
snake_case__ : List[str] = feat_extract(np_speech_inputs[0], return_tensors='np' ).input_values
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test batched
snake_case__ : Any = feat_extract(_snake_case, return_tensors='np' ).input_values
snake_case__ : Union[str, Any] = feat_extract(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
def lowercase_ ( self : int ) ->Optional[int]:
snake_case__ : List[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Tuple = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : int = ['longest', 'max_length', 'do_not_pad']
snake_case__ : List[str] = [None, 1_6_0_0, None]
for max_length, padding in zip(_snake_case, _snake_case ):
snake_case__ : Optional[int] = feat_extract(_snake_case, padding=_snake_case, max_length=_snake_case, return_tensors='np' )
snake_case__ : Optional[int] = processed.input_values
self._check_zero_mean_unit_variance(input_values[0][:8_0_0] )
self.assertTrue(input_values[0][8_0_0:].sum() < 1e-6 )
self._check_zero_mean_unit_variance(input_values[1][:1_0_0_0] )
self.assertTrue(input_values[0][1_0_0_0:].sum() < 1e-6 )
self._check_zero_mean_unit_variance(input_values[2][:1_2_0_0] )
def lowercase_ ( self : Union[str, Any] ) ->Optional[Any]:
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Tuple = range(8_0_0, 1_4_0_0, 2_0_0 )
snake_case__ : Optional[Any] = [floats_list((1, x) )[0] for x in lengths]
snake_case__ : Union[str, Any] = ['longest', 'max_length', 'do_not_pad']
snake_case__ : str = [None, 1_6_0_0, None]
for max_length, padding in zip(_snake_case, _snake_case ):
snake_case__ : List[str] = feat_extract(_snake_case, max_length=_snake_case, padding=_snake_case )
snake_case__ : Tuple = processed.input_values
self._check_zero_mean_unit_variance(input_values[0][:8_0_0] )
self._check_zero_mean_unit_variance(input_values[1][:1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2][:1_2_0_0] )
def lowercase_ ( self : List[Any] ) ->Optional[Any]:
snake_case__ : Any = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : str = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Optional[Any] = feat_extract(
_snake_case, truncation=_snake_case, max_length=1_0_0_0, padding='max_length', return_tensors='np' )
snake_case__ : int = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1] )
self._check_zero_mean_unit_variance(input_values[2] )
def lowercase_ ( self : int ) ->Union[str, Any]:
snake_case__ : str = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Dict = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : str = feat_extract(
_snake_case, truncation=_snake_case, max_length=1_0_0_0, padding='longest', return_tensors='np' )
snake_case__ : Dict = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1, :1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2] )
# make sure that if max_length < longest -> then pad to max_length
self.assertTrue(input_values.shape == (3, 1_0_0_0) )
snake_case__ : Tuple = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : List[str] = feat_extract(
_snake_case, truncation=_snake_case, max_length=2_0_0_0, padding='longest', return_tensors='np' )
snake_case__ : Optional[Any] = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1, :1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2] )
# make sure that if max_length > longest -> then pad to longest
self.assertTrue(input_values.shape == (3, 1_2_0_0) )
def lowercase_ ( self : List[str] ) ->Dict:
snake_case__ : Dict = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : List[Any] = np.random.rand(1_0_0 ).astype(np.floataa )
snake_case__ : int = np_speech_inputs.tolist()
for inputs in [py_speech_inputs, np_speech_inputs]:
snake_case__ : int = feature_extractor.pad([{'input_values': inputs}], return_tensors='np' )
self.assertTrue(np_processed.input_values.dtype == np.floataa )
snake_case__ : Optional[int] = feature_extractor.pad([{'input_values': inputs}], return_tensors='pt' )
self.assertTrue(pt_processed.input_values.dtype == torch.floataa )
def lowercase_ ( self : Optional[int] ) ->Optional[Any]:
# Tests that all call wrap to encode_plus and batch_encode_plus
snake_case__ : str = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
# create three inputs of length 800, 1000, and 1200
snake_case__ : List[Any] = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Dict = [np.asarray(_snake_case ) for speech_input in speech_inputs]
# Test feature size
snake_case__ : Optional[int] = feature_extractor(audio_target=_snake_case, padding=_snake_case, return_tensors='np' ).input_values
self.assertTrue(input_values.ndim == 3 )
self.assertTrue(input_values.shape[-1] == feature_extractor.num_mel_bins )
# Test not batched input
snake_case__ : Dict = feature_extractor(speech_inputs[0], return_tensors='np' ).input_values
snake_case__ : Any = feature_extractor(np_speech_inputs[0], return_tensors='np' ).input_values
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test batched
snake_case__ : Dict = feature_extractor(_snake_case, return_tensors='np' ).input_values
snake_case__ : Dict = feature_extractor(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test 2-D numpy arrays are batched.
snake_case__ : Optional[Any] = [floats_list((1, x) )[0] for x in (8_0_0, 8_0_0, 8_0_0)]
snake_case__ : int = np.asarray(_snake_case )
snake_case__ : Union[str, Any] = feature_extractor(_snake_case, return_tensors='np' ).input_values
snake_case__ : Union[str, Any] = feature_extractor(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
def lowercase_ ( self : Union[str, Any] ) ->str:
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : List[str] = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : Tuple = BatchFeature({input_name: speech_inputs} )
self.assertTrue(all(len(_snake_case ) == len(_snake_case ) for x, y in zip(_snake_case, processed_features[input_name] ) ) )
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target(equal_length=_snake_case )
snake_case__ : Union[str, Any] = BatchFeature({input_name: speech_inputs}, tensor_type='np' )
snake_case__ : Dict = processed_features[input_name]
if len(batch_features_input.shape ) < 3:
snake_case__ : List[str] = batch_features_input[:, :, None]
self.assertTrue(
batch_features_input.shape
== (self.feat_extract_tester.batch_size, len(speech_inputs[0] ), self.feat_extract_tester.num_mel_bins) )
@require_torch
def lowercase_ ( self : List[str] ) ->Any:
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target(equal_length=_snake_case )
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Tuple = feat_extract.model_input_names[0]
snake_case__ : List[Any] = BatchFeature({input_name: speech_inputs}, tensor_type='pt' )
snake_case__ : Tuple = processed_features[input_name]
if len(batch_features_input.shape ) < 3:
snake_case__ : Any = batch_features_input[:, :, None]
self.assertTrue(
batch_features_input.shape
== (self.feat_extract_tester.batch_size, len(speech_inputs[0] ), self.feat_extract_tester.num_mel_bins) )
@require_torch
def lowercase_ ( self : Optional[int] ) ->Tuple:
snake_case__ : Dict = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Union[str, Any] = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : List[str] = BatchFeature({input_name: speech_inputs} )
snake_case__ : int = feat_extract.num_mel_bins # hack!
snake_case__ : Tuple = feat_extract.pad(_snake_case, padding='longest', return_tensors='np' )[input_name]
snake_case__ : Union[str, Any] = feat_extract.pad(_snake_case, padding='longest', return_tensors='pt' )[input_name]
self.assertTrue(abs(input_np.astype(np.floataa ).sum() - input_pt.numpy().astype(np.floataa ).sum() ) < 1e-2 )
def lowercase_ ( self : int ) ->Any:
snake_case__ : Any = self.feat_extract_dict
snake_case__ : List[Any] = True
snake_case__ : Union[str, Any] = self.feature_extraction_class(**_snake_case )
snake_case__ : Any = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : List[Any] = [len(_snake_case ) for x in speech_inputs]
snake_case__ : Union[str, Any] = feat_extract.model_input_names[0]
snake_case__ : Optional[int] = BatchFeature({input_name: speech_inputs} )
snake_case__ : List[str] = feat_extract.num_mel_bins # hack!
snake_case__ : str = feat_extract.pad(_snake_case, padding='longest', return_tensors='np' )
self.assertIn('attention_mask', _snake_case )
self.assertListEqual(list(processed.attention_mask.shape ), list(processed[input_name].shape[:2] ) )
self.assertListEqual(processed.attention_mask.sum(-1 ).tolist(), _snake_case )
def lowercase_ ( self : Optional[int] ) ->str:
snake_case__ : int = self.feat_extract_dict
snake_case__ : List[str] = True
snake_case__ : Tuple = self.feature_extraction_class(**_snake_case )
snake_case__ : List[str] = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : str = [len(_snake_case ) for x in speech_inputs]
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : Optional[int] = BatchFeature({input_name: speech_inputs} )
snake_case__ : Optional[Any] = min(_snake_case )
snake_case__ : Union[str, Any] = feat_extract.num_mel_bins # hack!
snake_case__ : Tuple = feat_extract.pad(
_snake_case, padding='max_length', max_length=_snake_case, truncation=_snake_case, return_tensors='np' )
self.assertIn('attention_mask', _snake_case )
self.assertListEqual(
list(processed_pad.attention_mask.shape ), [processed_pad[input_name].shape[0], max_length] )
self.assertListEqual(
processed_pad.attention_mask[:, :max_length].sum(-1 ).tolist(), [max_length for x in speech_inputs] )
def lowercase_ ( self : List[Any], _snake_case : Optional[int] ) ->Optional[Any]:
from datasets import load_dataset
snake_case__ : str = load_dataset('hf-internal-testing/librispeech_asr_dummy', 'clean', split='validation' )
# automatic decoding with librispeech
snake_case__ : Dict = ds.sort('id' ).select(range(_snake_case ) )[:num_samples]['audio']
return [x["array"] for x in speech_samples]
def lowercase_ ( self : str ) ->str:
# fmt: off
snake_case__ : List[Any] = torch.tensor(
[2.3804e-03, 2.0752e-03, 1.9836e-03, 2.1057e-03, 1.6174e-03,
3.0518e-04, 9.1553e-05, 3.3569e-04, 9.7656e-04, 1.8311e-03,
2.0142e-03, 2.1057e-03, 1.7395e-03, 4.5776e-04, -3.9673e-04,
4.5776e-04, 1.0071e-03, 9.1553e-05, 4.8828e-04, 1.1597e-03,
7.3242e-04, 9.4604e-04, 1.8005e-03, 1.8311e-03, 8.8501e-04,
4.2725e-04, 4.8828e-04, 7.3242e-04, 1.0986e-03, 2.1057e-03] )
# fmt: on
snake_case__ : Union[str, Any] = self._load_datasamples(1 )
snake_case__ : Optional[int] = SpeechTaFeatureExtractor()
snake_case__ : List[Any] = feature_extractor(_snake_case, return_tensors='pt' ).input_values
self.assertEquals(input_values.shape, (1, 9_3_6_8_0) )
self.assertTrue(torch.allclose(input_values[0, :3_0], _snake_case, atol=1e-6 ) )
def lowercase_ ( self : Any ) ->str:
# fmt: off
snake_case__ : Optional[Any] = torch.tensor(
[-2.6_8_7_0, -3.0_1_0_4, -3.1_3_5_6, -3.5_3_5_2, -3.0_0_4_4, -3.0_3_5_3, -3.4_7_1_9, -3.6_7_7_7,
-3.1_5_2_0, -2.9_4_3_5, -2.6_5_5_3, -2.8_7_9_5, -2.9_9_4_4, -2.5_9_2_1, -3.0_2_7_9, -3.0_3_8_6,
-3.0_8_6_4, -3.1_2_9_1, -3.2_3_5_3, -2.7_4_4_4, -2.6_8_3_1, -2.7_2_8_7, -3.1_7_6_1, -3.1_5_7_1,
-3.2_7_2_6, -3.0_5_8_2, -3.1_0_0_7, -3.4_5_3_3, -3.4_6_9_5, -3.0_9_9_8] )
# fmt: on
snake_case__ : List[str] = self._load_datasamples(1 )
snake_case__ : str = SpeechTaFeatureExtractor()
snake_case__ : Optional[Any] = feature_extractor(audio_target=_snake_case, return_tensors='pt' ).input_values
self.assertEquals(input_values.shape, (1, 3_6_6, 8_0) )
self.assertTrue(torch.allclose(input_values[0, 0, :3_0], _snake_case, atol=1e-4 ) )
| 277 | 0 |
import gc
import random
import unittest
import numpy as np
import torch
from PIL import Image
from transformers import XLMRobertaTokenizerFast
from diffusers import DDIMScheduler, KandinskyImgaImgPipeline, KandinskyPriorPipeline, UNetaDConditionModel, VQModel
from diffusers.pipelines.kandinsky.text_encoder import MCLIPConfig, MultilingualCLIP
from diffusers.utils import floats_tensor, load_image, load_numpy, slow, torch_device
from diffusers.utils.testing_utils import enable_full_determinism, require_torch_gpu
from ..test_pipelines_common import PipelineTesterMixin, assert_mean_pixel_difference
enable_full_determinism()
class __lowerCAmelCase ( lowerCAmelCase_ , unittest.TestCase ):
UpperCamelCase = KandinskyImgaImgPipeline
UpperCamelCase = ['''prompt''', '''image_embeds''', '''negative_image_embeds''', '''image''']
UpperCamelCase = [
'''prompt''',
'''negative_prompt''',
'''image_embeds''',
'''negative_image_embeds''',
'''image''',
]
UpperCamelCase = [
'''generator''',
'''height''',
'''width''',
'''strength''',
'''guidance_scale''',
'''negative_prompt''',
'''num_inference_steps''',
'''return_dict''',
'''guidance_scale''',
'''num_images_per_prompt''',
'''output_type''',
'''return_dict''',
]
UpperCamelCase = False
@property
def _lowerCamelCase ( self : Dict) -> Optional[Any]:
"""simple docstring"""
return 32
@property
def _lowerCamelCase ( self : Dict) -> int:
"""simple docstring"""
return 32
@property
def _lowerCamelCase ( self : List[str]) -> Tuple:
"""simple docstring"""
return self.time_input_dim
@property
def _lowerCamelCase ( self : str) -> Optional[Any]:
"""simple docstring"""
return self.time_input_dim * 4
@property
def _lowerCamelCase ( self : str) -> Optional[Any]:
"""simple docstring"""
return 1_00
@property
def _lowerCamelCase ( self : Optional[Any]) -> List[Any]:
"""simple docstring"""
_UpperCAmelCase = XLMRobertaTokenizerFast.from_pretrained('YiYiXu/tiny-random-mclip-base')
return tokenizer
@property
def _lowerCamelCase ( self : Optional[int]) -> Tuple:
"""simple docstring"""
torch.manual_seed(0)
_UpperCAmelCase = MCLIPConfig(
numDims=self.cross_attention_dim , transformerDimensions=self.text_embedder_hidden_size , hidden_size=self.text_embedder_hidden_size , intermediate_size=37 , num_attention_heads=4 , num_hidden_layers=5 , vocab_size=10_05 , )
_UpperCAmelCase = MultilingualCLIP(_snake_case)
_UpperCAmelCase = text_encoder.eval()
return text_encoder
@property
def _lowerCamelCase ( self : Optional[int]) -> Dict:
"""simple docstring"""
torch.manual_seed(0)
_UpperCAmelCase = {
'in_channels': 4,
# Out channels is double in channels because predicts mean and variance
'out_channels': 8,
'addition_embed_type': 'text_image',
'down_block_types': ('ResnetDownsampleBlock2D', 'SimpleCrossAttnDownBlock2D'),
'up_block_types': ('SimpleCrossAttnUpBlock2D', 'ResnetUpsampleBlock2D'),
'mid_block_type': 'UNetMidBlock2DSimpleCrossAttn',
'block_out_channels': (self.block_out_channels_a, self.block_out_channels_a * 2),
'layers_per_block': 1,
'encoder_hid_dim': self.text_embedder_hidden_size,
'encoder_hid_dim_type': 'text_image_proj',
'cross_attention_dim': self.cross_attention_dim,
'attention_head_dim': 4,
'resnet_time_scale_shift': 'scale_shift',
'class_embed_type': None,
}
_UpperCAmelCase = UNetaDConditionModel(**_snake_case)
return model
@property
def _lowerCamelCase ( self : Optional[Any]) -> int:
"""simple docstring"""
return {
"block_out_channels": [32, 64],
"down_block_types": ["DownEncoderBlock2D", "AttnDownEncoderBlock2D"],
"in_channels": 3,
"latent_channels": 4,
"layers_per_block": 1,
"norm_num_groups": 8,
"norm_type": "spatial",
"num_vq_embeddings": 12,
"out_channels": 3,
"up_block_types": [
"AttnUpDecoderBlock2D",
"UpDecoderBlock2D",
],
"vq_embed_dim": 4,
}
@property
def _lowerCamelCase ( self : Tuple) -> List[str]:
"""simple docstring"""
torch.manual_seed(0)
_UpperCAmelCase = VQModel(**self.dummy_movq_kwargs)
return model
def _lowerCamelCase ( self : Union[str, Any]) -> Optional[int]:
"""simple docstring"""
_UpperCAmelCase = self.dummy_text_encoder
_UpperCAmelCase = self.dummy_tokenizer
_UpperCAmelCase = self.dummy_unet
_UpperCAmelCase = self.dummy_movq
_UpperCAmelCase = {
'num_train_timesteps': 10_00,
'beta_schedule': 'linear',
'beta_start': 0.0_0_0_8_5,
'beta_end': 0.0_1_2,
'clip_sample': False,
'set_alpha_to_one': False,
'steps_offset': 0,
'prediction_type': 'epsilon',
'thresholding': False,
}
_UpperCAmelCase = DDIMScheduler(**_snake_case)
_UpperCAmelCase = {
'text_encoder': text_encoder,
'tokenizer': tokenizer,
'unet': unet,
'scheduler': scheduler,
'movq': movq,
}
return components
def _lowerCamelCase ( self : int , A : Any , A : Dict=0) -> Dict:
"""simple docstring"""
_UpperCAmelCase = floats_tensor((1, self.cross_attention_dim) , rng=random.Random(_snake_case)).to(_snake_case)
_UpperCAmelCase = floats_tensor((1, self.cross_attention_dim) , rng=random.Random(seed + 1)).to(_snake_case)
# create init_image
_UpperCAmelCase = floats_tensor((1, 3, 64, 64) , rng=random.Random(_snake_case)).to(_snake_case)
_UpperCAmelCase = image.cpu().permute(0 , 2 , 3 , 1)[0]
_UpperCAmelCase = Image.fromarray(np.uinta(_snake_case)).convert('RGB').resize((2_56, 2_56))
if str(_snake_case).startswith('mps'):
_UpperCAmelCase = torch.manual_seed(_snake_case)
else:
_UpperCAmelCase = torch.Generator(device=_snake_case).manual_seed(_snake_case)
_UpperCAmelCase = {
'prompt': 'horse',
'image': init_image,
'image_embeds': image_embeds,
'negative_image_embeds': negative_image_embeds,
'generator': generator,
'height': 64,
'width': 64,
'num_inference_steps': 10,
'guidance_scale': 7.0,
'strength': 0.2,
'output_type': 'np',
}
return inputs
def _lowerCamelCase ( self : Optional[int]) -> Dict:
"""simple docstring"""
_UpperCAmelCase = 'cpu'
_UpperCAmelCase = self.get_dummy_components()
_UpperCAmelCase = self.pipeline_class(**_snake_case)
_UpperCAmelCase = pipe.to(_snake_case)
pipe.set_progress_bar_config(disable=_snake_case)
_UpperCAmelCase = pipe(**self.get_dummy_inputs(_snake_case))
_UpperCAmelCase = output.images
_UpperCAmelCase = pipe(
**self.get_dummy_inputs(_snake_case) , return_dict=_snake_case , )[0]
_UpperCAmelCase = image[0, -3:, -3:, -1]
_UpperCAmelCase = image_from_tuple[0, -3:, -3:, -1]
assert image.shape == (1, 64, 64, 3)
_UpperCAmelCase = np.array(
[0.6_1_4_7_4_9_4_3, 0.6_0_7_3_5_3_9, 0.4_3_3_0_8_5_4_4, 0.5_9_2_8_2_6_9, 0.4_7_4_9_3_5_9_5, 0.4_6_7_5_5_9_7_3, 0.4_6_1_3_8_3_8, 0.4_5_3_6_8_7_9_7, 0.5_0_1_1_9_2_3_3])
assert (
np.abs(image_slice.flatten() - expected_slice).max() < 1E-2
), F" expected_slice {expected_slice}, but got {image_slice.flatten()}"
assert (
np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 1E-2
), F" expected_slice {expected_slice}, but got {image_from_tuple_slice.flatten()}"
@slow
@require_torch_gpu
class __lowerCAmelCase ( unittest.TestCase ):
def _lowerCamelCase ( self : List[Any]) -> Any:
"""simple docstring"""
super().tearDown()
gc.collect()
torch.cuda.empty_cache()
def _lowerCamelCase ( self : Dict) -> List[str]:
"""simple docstring"""
_UpperCAmelCase = load_numpy(
'https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main'
'/kandinsky/kandinsky_img2img_frog.npy')
_UpperCAmelCase = load_image(
'https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main' '/kandinsky/cat.png')
_UpperCAmelCase = 'A red cartoon frog, 4k'
_UpperCAmelCase = KandinskyPriorPipeline.from_pretrained(
'kandinsky-community/kandinsky-2-1-prior' , torch_dtype=torch.floataa)
pipe_prior.to(_snake_case)
_UpperCAmelCase = KandinskyImgaImgPipeline.from_pretrained(
'kandinsky-community/kandinsky-2-1' , torch_dtype=torch.floataa)
_UpperCAmelCase = pipeline.to(_snake_case)
pipeline.set_progress_bar_config(disable=_snake_case)
_UpperCAmelCase = torch.Generator(device='cpu').manual_seed(0)
_UpperCAmelCase = pipe_prior(
_snake_case , generator=_snake_case , num_inference_steps=5 , negative_prompt='' , ).to_tuple()
_UpperCAmelCase = pipeline(
_snake_case , image=_snake_case , image_embeds=_snake_case , negative_image_embeds=_snake_case , generator=_snake_case , num_inference_steps=1_00 , height=7_68 , width=7_68 , strength=0.2 , output_type='np' , )
_UpperCAmelCase = output.images[0]
assert image.shape == (7_68, 7_68, 3)
assert_mean_pixel_difference(_snake_case , _snake_case)
| 339 |
# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from ..models.auto import AutoModelForSeqaSeqLM, AutoTokenizer
from .base import PipelineTool
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """philschmid/bart-large-cnn-samsum"""
_SCREAMING_SNAKE_CASE = (
"""This is a tool that summarizes an English text. It takes an input `text` containing the text to summarize, """
"""and returns a summary of the text."""
)
_SCREAMING_SNAKE_CASE = """summarizer"""
_SCREAMING_SNAKE_CASE = AutoTokenizer
_SCREAMING_SNAKE_CASE = AutoModelForSeqaSeqLM
_SCREAMING_SNAKE_CASE = ["""text"""]
_SCREAMING_SNAKE_CASE = ["""text"""]
def lowercase_ ( self : Optional[Any], _snake_case : str ) ->Any:
return self.pre_processor(_snake_case, return_tensors='pt', truncation=_snake_case )
def lowercase_ ( self : int, _snake_case : List[Any] ) ->Any:
return self.model.generate(**_snake_case )[0]
def lowercase_ ( self : int, _snake_case : int ) ->str:
return self.pre_processor.decode(_snake_case, skip_special_tokens=_snake_case, clean_up_tokenization_spaces=_snake_case )
| 277 | 0 |
import argparse
import json
import os
import torch
from transformers import LukeConfig, LukeModel, LukeTokenizer, RobertaTokenizer
from transformers.tokenization_utils_base import AddedToken
@torch.no_grad()
def __UpperCAmelCase ( a_ , a_ , a_ , a_ , a_):
# Load configuration defined in the metadata file
with open(a_) as metadata_file:
snake_case_ = json.load(a_)
snake_case_ = LukeConfig(use_entity_aware_attention=a_ , **metadata['model_config'])
# Load in the weights from the checkpoint_path
snake_case_ = torch.load(a_ , map_location='cpu')
# Load the entity vocab file
snake_case_ = load_entity_vocab(a_)
snake_case_ = RobertaTokenizer.from_pretrained(metadata['model_config']['bert_model_name'])
# Add special tokens to the token vocabulary for downstream tasks
snake_case_ = AddedToken('<ent>' , lstrip=a_ , rstrip=a_)
snake_case_ = AddedToken('<ent2>' , lstrip=a_ , rstrip=a_)
tokenizer.add_special_tokens({'additional_special_tokens': [entity_token_a, entity_token_a]})
config.vocab_size += 2
print(f'''Saving tokenizer to {pytorch_dump_folder_path}''')
tokenizer.save_pretrained(a_)
with open(os.path.join(a_ , LukeTokenizer.vocab_files_names['entity_vocab_file']) , 'w') as f:
json.dump(a_ , a_)
snake_case_ = LukeTokenizer.from_pretrained(a_)
# Initialize the embeddings of the special tokens
snake_case_ = state_dict['embeddings.word_embeddings.weight']
snake_case_ = word_emb[tokenizer.convert_tokens_to_ids(['@'])[0]].unsqueeze(0)
snake_case_ = word_emb[tokenizer.convert_tokens_to_ids(['#'])[0]].unsqueeze(0)
snake_case_ = torch.cat([word_emb, ent_emb, enta_emb])
# Initialize the query layers of the entity-aware self-attention mechanism
for layer_index in range(config.num_hidden_layers):
for matrix_name in ["query.weight", "query.bias"]:
snake_case_ = f'''encoder.layer.{layer_index}.attention.self.'''
snake_case_ = state_dict[prefix + matrix_name]
snake_case_ = state_dict[prefix + matrix_name]
snake_case_ = state_dict[prefix + matrix_name]
# Initialize the embedding of the [MASK2] entity using that of the [MASK] entity for downstream tasks
snake_case_ = state_dict['entity_embeddings.entity_embeddings.weight']
snake_case_ = entity_emb[entity_vocab['[MASK]']]
snake_case_ = LukeModel(config=a_).eval()
snake_case_ = model.load_state_dict(a_ , strict=a_)
if not (len(a_) == 1 and missing_keys[0] == "embeddings.position_ids"):
raise ValueError(f'''Missing keys {", ".join(a_)}. Expected only missing embeddings.position_ids''')
if not (all(key.startswith('entity_predictions') or key.startswith('lm_head') for key in unexpected_keys)):
raise ValueError(
'Unexpected keys'
f''' {", ".join([key for key in unexpected_keys if not (key.startswith("entity_predictions") or key.startswith("lm_head"))])}''')
# Check outputs
snake_case_ = LukeTokenizer.from_pretrained(a_ , task='entity_classification')
snake_case_ = (
'Top seed Ana Ivanovic said on Thursday she could hardly believe her luck as a fortuitous netcord helped the'
' new world number one avoid a humiliating second- round exit at Wimbledon .'
)
snake_case_ = (39, 42)
snake_case_ = tokenizer(a_ , entity_spans=[span] , add_prefix_space=a_ , return_tensors='pt')
snake_case_ = model(**a_)
# Verify word hidden states
if model_size == "large":
snake_case_ = torch.Size((1, 42, 10_24))
snake_case_ = torch.tensor(
[[0.01_33, 0.08_65, 0.00_95], [0.30_93, -0.25_76, -0.74_18], [-0.17_20, -0.21_17, -0.28_69]])
else: # base
snake_case_ = torch.Size((1, 42, 7_68))
snake_case_ = torch.tensor([[0.00_37, 0.13_68, -0.00_91], [0.10_99, 0.33_29, -0.10_95], [0.07_65, 0.53_35, 0.11_79]])
if not (outputs.last_hidden_state.shape == expected_shape):
raise ValueError(
f'''Outputs.last_hidden_state.shape is {outputs.last_hidden_state.shape}, Expected shape is {expected_shape}''')
if not torch.allclose(outputs.last_hidden_state[0, :3, :3] , a_ , atol=1E-4):
raise ValueError
# Verify entity hidden states
if model_size == "large":
snake_case_ = torch.Size((1, 1, 10_24))
snake_case_ = torch.tensor([[0.04_66, -0.01_06, -0.01_79]])
else: # base
snake_case_ = torch.Size((1, 1, 7_68))
snake_case_ = torch.tensor([[0.14_57, 0.10_44, 0.01_74]])
if not (outputs.entity_last_hidden_state.shape != expected_shape):
raise ValueError(
f'''Outputs.entity_last_hidden_state.shape is {outputs.entity_last_hidden_state.shape}, Expected shape is'''
f''' {expected_shape}''')
if not torch.allclose(outputs.entity_last_hidden_state[0, :3, :3] , a_ , atol=1E-4):
raise ValueError
# Finally, save our PyTorch model and tokenizer
print('Saving PyTorch model to {}'.format(a_))
model.save_pretrained(a_)
def __UpperCAmelCase ( a_):
snake_case_ = {}
with open(a_ , 'r' , encoding='utf-8') as f:
for index, line in enumerate(a_):
snake_case_ = line.rstrip().split('\t')
snake_case_ = index
return entity_vocab
if __name__ == "__main__":
lowercase = argparse.ArgumentParser()
# Required parameters
parser.add_argument("--checkpoint_path", type=str, help="Path to a pytorch_model.bin file.")
parser.add_argument(
"--metadata_path", default=None, type=str, help="Path to a metadata.json file, defining the configuration."
)
parser.add_argument(
"--entity_vocab_path",
default=None,
type=str,
help="Path to an entity_vocab.tsv file, containing the entity vocabulary.",
)
parser.add_argument(
"--pytorch_dump_folder_path", default=None, type=str, help="Path to where to dump the output PyTorch model."
)
parser.add_argument(
"--model_size", default="base", type=str, choices=["base", "large"], help="Size of the model to be converted."
)
lowercase = parser.parse_args()
convert_luke_checkpoint(
args.checkpoint_path,
args.metadata_path,
args.entity_vocab_path,
args.pytorch_dump_folder_path,
args.model_size,
)
| 178 |
import argparse
import torch
from transformers import LxmertConfig, LxmertForPreTraining, load_tf_weights_in_lxmert
from transformers.utils import logging
logging.set_verbosity_info()
def lowercase_ (A : str , A : List[Any] , A : Any ):
# Initialise PyTorch model
snake_case__ : List[Any] = LxmertConfig.from_json_file(A )
print(F'''Building PyTorch model from configuration: {config}''' )
snake_case__ : List[str] = LxmertForPreTraining(A )
# Load weights from tf checkpoint
load_tf_weights_in_lxmert(A , A , A )
# Save pytorch-model
print(F'''Save PyTorch model to {pytorch_dump_path}''' )
torch.save(model.state_dict() , A )
if __name__ == "__main__":
a_ :Union[str, Any] = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--tf_checkpoint_path", default=None, type=str, required=True, help="Path to the TensorFlow checkpoint path."
)
parser.add_argument(
"--config_file",
default=None,
type=str,
required=True,
help="The config json file corresponding to the pre-trained model. \nThis specifies the model architecture.",
)
parser.add_argument(
"--pytorch_dump_path", default=None, type=str, required=True, help="Path to the output PyTorch model."
)
a_ :Optional[int] = parser.parse_args()
convert_tf_checkpoint_to_pytorch(args.tf_checkpoint_path, args.config_file, args.pytorch_dump_path)
| 277 | 0 |
"""simple docstring"""
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_tf_available,
is_tokenizers_available,
is_torch_available,
)
__UpperCamelCase = {
"configuration_funnel": ["FUNNEL_PRETRAINED_CONFIG_ARCHIVE_MAP", "FunnelConfig"],
"convert_funnel_original_tf_checkpoint_to_pytorch": [],
"tokenization_funnel": ["FunnelTokenizer"],
}
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__UpperCamelCase = ["FunnelTokenizerFast"]
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__UpperCamelCase = [
"FUNNEL_PRETRAINED_MODEL_ARCHIVE_LIST",
"FunnelBaseModel",
"FunnelForMaskedLM",
"FunnelForMultipleChoice",
"FunnelForPreTraining",
"FunnelForQuestionAnswering",
"FunnelForSequenceClassification",
"FunnelForTokenClassification",
"FunnelModel",
"FunnelPreTrainedModel",
"load_tf_weights_in_funnel",
]
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__UpperCamelCase = [
"TF_FUNNEL_PRETRAINED_MODEL_ARCHIVE_LIST",
"TFFunnelBaseModel",
"TFFunnelForMaskedLM",
"TFFunnelForMultipleChoice",
"TFFunnelForPreTraining",
"TFFunnelForQuestionAnswering",
"TFFunnelForSequenceClassification",
"TFFunnelForTokenClassification",
"TFFunnelModel",
"TFFunnelPreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_funnel import FUNNEL_PRETRAINED_CONFIG_ARCHIVE_MAP, FunnelConfig
from .tokenization_funnel import FunnelTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_funnel_fast import FunnelTokenizerFast
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_funnel import (
FUNNEL_PRETRAINED_MODEL_ARCHIVE_LIST,
FunnelBaseModel,
FunnelForMaskedLM,
FunnelForMultipleChoice,
FunnelForPreTraining,
FunnelForQuestionAnswering,
FunnelForSequenceClassification,
FunnelForTokenClassification,
FunnelModel,
FunnelPreTrainedModel,
load_tf_weights_in_funnel,
)
try:
if not is_tf_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_tf_funnel import (
TF_FUNNEL_PRETRAINED_MODEL_ARCHIVE_LIST,
TFFunnelBaseModel,
TFFunnelForMaskedLM,
TFFunnelForMultipleChoice,
TFFunnelForPreTraining,
TFFunnelForQuestionAnswering,
TFFunnelForSequenceClassification,
TFFunnelForTokenClassification,
TFFunnelModel,
TFFunnelPreTrainedModel,
)
else:
import sys
__UpperCamelCase = _LazyModule(__name__, globals()['''__file__'''], _import_structure, module_spec=__spec__)
| 113 |
import argparse
import fairseq
import torch
from torch import nn
from transformers import (
MBartaaTokenizer,
MBartConfig,
MBartForCausalLM,
SpeechEncoderDecoderConfig,
SpeechEncoderDecoderModel,
WavaVecaConfig,
WavaVecaFeatureExtractor,
WavaVecaModel,
logging,
)
logging.set_verbosity_info()
a_ :Tuple = logging.get_logger(__name__)
a_ :List[Any] = {
"post_extract_proj": "feature_projection.projection",
"encoder.pos_conv.0": "encoder.pos_conv_embed.conv",
"self_attn.k_proj": "encoder.layers.*.attention.k_proj",
"self_attn.v_proj": "encoder.layers.*.attention.v_proj",
"self_attn.q_proj": "encoder.layers.*.attention.q_proj",
"self_attn.out_proj": "encoder.layers.*.attention.out_proj",
"self_attn_layer_norm": "encoder.layers.*.layer_norm",
"fc1": "encoder.layers.*.feed_forward.intermediate_dense",
"fc2": "encoder.layers.*.feed_forward.output_dense",
"final_layer_norm": "encoder.layers.*.final_layer_norm",
"encoder.layer_norm": "encoder.layer_norm",
"w2v_model.layer_norm": "feature_projection.layer_norm",
"quantizer.weight_proj": "quantizer.weight_proj",
"quantizer.vars": "quantizer.codevectors",
"project_q": "project_q",
"final_proj": "project_hid",
"w2v_encoder.proj": "lm_head",
"mask_emb": "masked_spec_embed",
}
a_ :Optional[int] = [
"lm_head",
"quantizer.weight_proj",
"quantizer.codevectors",
"project_q",
"project_hid",
]
def lowercase_ (A : Union[str, Any] , A : str , A : Dict , A : Optional[Any] , A : Optional[Any] ):
for attribute in key.split('.' ):
snake_case__ : Any = getattr(A , A )
if weight_type is not None:
snake_case__ : Optional[Any] = getattr(A , A ).shape
else:
snake_case__ : Optional[int] = hf_pointer.shape
assert hf_shape == value.shape, (
F'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
F''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
snake_case__ : Tuple = value
elif weight_type == "weight_g":
snake_case__ : Tuple = value
elif weight_type == "weight_v":
snake_case__ : List[Any] = value
elif weight_type == "bias":
snake_case__ : List[Any] = value
else:
snake_case__ : Optional[Any] = value
logger.info(F'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def lowercase_ (A : str , A : Any ):
snake_case__ : Union[str, Any] = []
snake_case__ : Union[str, Any] = fairseq_model.state_dict()
snake_case__ : Union[str, Any] = hf_model.feature_extractor
snake_case__ : Any = hf_model.adapter
for name, value in fairseq_dict.items():
snake_case__ : Any = False
if "conv_layers" in name:
load_conv_layer(
A , A , A , A , hf_model.config.feat_extract_norm == 'group' , )
snake_case__ : List[Any] = True
elif any(x in name for x in ['adaptor', 'w2v_encoder.proj.', 'w2v_proj_ln.'] ):
load_adapter(A , A , A , A )
snake_case__ : Optional[Any] = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split('w2v_model.' )[-1] == name.split('.' )[0]:
snake_case__ : Tuple = True
if "*" in mapped_key:
snake_case__ : List[Any] = name.split(A )[0].split('.' )[-2]
snake_case__ : Optional[int] = mapped_key.replace('*' , A )
if "weight_g" in name:
snake_case__ : Optional[int] = 'weight_g'
elif "weight_v" in name:
snake_case__ : Optional[Any] = 'weight_v'
elif "bias" in name:
snake_case__ : Union[str, Any] = 'bias'
elif "weight" in name:
snake_case__ : Optional[int] = 'weight'
else:
snake_case__ : Tuple = None
set_recursively(A , A , A , A , A )
continue
if not is_used:
unused_weights.append(A )
logger.warning(F'''Unused weights: {unused_weights}''' )
def lowercase_ (A : Union[str, Any] , A : Any , A : str , A : str , A : int ):
snake_case__ : str = full_name.split('conv_layers.' )[-1]
snake_case__ : Optional[int] = name.split('.' )
snake_case__ : Tuple = int(items[0] )
snake_case__ : Any = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
snake_case__ : Union[str, Any] = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
snake_case__ : Union[str, Any] = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
F'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
snake_case__ : Optional[int] = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
snake_case__ : Optional[Any] = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(A )
def lowercase_ (A : Optional[Any] , A : Any , A : Tuple , A : Any ):
snake_case__ : List[str] = full_name.split('adaptor.' )[-1]
snake_case__ : Tuple = name.split('.' )
if items[1].isdigit():
snake_case__ : Optional[int] = int(items[1] )
else:
snake_case__ : Any = None
if "adaptor" not in full_name:
if "proj_ln" in full_name:
# has to be layer norm
if "bias" in name:
assert (
value.shape == adapter.proj_layer_norm.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj_layer_norm.bias.data.shape} was found.'''
snake_case__ : List[Any] = value
logger.info(F'''Adapter proj layer norm bias was initialized from {full_name}.''' )
if "weight" in name:
assert (
value.shape == adapter.proj_layer_norm.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj_layer_norm.weight.data.shape} was found.'''
snake_case__ : int = value
else:
# has to be projection layer
if "bias" in name:
assert (
value.shape == adapter.proj.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj.bias.data.shape} was found.'''
snake_case__ : str = value
logger.info(F'''Adapter proj layer bias was initialized from {full_name}.''' )
if "weight" in name:
assert (
value.shape == adapter.proj.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj.weight.data.shape} was found.'''
snake_case__ : Dict = value
logger.info(F'''Adapter proj layer weight was initialized from {full_name}.''' )
elif isinstance(A , A ):
if "bias" in name:
assert (
value.shape == adapter.layers[layer_id].conv.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.layers[layer_id].conv.bias.data.shape} was found.'''
snake_case__ : List[str] = value
logger.info(F'''Adapter layer {layer_id} bias was initialized from {full_name}.''' )
elif "weight" in name:
assert (
value.shape == adapter.layers[layer_id].conv.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.layers[layer_id].conv.weight.data.shape} was found.'''
snake_case__ : List[str] = value
logger.info(F'''Adapter layer {layer_id} bias was initialized from {full_name}.''' )
else:
unused_weights.append(A )
def lowercase_ (A : int ):
snake_case__ , snake_case__ : Union[str, Any] = emb.weight.shape
snake_case__ : int = nn.Linear(A , A , bias=A )
snake_case__ : Optional[Any] = emb.weight.data
return lin_layer
@torch.no_grad()
def lowercase_ (A : Tuple , A : Tuple , A : Any , A : Optional[Any] , A : int , A : Optional[Any] , A : Union[str, Any] , A : Union[str, Any] , A : Optional[Any] , A : List[Any] , A : Union[str, Any] , ):
snake_case__ : Optional[Any] = WavaVecaConfig.from_pretrained(
A , add_adapter=A , adapter_stride=A , adapter_kernel_size=A , use_auth_token=A , output_hidden_size=A , )
snake_case__ : Dict = MBartConfig.from_pretrained(A )
# load model
snake_case__ , snake_case__ , snake_case__ : Any = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] , arg_overrides={
'config_yaml': config_yaml_path,
'data': '/'.join(dict_path.split('/' )[:-1] ),
'w2v_path': checkpoint_path,
'load_pretrained_decoder_from': None,
} , )
snake_case__ : List[Any] = model[0].eval()
# load feature extractor
snake_case__ : str = WavaVecaFeatureExtractor.from_pretrained(A , use_auth_token=A )
# set weights for wav2vec2 encoder
snake_case__ : List[str] = WavaVecaModel(A )
recursively_load_weights_wavaveca(model.encoder , A )
# load decoder weights
snake_case__ : Any = MBartForCausalLM(A )
snake_case__ , snake_case__ : int = hf_decoder.model.decoder.load_state_dict(model.decoder.state_dict() , strict=A )
logger.warning(F'''The following keys are missing when loading the decoder weights: {missing_keys}''' )
logger.warning(F'''The following keys are unexpected when loading the decoder weights: {unexpected_keys}''' )
snake_case__ : Union[str, Any] = SpeechEncoderDecoderModel(encoder=A , decoder=A )
snake_case__ : str = False
snake_case__ : int = MBartaaTokenizer(A )
tokenizer.save_pretrained(A )
snake_case__ : Any = hf_wavavec.config.to_dict()
snake_case__ : Tuple = tokenizer.pad_token_id
snake_case__ : Union[str, Any] = tokenizer.bos_token_id
snake_case__ : Dict = tokenizer.eos_token_id
snake_case__ : Optional[int] = 'mbart50'
snake_case__ : Union[str, Any] = 'wav2vec2'
snake_case__ : List[str] = tokenizer.eos_token_id
snake_case__ : Union[str, Any] = 2_5_0_0_0_4
snake_case__ : int = tokenizer.eos_token_id
snake_case__ : Union[str, Any] = SpeechEncoderDecoderConfig.from_dict(A )
hf_wavavec.save_pretrained(A )
feature_extractor.save_pretrained(A )
if __name__ == "__main__":
a_ :str = argparse.ArgumentParser()
parser.add_argument("--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model.")
parser.add_argument("--checkpoint_path", default=None, type=str, help="Path to fairseq checkpoint")
parser.add_argument("--dict_path", default=None, type=str, help="Path to dict of fine-tuned model")
parser.add_argument("--config_yaml_path", default=None, type=str, help="Path to yaml file of fine-tuned model")
parser.add_argument(
"--encoder_config_path",
default="facebook/wav2vec2-xls-r-1b",
type=str,
help="Path to hf encoder wav2vec2 checkpoint config",
)
parser.add_argument(
"--decoder_config_path",
default="facebook/mbart-large-50-one-to-many-mmt",
type=str,
help="Path to hf decoder checkpoint config",
)
parser.add_argument("--add_adapter", default=True, type=bool, help="whethere to add model adapter layers")
parser.add_argument("--adapter_stride", default=2, type=int, help="stride of adapter layers")
parser.add_argument("--adapter_kernel_size", default=3, type=int, help="kernel size of adapter layers")
parser.add_argument("--encoder_output_dim", default=1_024, type=int, help="encoder output dim")
parser.add_argument("--start_token_id", default=250_004, type=int, help="`decoder_start_token_id` of model config")
a_ :Union[str, Any] = parser.parse_args()
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.dict_path,
args.config_yaml_path,
encoder_config_path=args.encoder_config_path,
decoder_config_path=args.decoder_config_path,
add_adapter=args.add_adapter,
adapter_kernel_size=args.adapter_kernel_size,
adapter_stride=args.adapter_stride,
decoder_start_token_id=args.start_token_id,
encoder_output_dim=args.encoder_output_dim,
)
| 277 | 0 |
'''simple docstring'''
import os
from shutil import copyfile
from typing import List, Optional, Tuple
from ...tokenization_utils import AddedToken
from ...tokenization_utils_fast import PreTrainedTokenizerFast
from ...utils import is_sentencepiece_available, logging
if is_sentencepiece_available():
from .tokenization_fnet import FNetTokenizer
else:
_A : List[Any] = None
_A : List[str] = logging.get_logger(__name__)
_A : Optional[Any] = {"vocab_file": "spiece.model", "tokenizer_file": "tokenizer.json"}
_A : Dict = {
"vocab_file": {
"google/fnet-base": "https://huggingface.co/google/fnet-base/resolve/main/spiece.model",
"google/fnet-large": "https://huggingface.co/google/fnet-large/resolve/main/spiece.model",
},
"tokenizer_file": {
"google/fnet-base": "https://huggingface.co/google/fnet-base/resolve/main/tokenizer.json",
"google/fnet-large": "https://huggingface.co/google/fnet-large/resolve/main/tokenizer.json",
},
}
_A : Optional[Any] = {
"google/fnet-base": 512,
"google/fnet-large": 512,
}
_A : Optional[Any] = "▁"
class _lowercase ( lowerCAmelCase_ ):
'''simple docstring'''
_SCREAMING_SNAKE_CASE : Any = VOCAB_FILES_NAMES
_SCREAMING_SNAKE_CASE : Dict = PRETRAINED_VOCAB_FILES_MAP
_SCREAMING_SNAKE_CASE : str = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
_SCREAMING_SNAKE_CASE : int = ["""input_ids""", """token_type_ids"""]
_SCREAMING_SNAKE_CASE : List[Any] = FNetTokenizer
def __init__( self : Tuple , SCREAMING_SNAKE_CASE__ : Dict=None , SCREAMING_SNAKE_CASE__ : str=None , SCREAMING_SNAKE_CASE__ : int=False , SCREAMING_SNAKE_CASE__ : Optional[int]=True , SCREAMING_SNAKE_CASE__ : Tuple=True , SCREAMING_SNAKE_CASE__ : Dict="<unk>" , SCREAMING_SNAKE_CASE__ : Any="[SEP]" , SCREAMING_SNAKE_CASE__ : List[str]="<pad>" , SCREAMING_SNAKE_CASE__ : Union[str, Any]="[CLS]" , SCREAMING_SNAKE_CASE__ : List[str]="[MASK]" , **SCREAMING_SNAKE_CASE__ : Any , ) -> Tuple:
# Mask token behave like a normal word, i.e. include the space before it and
# is included in the raw text, there should be a match in a non-normalized sentence.
__lowerCAmelCase = (
AddedToken(_snake_case , lstrip=_snake_case , rstrip=_snake_case , normalized=_snake_case )
if isinstance(_snake_case , _snake_case )
else mask_token
)
super().__init__(
_snake_case , tokenizer_file=_snake_case , do_lower_case=_snake_case , remove_space=_snake_case , keep_accents=_snake_case , unk_token=_snake_case , sep_token=_snake_case , pad_token=_snake_case , cls_token=_snake_case , mask_token=_snake_case , **_snake_case , )
__lowerCAmelCase = do_lower_case
__lowerCAmelCase = remove_space
__lowerCAmelCase = keep_accents
__lowerCAmelCase = vocab_file
__lowerCAmelCase = False if not self.vocab_file else True
def a ( self : Any , SCREAMING_SNAKE_CASE__ : List[int] , SCREAMING_SNAKE_CASE__ : Optional[List[int]] = None ) -> List[int]:
__lowerCAmelCase = [self.sep_token_id]
__lowerCAmelCase = [self.cls_token_id]
if token_ids_a is None:
return cls + token_ids_a + sep
return cls + token_ids_a + sep + token_ids_a + sep
def a ( self : Optional[Any] , SCREAMING_SNAKE_CASE__ : List[int] , SCREAMING_SNAKE_CASE__ : Optional[List[int]] = None ) -> List[int]:
__lowerCAmelCase = [self.sep_token_id]
__lowerCAmelCase = [self.cls_token_id]
if token_ids_a is None:
return len(cls + token_ids_a + sep ) * [0]
return len(cls + token_ids_a + sep ) * [0] + len(token_ids_a + sep ) * [1]
def a ( self : Optional[int] , SCREAMING_SNAKE_CASE__ : str , SCREAMING_SNAKE_CASE__ : Optional[str] = None ) -> Tuple[str]:
if not os.path.isdir(_snake_case ):
logger.error(f"""Vocabulary path ({save_directory}) should be a directory""" )
return
__lowerCAmelCase = os.path.join(
_snake_case , (filename_prefix + """-""" if filename_prefix else """""") + VOCAB_FILES_NAMES["""vocab_file"""] )
if os.path.abspath(self.vocab_file ) != os.path.abspath(_snake_case ):
copyfile(self.vocab_file , _snake_case )
return (out_vocab_file,)
| 229 |
from collections import OrderedDict
from typing import TYPE_CHECKING, Any, Mapping, Optional, Union
from ...configuration_utils import PretrainedConfig
from ...onnx import OnnxConfig
from ...utils import logging
if TYPE_CHECKING:
from ... import FeatureExtractionMixin, PreTrainedTokenizerBase, TensorType
a_ :Tuple = logging.get_logger(__name__)
a_ :Union[str, Any] = {
"microsoft/deberta-v2-xlarge": "https://huggingface.co/microsoft/deberta-v2-xlarge/resolve/main/config.json",
"microsoft/deberta-v2-xxlarge": "https://huggingface.co/microsoft/deberta-v2-xxlarge/resolve/main/config.json",
"microsoft/deberta-v2-xlarge-mnli": (
"https://huggingface.co/microsoft/deberta-v2-xlarge-mnli/resolve/main/config.json"
),
"microsoft/deberta-v2-xxlarge-mnli": (
"https://huggingface.co/microsoft/deberta-v2-xxlarge-mnli/resolve/main/config.json"
),
}
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """deberta-v2"""
def __init__( self : Union[str, Any], _snake_case : Dict=1_2_8_1_0_0, _snake_case : Any=1_5_3_6, _snake_case : Tuple=2_4, _snake_case : int=2_4, _snake_case : Optional[int]=6_1_4_4, _snake_case : Optional[int]="gelu", _snake_case : Optional[int]=0.1, _snake_case : List[str]=0.1, _snake_case : str=5_1_2, _snake_case : Optional[int]=0, _snake_case : Optional[int]=0.0_2, _snake_case : Dict=1e-7, _snake_case : int=False, _snake_case : Any=-1, _snake_case : List[str]=0, _snake_case : Tuple=True, _snake_case : Any=None, _snake_case : Union[str, Any]=0, _snake_case : Tuple="gelu", **_snake_case : Union[str, Any], ) ->Optional[int]:
super().__init__(**_snake_case )
snake_case__ : Dict = hidden_size
snake_case__ : Optional[int] = num_hidden_layers
snake_case__ : Any = num_attention_heads
snake_case__ : List[Any] = intermediate_size
snake_case__ : List[Any] = hidden_act
snake_case__ : Union[str, Any] = hidden_dropout_prob
snake_case__ : Dict = attention_probs_dropout_prob
snake_case__ : List[str] = max_position_embeddings
snake_case__ : List[str] = type_vocab_size
snake_case__ : Optional[Any] = initializer_range
snake_case__ : Optional[int] = relative_attention
snake_case__ : Tuple = max_relative_positions
snake_case__ : Union[str, Any] = pad_token_id
snake_case__ : Optional[int] = position_biased_input
# Backwards compatibility
if type(_snake_case ) == str:
snake_case__ : int = [x.strip() for x in pos_att_type.lower().split('|' )]
snake_case__ : List[str] = pos_att_type
snake_case__ : Union[str, Any] = vocab_size
snake_case__ : Optional[int] = layer_norm_eps
snake_case__ : Optional[int] = kwargs.get('pooler_hidden_size', _snake_case )
snake_case__ : int = pooler_dropout
snake_case__ : str = pooler_hidden_act
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
@property
def lowercase_ ( self : Optional[int] ) ->Mapping[str, Mapping[int, str]]:
if self.task == "multiple-choice":
snake_case__ : List[Any] = {0: 'batch', 1: 'choice', 2: 'sequence'}
else:
snake_case__ : int = {0: 'batch', 1: 'sequence'}
if self._config.type_vocab_size > 0:
return OrderedDict(
[('input_ids', dynamic_axis), ('attention_mask', dynamic_axis), ('token_type_ids', dynamic_axis)] )
else:
return OrderedDict([('input_ids', dynamic_axis), ('attention_mask', dynamic_axis)] )
@property
def lowercase_ ( self : Dict ) ->int:
return 1_2
def lowercase_ ( self : Tuple, _snake_case : Union["PreTrainedTokenizerBase", "FeatureExtractionMixin"], _snake_case : int = -1, _snake_case : int = -1, _snake_case : int = -1, _snake_case : bool = False, _snake_case : Optional["TensorType"] = None, _snake_case : int = 3, _snake_case : int = 4_0, _snake_case : int = 4_0, _snake_case : "PreTrainedTokenizerBase" = None, ) ->Mapping[str, Any]:
snake_case__ : Union[str, Any] = super().generate_dummy_inputs(preprocessor=_snake_case, framework=_snake_case )
if self._config.type_vocab_size == 0 and "token_type_ids" in dummy_inputs:
del dummy_inputs["token_type_ids"]
return dummy_inputs
| 277 | 0 |
"""simple docstring"""
import os
import unittest
from transformers.models.transfo_xl.tokenization_transfo_xl import VOCAB_FILES_NAMES, TransfoXLTokenizer
from ...test_tokenization_common import TokenizerTesterMixin
class UpperCAmelCase_ ( lowerCAmelCase_ , unittest.TestCase):
snake_case__ = TransfoXLTokenizer
snake_case__ = False
snake_case__ = False
def _UpperCamelCase ( self : Optional[int] ) -> Any:
super().setUp()
_UpperCamelCase = [
'<unk>',
'[CLS]',
'[SEP]',
'want',
'unwanted',
'wa',
'un',
'running',
',',
'low',
'l',
]
_UpperCamelCase = os.path.join(self.tmpdirname , VOCAB_FILES_NAMES['''vocab_file'''] )
with open(self.vocab_file , '''w''' , encoding='''utf-8''' ) as vocab_writer:
vocab_writer.write(''''''.join([x + '''\n''' for x in vocab_tokens] ) )
def _UpperCamelCase ( self : Union[str, Any] , **__UpperCamelCase : List[Any] ) -> Dict:
_UpperCamelCase = True
return TransfoXLTokenizer.from_pretrained(self.tmpdirname , **_snake_case )
def _UpperCamelCase ( self : Optional[Any] , __UpperCamelCase : str ) -> Dict:
_UpperCamelCase = '<unk> UNwanted , running'
_UpperCamelCase = '<unk> unwanted, running'
return input_text, output_text
def _UpperCamelCase ( self : List[Any] ) -> Tuple:
_UpperCamelCase = TransfoXLTokenizer(vocab_file=self.vocab_file , lower_case=_snake_case )
_UpperCamelCase = tokenizer.tokenize('''<unk> UNwanted , running''' )
self.assertListEqual(_snake_case , ['''<unk>''', '''unwanted''', ''',''', '''running'''] )
self.assertListEqual(tokenizer.convert_tokens_to_ids(_snake_case ) , [0, 4, 8, 7] )
def _UpperCamelCase ( self : List[str] ) -> List[Any]:
_UpperCamelCase = TransfoXLTokenizer(lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(''' \tHeLLo ! how \n Are yoU ? ''' ) , ['''hello''', '''!''', '''how''', '''are''', '''you''', '''?'''] )
def _UpperCamelCase ( self : Optional[int] ) -> Optional[Any]:
_UpperCamelCase = TransfoXLTokenizer(lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(''' \tHeLLo ! how \n Are yoU ? ''' ) , ['''HeLLo''', '''!''', '''how''', '''Are''', '''yoU''', '''?'''] )
def _UpperCamelCase ( self : Optional[int] ) -> Union[str, Any]:
_UpperCamelCase = TransfoXLTokenizer(lower_case=_snake_case )
_UpperCamelCase = 'Hello (bracket) and side-scrolled [and] Henry\'s $5,000 with 3.34 m. What\'s up!?'
_UpperCamelCase = [
'Hello',
'(',
'bracket',
')',
'and',
'side',
'@-@',
'scrolled',
'[',
'and',
']',
'Henry',
'\'s',
'$',
'5',
'@,@',
'000',
'with',
'3',
'@.@',
'34',
'm',
'.',
'What',
'\'s',
'up',
'!',
'?',
]
self.assertListEqual(tokenizer.tokenize(_snake_case ) , _snake_case )
self.assertEqual(tokenizer.convert_tokens_to_string(_snake_case ) , _snake_case )
def _UpperCamelCase ( self : Dict ) -> Any:
_UpperCamelCase = self.get_tokenizer()
_UpperCamelCase = len(_snake_case )
tokenizer.add_tokens(['''new1''', '''new2'''] )
tokenizer.move_added_token('''new1''' , 1 )
# Check that moved token is not copied (duplicate)
self.assertEqual(len(_snake_case ) , original_len + 2 )
# Check that token is moved to specified id
self.assertEqual(tokenizer.encode('''new1''' ) , [1] )
self.assertEqual(tokenizer.decode([1] ) , '''new1''' )
| 256 |
import argparse
import json
import pickle
from pathlib import Path
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import MaskFormerConfig, MaskFormerForInstanceSegmentation, MaskFormerImageProcessor, SwinConfig
from transformers.utils import logging
logging.set_verbosity_info()
a_ :str = logging.get_logger(__name__)
def lowercase_ (A : str ):
snake_case__ : Tuple = SwinConfig.from_pretrained(
'microsoft/swin-tiny-patch4-window7-224' , out_features=['stage1', 'stage2', 'stage3', 'stage4'] )
snake_case__ : List[Any] = MaskFormerConfig(backbone_config=A )
snake_case__ : Union[str, Any] = 'huggingface/label-files'
if "ade20k-full" in model_name:
# this should be ok
snake_case__ : Dict = 8_4_7
snake_case__ : List[str] = 'maskformer-ade20k-full-id2label.json'
elif "ade" in model_name:
# this should be ok
snake_case__ : Union[str, Any] = 1_5_0
snake_case__ : Any = 'ade20k-id2label.json'
elif "coco-stuff" in model_name:
# this should be ok
snake_case__ : List[str] = 1_7_1
snake_case__ : Union[str, Any] = 'maskformer-coco-stuff-id2label.json'
elif "coco" in model_name:
# TODO
snake_case__ : Dict = 1_3_3
snake_case__ : str = 'coco-panoptic-id2label.json'
elif "cityscapes" in model_name:
# this should be ok
snake_case__ : List[str] = 1_9
snake_case__ : Union[str, Any] = 'cityscapes-id2label.json'
elif "vistas" in model_name:
# this should be ok
snake_case__ : Tuple = 6_5
snake_case__ : List[str] = 'mapillary-vistas-id2label.json'
snake_case__ : Dict = json.load(open(hf_hub_download(A , A , repo_type='dataset' ) , 'r' ) )
snake_case__ : List[str] = {int(A ): v for k, v in idalabel.items()}
return config
def lowercase_ (A : Any ):
snake_case__ : Optional[int] = []
# stem
# fmt: off
rename_keys.append(('backbone.patch_embed.proj.weight', 'model.pixel_level_module.encoder.model.embeddings.patch_embeddings.projection.weight') )
rename_keys.append(('backbone.patch_embed.proj.bias', 'model.pixel_level_module.encoder.model.embeddings.patch_embeddings.projection.bias') )
rename_keys.append(('backbone.patch_embed.norm.weight', 'model.pixel_level_module.encoder.model.embeddings.norm.weight') )
rename_keys.append(('backbone.patch_embed.norm.bias', 'model.pixel_level_module.encoder.model.embeddings.norm.bias') )
# stages
for i in range(len(config.backbone_config.depths ) ):
for j in range(config.backbone_config.depths[i] ):
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm1.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_before.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm1.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_before.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.relative_position_bias_table''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.self.relative_position_bias_table''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.relative_position_index''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.self.relative_position_index''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.proj.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.output.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.proj.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.output.dense.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm2.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_after.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm2.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_after.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc1.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.intermediate.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc1.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.intermediate.dense.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc2.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.output.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc2.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.output.dense.bias''') )
if i < 3:
rename_keys.append((F'''backbone.layers.{i}.downsample.reduction.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.reduction.weight''') )
rename_keys.append((F'''backbone.layers.{i}.downsample.norm.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.norm.weight''') )
rename_keys.append((F'''backbone.layers.{i}.downsample.norm.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.norm.bias''') )
rename_keys.append((F'''backbone.norm{i}.weight''', F'''model.pixel_level_module.encoder.hidden_states_norms.{i}.weight''') )
rename_keys.append((F'''backbone.norm{i}.bias''', F'''model.pixel_level_module.encoder.hidden_states_norms.{i}.bias''') )
# FPN
rename_keys.append(('sem_seg_head.layer_4.weight', 'model.pixel_level_module.decoder.fpn.stem.0.weight') )
rename_keys.append(('sem_seg_head.layer_4.norm.weight', 'model.pixel_level_module.decoder.fpn.stem.1.weight') )
rename_keys.append(('sem_seg_head.layer_4.norm.bias', 'model.pixel_level_module.decoder.fpn.stem.1.bias') )
for source_index, target_index in zip(range(3 , 0 , -1 ) , range(0 , 3 ) ):
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.0.weight''') )
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.norm.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.1.weight''') )
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.norm.bias''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.1.bias''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.0.weight''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.norm.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.1.weight''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.norm.bias''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.1.bias''') )
rename_keys.append(('sem_seg_head.mask_features.weight', 'model.pixel_level_module.decoder.mask_projection.weight') )
rename_keys.append(('sem_seg_head.mask_features.bias', 'model.pixel_level_module.decoder.mask_projection.bias') )
# Transformer decoder
for idx in range(config.decoder_config.decoder_layers ):
# self-attention out projection
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.out_proj.weight''', F'''model.transformer_module.decoder.layers.{idx}.self_attn.out_proj.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.out_proj.bias''', F'''model.transformer_module.decoder.layers.{idx}.self_attn.out_proj.bias''') )
# cross-attention out projection
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.out_proj.weight''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn.out_proj.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.out_proj.bias''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn.out_proj.bias''') )
# MLP 1
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear1.weight''', F'''model.transformer_module.decoder.layers.{idx}.fc1.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear1.bias''', F'''model.transformer_module.decoder.layers.{idx}.fc1.bias''') )
# MLP 2
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear2.weight''', F'''model.transformer_module.decoder.layers.{idx}.fc2.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear2.bias''', F'''model.transformer_module.decoder.layers.{idx}.fc2.bias''') )
# layernorm 1 (self-attention layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm1.weight''', F'''model.transformer_module.decoder.layers.{idx}.self_attn_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm1.bias''', F'''model.transformer_module.decoder.layers.{idx}.self_attn_layer_norm.bias''') )
# layernorm 2 (cross-attention layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm2.weight''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm2.bias''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn_layer_norm.bias''') )
# layernorm 3 (final layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm3.weight''', F'''model.transformer_module.decoder.layers.{idx}.final_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm3.bias''', F'''model.transformer_module.decoder.layers.{idx}.final_layer_norm.bias''') )
rename_keys.append(('sem_seg_head.predictor.transformer.decoder.norm.weight', 'model.transformer_module.decoder.layernorm.weight') )
rename_keys.append(('sem_seg_head.predictor.transformer.decoder.norm.bias', 'model.transformer_module.decoder.layernorm.bias') )
# heads on top
rename_keys.append(('sem_seg_head.predictor.query_embed.weight', 'model.transformer_module.queries_embedder.weight') )
rename_keys.append(('sem_seg_head.predictor.input_proj.weight', 'model.transformer_module.input_projection.weight') )
rename_keys.append(('sem_seg_head.predictor.input_proj.bias', 'model.transformer_module.input_projection.bias') )
rename_keys.append(('sem_seg_head.predictor.class_embed.weight', 'class_predictor.weight') )
rename_keys.append(('sem_seg_head.predictor.class_embed.bias', 'class_predictor.bias') )
for i in range(3 ):
rename_keys.append((F'''sem_seg_head.predictor.mask_embed.layers.{i}.weight''', F'''mask_embedder.{i}.0.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.mask_embed.layers.{i}.bias''', F'''mask_embedder.{i}.0.bias''') )
# fmt: on
return rename_keys
def lowercase_ (A : Tuple , A : Tuple , A : Optional[Any] ):
snake_case__ : Optional[int] = dct.pop(A )
snake_case__ : Union[str, Any] = val
def lowercase_ (A : Optional[Any] , A : Tuple ):
snake_case__ : Optional[int] = [int(backbone_config.embed_dim * 2**i ) for i in range(len(backbone_config.depths ) )]
for i in range(len(backbone_config.depths ) ):
snake_case__ : Optional[int] = num_features[i]
for j in range(backbone_config.depths[i] ):
# fmt: off
# read in weights + bias of input projection layer (in original implementation, this is a single matrix + bias)
snake_case__ : int = state_dict.pop(F'''backbone.layers.{i}.blocks.{j}.attn.qkv.weight''' )
snake_case__ : Tuple = state_dict.pop(F'''backbone.layers.{i}.blocks.{j}.attn.qkv.bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : str = in_proj_weight[:dim, :]
snake_case__ : int = in_proj_bias[: dim]
snake_case__ : List[Any] = in_proj_weight[
dim : dim * 2, :
]
snake_case__ : List[str] = in_proj_bias[
dim : dim * 2
]
snake_case__ : List[Any] = in_proj_weight[
-dim :, :
]
snake_case__ : Dict = in_proj_bias[-dim :]
# fmt: on
def lowercase_ (A : List[str] , A : List[Any] ):
# fmt: off
snake_case__ : str = config.decoder_config.hidden_size
for idx in range(config.decoder_config.decoder_layers ):
# read in weights + bias of self-attention input projection layer (in the original implementation, this is a single matrix + bias)
snake_case__ : List[Any] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.in_proj_weight''' )
snake_case__ : int = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.in_proj_bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : Any = in_proj_weight[: hidden_size, :]
snake_case__ : Tuple = in_proj_bias[:config.hidden_size]
snake_case__ : List[str] = in_proj_weight[hidden_size : hidden_size * 2, :]
snake_case__ : Dict = in_proj_bias[hidden_size : hidden_size * 2]
snake_case__ : Any = in_proj_weight[-hidden_size :, :]
snake_case__ : int = in_proj_bias[-hidden_size :]
# read in weights + bias of cross-attention input projection layer (in the original implementation, this is a single matrix + bias)
snake_case__ : List[Any] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.in_proj_weight''' )
snake_case__ : List[str] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.in_proj_bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : Optional[int] = in_proj_weight[: hidden_size, :]
snake_case__ : Optional[Any] = in_proj_bias[:config.hidden_size]
snake_case__ : int = in_proj_weight[hidden_size : hidden_size * 2, :]
snake_case__ : List[str] = in_proj_bias[hidden_size : hidden_size * 2]
snake_case__ : List[str] = in_proj_weight[-hidden_size :, :]
snake_case__ : str = in_proj_bias[-hidden_size :]
# fmt: on
def lowercase_ ():
snake_case__ : Any = 'http://images.cocodataset.org/val2017/000000039769.jpg'
snake_case__ : int = Image.open(requests.get(A , stream=A ).raw )
return im
@torch.no_grad()
def lowercase_ (A : str , A : str , A : str , A : bool = False ):
snake_case__ : Optional[int] = get_maskformer_config(A )
# load original state_dict
with open(A , 'rb' ) as f:
snake_case__ : List[Any] = pickle.load(A )
snake_case__ : Optional[int] = data['model']
# for name, param in state_dict.items():
# print(name, param.shape)
# rename keys
snake_case__ : List[str] = create_rename_keys(A )
for src, dest in rename_keys:
rename_key(A , A , A )
read_in_swin_q_k_v(A , config.backbone_config )
read_in_decoder_q_k_v(A , A )
# update to torch tensors
for key, value in state_dict.items():
snake_case__ : int = torch.from_numpy(A )
# load 🤗 model
snake_case__ : str = MaskFormerForInstanceSegmentation(A )
model.eval()
for name, param in model.named_parameters():
print(A , param.shape )
snake_case__ , snake_case__ : Union[str, Any] = model.load_state_dict(A , strict=A )
assert missing_keys == [
"model.pixel_level_module.encoder.model.layernorm.weight",
"model.pixel_level_module.encoder.model.layernorm.bias",
]
assert len(A ) == 0, F'''Unexpected keys: {unexpected_keys}'''
# verify results
snake_case__ : Optional[Any] = prepare_img()
if "vistas" in model_name:
snake_case__ : int = 6_5
elif "cityscapes" in model_name:
snake_case__ : Dict = 6_5_5_3_5
else:
snake_case__ : Tuple = 2_5_5
snake_case__ : Optional[int] = True if 'ade' in model_name else False
snake_case__ : Dict = MaskFormerImageProcessor(ignore_index=A , reduce_labels=A )
snake_case__ : Any = image_processor(A , return_tensors='pt' )
snake_case__ : Any = model(**A )
print('Logits:' , outputs.class_queries_logits[0, :3, :3] )
if model_name == "maskformer-swin-tiny-ade":
snake_case__ : Tuple = torch.tensor(
[[3.6353, -4.4770, -2.6065], [0.5081, -4.2394, -3.5343], [2.1909, -5.0353, -1.9323]] )
assert torch.allclose(outputs.class_queries_logits[0, :3, :3] , A , atol=1e-4 )
print('Looks ok!' )
if pytorch_dump_folder_path is not None:
print(F'''Saving model and image processor to {pytorch_dump_folder_path}''' )
Path(A ).mkdir(exist_ok=A )
model.save_pretrained(A )
image_processor.save_pretrained(A )
if push_to_hub:
print('Pushing model and image processor to the hub...' )
model.push_to_hub(F'''nielsr/{model_name}''' )
image_processor.push_to_hub(F'''nielsr/{model_name}''' )
if __name__ == "__main__":
a_ :Optional[int] = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--model_name",
default="maskformer-swin-tiny-ade",
type=str,
help=("Name of the MaskFormer model you'd like to convert",),
)
parser.add_argument(
"--checkpoint_path",
default="/Users/nielsrogge/Documents/MaskFormer_checkpoints/MaskFormer-Swin-tiny-ADE20k/model.pkl",
type=str,
help="Path to the original state dict (.pth file).",
)
parser.add_argument(
"--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model directory."
)
parser.add_argument(
"--push_to_hub", action="store_true", help="Whether or not to push the converted model to the 🤗 hub."
)
a_ :Dict = parser.parse_args()
convert_maskformer_checkpoint(
args.model_name, args.checkpoint_path, args.pytorch_dump_folder_path, args.push_to_hub
)
| 277 | 0 |
from __future__ import annotations
_a = 1_0
def _a ( SCREAMING_SNAKE_CASE : list[int] ) -> Any:
"""simple docstring"""
__lowerCAmelCase: Any = 1
__lowerCAmelCase: Optional[int] = max(SCREAMING_SNAKE_CASE )
while placement <= max_digit:
# declare and initialize empty buckets
__lowerCAmelCase: list[list] = [[] for _ in range(SCREAMING_SNAKE_CASE )]
# split list_of_ints between the buckets
for i in list_of_ints:
__lowerCAmelCase: Dict = int((i / placement) % RADIX )
buckets[tmp].append(SCREAMING_SNAKE_CASE )
# put each buckets' contents into list_of_ints
__lowerCAmelCase: str = 0
for b in range(SCREAMING_SNAKE_CASE ):
for i in buckets[b]:
__lowerCAmelCase: str = i
a += 1
# move to next
placement *= RADIX
return list_of_ints
if __name__ == "__main__":
import doctest
doctest.testmod()
| 322 |
import unittest
from transformers import is_torch_available
from transformers.testing_utils import require_torch, slow, torch_device
from ...generation.test_utils import GenerationTesterMixin
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers import (
OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST,
OpenAIGPTConfig,
OpenAIGPTDoubleHeadsModel,
OpenAIGPTForSequenceClassification,
OpenAIGPTLMHeadModel,
OpenAIGPTModel,
)
class snake_case__ :
"""simple docstring"""
def __init__( self : List[str], _snake_case : Any, _snake_case : int=1_3, _snake_case : Optional[int]=7, _snake_case : int=True, _snake_case : Optional[Any]=True, _snake_case : Optional[Any]=True, _snake_case : Union[str, Any]=9_9, _snake_case : Optional[Any]=3_2, _snake_case : Tuple=5, _snake_case : str=4, _snake_case : Any=3_7, _snake_case : int="gelu", _snake_case : Optional[Any]=0.1, _snake_case : str=0.1, _snake_case : str=5_1_2, _snake_case : Dict=1_6, _snake_case : str=2, _snake_case : Union[str, Any]=0.0_2, _snake_case : Optional[int]=3, _snake_case : Union[str, Any]=4, _snake_case : Tuple=None, ) ->Optional[Any]:
snake_case__ : Optional[int] = parent
snake_case__ : List[Any] = batch_size
snake_case__ : Tuple = seq_length
snake_case__ : str = is_training
snake_case__ : Optional[int] = use_token_type_ids
snake_case__ : Any = use_labels
snake_case__ : Dict = vocab_size
snake_case__ : str = hidden_size
snake_case__ : Union[str, Any] = num_hidden_layers
snake_case__ : List[str] = num_attention_heads
snake_case__ : Union[str, Any] = intermediate_size
snake_case__ : List[Any] = hidden_act
snake_case__ : int = hidden_dropout_prob
snake_case__ : str = attention_probs_dropout_prob
snake_case__ : Any = max_position_embeddings
snake_case__ : Union[str, Any] = type_vocab_size
snake_case__ : Optional[Any] = type_sequence_label_size
snake_case__ : Optional[int] = initializer_range
snake_case__ : Optional[int] = num_labels
snake_case__ : str = num_choices
snake_case__ : int = scope
snake_case__ : List[str] = self.vocab_size - 1
def lowercase_ ( self : Union[str, Any] ) ->Tuple:
snake_case__ : List[str] = ids_tensor([self.batch_size, self.seq_length], self.vocab_size )
snake_case__ : List[str] = None
if self.use_token_type_ids:
snake_case__ : Optional[Any] = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size )
snake_case__ : Tuple = None
snake_case__ : str = None
snake_case__ : List[Any] = None
if self.use_labels:
snake_case__ : Dict = ids_tensor([self.batch_size], self.type_sequence_label_size )
snake_case__ : int = ids_tensor([self.batch_size, self.seq_length], self.num_labels )
snake_case__ : List[str] = ids_tensor([self.batch_size], self.num_choices )
snake_case__ : Union[str, Any] = OpenAIGPTConfig(
vocab_size=self.vocab_size, n_embd=self.hidden_size, n_layer=self.num_hidden_layers, n_head=self.num_attention_heads, n_positions=self.max_position_embeddings, pad_token_id=self.pad_token_id, )
snake_case__ : List[str] = ids_tensor([self.num_hidden_layers, self.num_attention_heads], 2 )
return (
config,
input_ids,
head_mask,
token_type_ids,
sequence_labels,
token_labels,
choice_labels,
)
def lowercase_ ( self : Any, _snake_case : List[str], _snake_case : Any, _snake_case : List[Any], _snake_case : Tuple, *_snake_case : Optional[Any] ) ->Tuple:
snake_case__ : Union[str, Any] = OpenAIGPTModel(config=_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Optional[Any] = model(_snake_case, token_type_ids=_snake_case, head_mask=_snake_case )
snake_case__ : Union[str, Any] = model(_snake_case, token_type_ids=_snake_case )
snake_case__ : Optional[Any] = model(_snake_case )
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size) )
def lowercase_ ( self : Optional[int], _snake_case : Optional[Any], _snake_case : Union[str, Any], _snake_case : Optional[int], _snake_case : List[Any], *_snake_case : Dict ) ->Optional[int]:
snake_case__ : Optional[Any] = OpenAIGPTLMHeadModel(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Tuple = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.loss.shape, () )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size) )
def lowercase_ ( self : int, _snake_case : Tuple, _snake_case : List[str], _snake_case : List[Any], _snake_case : List[Any], *_snake_case : List[Any] ) ->Optional[int]:
snake_case__ : List[str] = OpenAIGPTDoubleHeadsModel(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Optional[Any] = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.loss.shape, () )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size) )
def lowercase_ ( self : Optional[int], _snake_case : Tuple, _snake_case : Dict, _snake_case : List[str], _snake_case : Optional[Any], *_snake_case : Union[str, Any] ) ->str:
snake_case__ : List[str] = self.num_labels
snake_case__ : Dict = OpenAIGPTForSequenceClassification(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : List[str] = ids_tensor([self.batch_size], self.type_sequence_label_size )
snake_case__ : List[str] = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels) )
def lowercase_ ( self : Dict ) ->int:
snake_case__ : List[Any] = self.prepare_config_and_inputs()
(
(
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) ,
) : Optional[Any] = config_and_inputs
snake_case__ : str = {
'input_ids': input_ids,
'token_type_ids': token_type_ids,
'head_mask': head_mask,
}
return config, inputs_dict
@require_torch
class snake_case__ ( lowerCAmelCase_ , lowerCAmelCase_ , lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = (
(OpenAIGPTModel, OpenAIGPTLMHeadModel, OpenAIGPTDoubleHeadsModel, OpenAIGPTForSequenceClassification)
if is_torch_available()
else ()
)
_SCREAMING_SNAKE_CASE = (
(OpenAIGPTLMHeadModel,) if is_torch_available() else ()
) # TODO (PVP): Add Double HeadsModel when generate() function is changed accordingly
_SCREAMING_SNAKE_CASE = (
{
"""feature-extraction""": OpenAIGPTModel,
"""text-classification""": OpenAIGPTForSequenceClassification,
"""text-generation""": OpenAIGPTLMHeadModel,
"""zero-shot""": OpenAIGPTForSequenceClassification,
}
if is_torch_available()
else {}
)
def lowercase_ ( self : Optional[int], _snake_case : Union[str, Any], _snake_case : int, _snake_case : Tuple, _snake_case : Tuple, _snake_case : List[str] ) ->Optional[Any]:
if pipeline_test_casse_name == "ZeroShotClassificationPipelineTests":
# Get `tokenizer does not have a padding token` error for both fast/slow tokenizers.
# `OpenAIGPTConfig` was never used in pipeline tests, either because of a missing checkpoint or because a
# tiny config could not be created.
return True
return False
def lowercase_ ( self : Optional[Any], _snake_case : Union[str, Any], _snake_case : List[str], _snake_case : Any=False ) ->Tuple:
snake_case__ : Optional[int] = super()._prepare_for_class(_snake_case, _snake_case, return_labels=_snake_case )
if return_labels:
if model_class.__name__ == "OpenAIGPTDoubleHeadsModel":
snake_case__ : Union[str, Any] = torch.zeros(
(self.model_tester.batch_size, self.model_tester.num_choices, self.model_tester.seq_length), dtype=torch.long, device=_snake_case, )
snake_case__ : List[Any] = inputs_dict['labels']
snake_case__ : List[Any] = inputs_dict['labels']
snake_case__ : Any = torch.zeros(
(self.model_tester.batch_size, self.model_tester.num_choices), dtype=torch.long, device=_snake_case, )
snake_case__ : Tuple = torch.zeros(
self.model_tester.batch_size, dtype=torch.long, device=_snake_case )
return inputs_dict
def lowercase_ ( self : Union[str, Any] ) ->List[str]:
snake_case__ : List[str] = OpenAIGPTModelTester(self )
snake_case__ : Any = ConfigTester(self, config_class=_snake_case, n_embd=3_7 )
def lowercase_ ( self : Optional[int] ) ->str:
self.config_tester.run_common_tests()
def lowercase_ ( self : int ) ->Tuple:
snake_case__ : str = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_openai_gpt_model(*_snake_case )
def lowercase_ ( self : Tuple ) ->List[str]:
snake_case__ : Optional[int] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_lm_head_model(*_snake_case )
def lowercase_ ( self : Dict ) ->int:
snake_case__ : Optional[Any] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_double_lm_head_model(*_snake_case )
def lowercase_ ( self : int ) ->str:
snake_case__ : List[Any] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_openai_gpt_for_sequence_classification(*_snake_case )
@slow
def lowercase_ ( self : Optional[Any] ) ->str:
for model_name in OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
snake_case__ : Optional[int] = OpenAIGPTModel.from_pretrained(_snake_case )
self.assertIsNotNone(_snake_case )
@require_torch
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
@slow
def lowercase_ ( self : Tuple ) ->Optional[int]:
snake_case__ : Union[str, Any] = OpenAIGPTLMHeadModel.from_pretrained('openai-gpt' )
model.to(_snake_case )
snake_case__ : Tuple = torch.tensor([[4_8_1, 4_7_3_5, 5_4_4]], dtype=torch.long, device=_snake_case ) # the president is
snake_case__ : int = [
4_8_1,
4_7_3_5,
5_4_4,
2_4_6,
9_6_3,
8_7_0,
7_6_2,
2_3_9,
2_4_4,
4_0_4_7_7,
2_4_4,
2_4_9,
7_1_9,
8_8_1,
4_8_7,
5_4_4,
2_4_0,
2_4_4,
6_0_3,
4_8_1,
] # the president is a very good man. " \n " i\'m sure he is, " said the
snake_case__ : Optional[int] = model.generate(_snake_case, do_sample=_snake_case )
self.assertListEqual(output_ids[0].tolist(), _snake_case )
| 277 | 0 |
from ..utils import DummyObject, requires_backends
class __lowerCAmelCase ( metaclass=lowerCAmelCase_):
_lowercase : Tuple = ["""torch""", """torchsde"""]
def __init__( self , *lowerCAmelCase__ , **lowerCAmelCase__ ) -> Optional[Any]:
'''simple docstring'''
requires_backends(self , ["torch", "torchsde"] )
@classmethod
def _lowercase ( cls , *lowerCAmelCase__ , **lowerCAmelCase__ ) -> Union[str, Any]:
'''simple docstring'''
requires_backends(cls , ["torch", "torchsde"] )
@classmethod
def _lowercase ( cls , *lowerCAmelCase__ , **lowerCAmelCase__ ) -> Optional[Any]:
'''simple docstring'''
requires_backends(cls , ["torch", "torchsde"] )
| 95 |
import os
import unittest
from transformers.models.transfo_xl.tokenization_transfo_xl import VOCAB_FILES_NAMES, TransfoXLTokenizer
from ...test_tokenization_common import TokenizerTesterMixin
class snake_case__ ( lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = TransfoXLTokenizer
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
def lowercase_ ( self : Optional[int] ) ->Any:
super().setUp()
snake_case__ : Tuple = [
'<unk>',
'[CLS]',
'[SEP]',
'want',
'unwanted',
'wa',
'un',
'running',
',',
'low',
'l',
]
snake_case__ : Any = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['vocab_file'] )
with open(self.vocab_file, 'w', encoding='utf-8' ) as vocab_writer:
vocab_writer.write(''.join([x + '\n' for x in vocab_tokens] ) )
def lowercase_ ( self : Union[str, Any], **_snake_case : List[Any] ) ->Dict:
snake_case__ : str = True
return TransfoXLTokenizer.from_pretrained(self.tmpdirname, **_snake_case )
def lowercase_ ( self : Optional[Any], _snake_case : str ) ->Dict:
snake_case__ : List[Any] = '<unk> UNwanted , running'
snake_case__ : List[Any] = '<unk> unwanted, running'
return input_text, output_text
def lowercase_ ( self : List[Any] ) ->Tuple:
snake_case__ : Dict = TransfoXLTokenizer(vocab_file=self.vocab_file, lower_case=_snake_case )
snake_case__ : str = tokenizer.tokenize('<unk> UNwanted , running' )
self.assertListEqual(_snake_case, ['<unk>', 'unwanted', ',', 'running'] )
self.assertListEqual(tokenizer.convert_tokens_to_ids(_snake_case ), [0, 4, 8, 7] )
def lowercase_ ( self : List[str] ) ->List[Any]:
snake_case__ : str = TransfoXLTokenizer(lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHeLLo ! how \n Are yoU ? ' ), ['hello', '!', 'how', 'are', 'you', '?'] )
def lowercase_ ( self : Optional[int] ) ->Optional[Any]:
snake_case__ : Optional[int] = TransfoXLTokenizer(lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHeLLo ! how \n Are yoU ? ' ), ['HeLLo', '!', 'how', 'Are', 'yoU', '?'] )
def lowercase_ ( self : Optional[int] ) ->Union[str, Any]:
snake_case__ : List[Any] = TransfoXLTokenizer(lower_case=_snake_case )
snake_case__ : Dict = 'Hello (bracket) and side-scrolled [and] Henry\'s $5,000 with 3.34 m. What\'s up!?'
snake_case__ : List[Any] = [
'Hello',
'(',
'bracket',
')',
'and',
'side',
'@-@',
'scrolled',
'[',
'and',
']',
'Henry',
'\'s',
'$',
'5',
'@,@',
'000',
'with',
'3',
'@.@',
'34',
'm',
'.',
'What',
'\'s',
'up',
'!',
'?',
]
self.assertListEqual(tokenizer.tokenize(_snake_case ), _snake_case )
self.assertEqual(tokenizer.convert_tokens_to_string(_snake_case ), _snake_case )
def lowercase_ ( self : Dict ) ->Any:
snake_case__ : Dict = self.get_tokenizer()
snake_case__ : Optional[Any] = len(_snake_case )
tokenizer.add_tokens(['new1', 'new2'] )
tokenizer.move_added_token('new1', 1 )
# Check that moved token is not copied (duplicate)
self.assertEqual(len(_snake_case ), original_len + 2 )
# Check that token is moved to specified id
self.assertEqual(tokenizer.encode('new1' ), [1] )
self.assertEqual(tokenizer.decode([1] ), 'new1' )
| 277 | 0 |
import unittest
from parameterized import parameterized
from transformers import AutoTokenizer, GPTNeoXConfig, is_torch_available, set_seed
from transformers.testing_utils import require_torch, slow, torch_device
from ...generation.test_utils import GenerationTesterMixin
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers import (
GPTNeoXForCausalLM,
GPTNeoXForQuestionAnswering,
GPTNeoXForSequenceClassification,
GPTNeoXForTokenClassification,
GPTNeoXModel,
)
class lowercase__:
"""simple docstring"""
def __init__( self : int , SCREAMING_SNAKE_CASE_ : str , SCREAMING_SNAKE_CASE_ : List[str]=1_3 , SCREAMING_SNAKE_CASE_ : Union[str, Any]=7 , SCREAMING_SNAKE_CASE_ : Tuple=True , SCREAMING_SNAKE_CASE_ : str=True , SCREAMING_SNAKE_CASE_ : List[Any]=True , SCREAMING_SNAKE_CASE_ : List[Any]=True , SCREAMING_SNAKE_CASE_ : List[Any]=9_9 , SCREAMING_SNAKE_CASE_ : Any=6_4 , SCREAMING_SNAKE_CASE_ : Optional[Any]=5 , SCREAMING_SNAKE_CASE_ : Any=4 , SCREAMING_SNAKE_CASE_ : Optional[int]=3_7 , SCREAMING_SNAKE_CASE_ : Tuple="gelu" , SCREAMING_SNAKE_CASE_ : Any=0.1 , SCREAMING_SNAKE_CASE_ : Optional[Any]=0.1 , SCREAMING_SNAKE_CASE_ : Any=5_1_2 , SCREAMING_SNAKE_CASE_ : Optional[Any]=1_6 , SCREAMING_SNAKE_CASE_ : List[str]=2 , SCREAMING_SNAKE_CASE_ : Dict=0.02 , SCREAMING_SNAKE_CASE_ : Dict=3 , SCREAMING_SNAKE_CASE_ : int=4 , SCREAMING_SNAKE_CASE_ : Optional[int]=None , ) -> Any:
lowercase_ = parent
lowercase_ = batch_size
lowercase_ = seq_length
lowercase_ = is_training
lowercase_ = use_input_mask
lowercase_ = use_token_type_ids
lowercase_ = use_labels
lowercase_ = vocab_size
lowercase_ = hidden_size
lowercase_ = num_hidden_layers
lowercase_ = num_attention_heads
lowercase_ = intermediate_size
lowercase_ = hidden_act
lowercase_ = hidden_dropout_prob
lowercase_ = attention_probs_dropout_prob
lowercase_ = max_position_embeddings
lowercase_ = type_vocab_size
lowercase_ = type_sequence_label_size
lowercase_ = initializer_range
lowercase_ = num_labels
lowercase_ = num_choices
lowercase_ = scope
lowercase_ = vocab_size - 1
def _lowercase ( self : int ) -> int:
lowercase_ = ids_tensor([self.batch_size, self.seq_length] , self.vocab_size )
lowercase_ = None
if self.use_input_mask:
lowercase_ = random_attention_mask([self.batch_size, self.seq_length] )
lowercase_ = None
if self.use_labels:
lowercase_ = ids_tensor([self.batch_size, self.seq_length] , self.num_labels )
lowercase_ = self.get_config()
return config, input_ids, input_mask, token_labels
def _lowercase ( self : Optional[int] ) -> Optional[Any]:
return GPTNeoXConfig(
vocab_size=self.vocab_size , hidden_size=self.hidden_size , num_hidden_layers=self.num_hidden_layers , num_attention_heads=self.num_attention_heads , intermediate_size=self.intermediate_size , hidden_act=self.hidden_act , hidden_dropout_prob=self.hidden_dropout_prob , attention_probs_dropout_prob=self.attention_probs_dropout_prob , max_position_embeddings=self.max_position_embeddings , type_vocab_size=self.type_vocab_size , is_decoder=_snake_case , initializer_range=self.initializer_range , pad_token_id=self.pad_token_id , )
def _lowercase ( self : Optional[Any] ) -> Optional[Any]:
lowercase_ = self.prepare_config_and_inputs()
lowercase_ = True
return config, input_ids, input_mask, token_labels
def _lowercase ( self : Any , SCREAMING_SNAKE_CASE_ : List[str] , SCREAMING_SNAKE_CASE_ : Optional[Any] , SCREAMING_SNAKE_CASE_ : Optional[Any] ) -> List[str]:
lowercase_ = GPTNeoXModel(config=_snake_case )
model.to(_snake_case )
model.eval()
lowercase_ = model(_snake_case , attention_mask=_snake_case )
lowercase_ = model(_snake_case )
self.parent.assertEqual(result.last_hidden_state.shape , (self.batch_size, self.seq_length, self.hidden_size) )
def _lowercase ( self : Union[str, Any] , SCREAMING_SNAKE_CASE_ : Optional[Any] , SCREAMING_SNAKE_CASE_ : Tuple , SCREAMING_SNAKE_CASE_ : List[str] ) -> Optional[Any]:
lowercase_ = True
lowercase_ = GPTNeoXModel(_snake_case )
model.to(_snake_case )
model.eval()
lowercase_ = model(_snake_case , attention_mask=_snake_case )
self.parent.assertEqual(result.last_hidden_state.shape , (self.batch_size, self.seq_length, self.hidden_size) )
def _lowercase ( self : int , SCREAMING_SNAKE_CASE_ : int , SCREAMING_SNAKE_CASE_ : List[Any] , SCREAMING_SNAKE_CASE_ : List[str] , SCREAMING_SNAKE_CASE_ : List[str] ) -> Tuple:
lowercase_ = GPTNeoXForCausalLM(config=_snake_case )
model.to(_snake_case )
model.eval()
lowercase_ = model(_snake_case , attention_mask=_snake_case , labels=_snake_case )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.seq_length, self.vocab_size) )
def _lowercase ( self : Optional[int] , SCREAMING_SNAKE_CASE_ : List[str] , SCREAMING_SNAKE_CASE_ : Union[str, Any] , SCREAMING_SNAKE_CASE_ : Optional[Any] , SCREAMING_SNAKE_CASE_ : Any ) -> List[str]:
lowercase_ = self.num_labels
lowercase_ = GPTNeoXForQuestionAnswering(_snake_case )
model.to(_snake_case )
model.eval()
lowercase_ = model(_snake_case , attention_mask=_snake_case )
self.parent.assertEqual(result.start_logits.shape , (self.batch_size, self.seq_length) )
self.parent.assertEqual(result.end_logits.shape , (self.batch_size, self.seq_length) )
def _lowercase ( self : Any , SCREAMING_SNAKE_CASE_ : Any , SCREAMING_SNAKE_CASE_ : Dict , SCREAMING_SNAKE_CASE_ : Optional[Any] , SCREAMING_SNAKE_CASE_ : str ) -> Union[str, Any]:
lowercase_ = self.num_labels
lowercase_ = GPTNeoXForSequenceClassification(_snake_case )
model.to(_snake_case )
model.eval()
lowercase_ = ids_tensor([self.batch_size] , self.type_sequence_label_size )
lowercase_ = model(_snake_case , attention_mask=_snake_case , labels=_snake_case )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.num_labels) )
def _lowercase ( self : Dict , SCREAMING_SNAKE_CASE_ : str , SCREAMING_SNAKE_CASE_ : Optional[Any] , SCREAMING_SNAKE_CASE_ : Optional[int] , SCREAMING_SNAKE_CASE_ : List[Any] ) -> str:
lowercase_ = self.num_labels
lowercase_ = GPTNeoXForTokenClassification(_snake_case )
model.to(_snake_case )
model.eval()
lowercase_ = model(_snake_case , attention_mask=_snake_case , labels=_snake_case )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.seq_length, self.num_labels) )
def _lowercase ( self : Any , SCREAMING_SNAKE_CASE_ : List[str] , SCREAMING_SNAKE_CASE_ : str , SCREAMING_SNAKE_CASE_ : int ) -> str:
lowercase_ = True
lowercase_ = GPTNeoXForCausalLM(config=_snake_case )
model.to(_snake_case )
model.eval()
# first forward pass
lowercase_ = model(_snake_case , attention_mask=_snake_case , use_cache=_snake_case )
lowercase_ = outputs.past_key_values
# create hypothetical multiple next token and extent to next_input_ids
lowercase_ = ids_tensor((self.batch_size, 3) , config.vocab_size )
lowercase_ = ids_tensor((self.batch_size, 3) , vocab_size=2 )
# append to next input_ids and
lowercase_ = torch.cat([input_ids, next_tokens] , dim=-1 )
lowercase_ = torch.cat([input_mask, next_mask] , dim=-1 )
lowercase_ = model(_snake_case , attention_mask=_snake_case , output_hidden_states=_snake_case )
lowercase_ = output_from_no_past['hidden_states'][0]
lowercase_ = model(
_snake_case , attention_mask=_snake_case , past_key_values=_snake_case , output_hidden_states=_snake_case , )['hidden_states'][0]
# select random slice
lowercase_ = ids_tensor((1,) , output_from_past.shape[-1] ).item()
lowercase_ = output_from_no_past[:, -3:, random_slice_idx].detach()
lowercase_ = output_from_past[:, :, random_slice_idx].detach()
self.parent.assertTrue(output_from_past_slice.shape[1] == next_tokens.shape[1] )
# test that outputs are equal for slice
self.parent.assertTrue(torch.allclose(_snake_case , _snake_case , atol=1e-3 ) )
def _lowercase ( self : str ) -> Union[str, Any]:
lowercase_ = self.prepare_config_and_inputs()
lowercase_ = config_and_inputs
lowercase_ = {'input_ids': input_ids, 'attention_mask': input_mask}
return config, inputs_dict
@require_torch
class lowercase__( lowerCAmelCase_ , lowerCAmelCase_ , lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
a :int = (
(
GPTNeoXModel,
GPTNeoXForCausalLM,
GPTNeoXForQuestionAnswering,
GPTNeoXForSequenceClassification,
GPTNeoXForTokenClassification,
)
if is_torch_available()
else ()
)
a :str = (GPTNeoXForCausalLM,) if is_torch_available() else ()
a :Dict = (
{
'feature-extraction': GPTNeoXModel,
'question-answering': GPTNeoXForQuestionAnswering,
'text-classification': GPTNeoXForSequenceClassification,
'text-generation': GPTNeoXForCausalLM,
'token-classification': GPTNeoXForTokenClassification,
'zero-shot': GPTNeoXForSequenceClassification,
}
if is_torch_available()
else {}
)
a :Any = False
a :int = False
a :Any = False
a :Dict = False
def _lowercase ( self : Optional[int] ) -> List[Any]:
lowercase_ = GPTNeoXModelTester(self )
lowercase_ = ConfigTester(self , config_class=_snake_case , hidden_size=6_4 , num_attention_heads=8 )
def _lowercase ( self : List[str] ) -> Any:
self.config_tester.run_common_tests()
def _lowercase ( self : Union[str, Any] ) -> List[str]:
lowercase_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(_snake_case , _snake_case , _snake_case )
def _lowercase ( self : Tuple ) -> int:
lowercase_ = self.model_tester.prepare_config_and_inputs_for_decoder()
self.model_tester.create_and_check_model_as_decoder(_snake_case , _snake_case , _snake_case )
def _lowercase ( self : Optional[Any] ) -> str:
# This regression test was failing with PyTorch < 1.3
lowercase_ = self.model_tester.prepare_config_and_inputs_for_decoder()
lowercase_ = None
self.model_tester.create_and_check_model_as_decoder(_snake_case , _snake_case , _snake_case )
def _lowercase ( self : Union[str, Any] ) -> Tuple:
lowercase_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_decoder_model_past_large_inputs(_snake_case , _snake_case , _snake_case )
def _lowercase ( self : Any ) -> List[Any]:
lowercase_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_causal_lm(*_snake_case )
def _lowercase ( self : Tuple ) -> List[str]:
lowercase_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_question_answering(*_snake_case )
def _lowercase ( self : Tuple ) -> Dict:
lowercase_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_sequence_classification(*_snake_case )
def _lowercase ( self : List[str] ) -> int:
lowercase_ = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_token_classification(*_snake_case )
@unittest.skip(reason='''Feed forward chunking is not implemented''' )
def _lowercase ( self : Any ) -> List[str]:
pass
@parameterized.expand([('''linear''',), ('''dynamic''',)] )
def _lowercase ( self : List[str] , SCREAMING_SNAKE_CASE_ : Tuple ) -> List[Any]:
lowercase_ = self.model_tester.prepare_config_and_inputs_for_common()
lowercase_ = ids_tensor([1, 1_0] , config.vocab_size )
lowercase_ = ids_tensor([1, int(config.max_position_embeddings * 1.5 )] , config.vocab_size )
set_seed(4_2 ) # Fixed seed at init time so the two models get the same random weights
lowercase_ = GPTNeoXModel(_snake_case )
original_model.to(_snake_case )
original_model.eval()
lowercase_ = original_model(_snake_case ).last_hidden_state
lowercase_ = original_model(_snake_case ).last_hidden_state
set_seed(4_2 ) # Fixed seed at init time so the two models get the same random weights
lowercase_ = {'type': scaling_type, 'factor': 1_0.0}
lowercase_ = GPTNeoXModel(_snake_case )
scaled_model.to(_snake_case )
scaled_model.eval()
lowercase_ = scaled_model(_snake_case ).last_hidden_state
lowercase_ = scaled_model(_snake_case ).last_hidden_state
# Dynamic scaling does not change the RoPE embeddings until it receives an input longer than the original
# maximum sequence length, so the outputs for the short input should match.
if scaling_type == "dynamic":
self.assertTrue(torch.allclose(_snake_case , _snake_case , atol=1e-5 ) )
else:
self.assertFalse(torch.allclose(_snake_case , _snake_case , atol=1e-5 ) )
# The output should be different for long inputs
self.assertFalse(torch.allclose(_snake_case , _snake_case , atol=1e-5 ) )
@require_torch
class lowercase__( unittest.TestCase ):
"""simple docstring"""
@slow
def _lowercase ( self : List[Any] ) -> Union[str, Any]:
lowercase_ = AutoTokenizer.from_pretrained('''EleutherAI/pythia-410m-deduped''' )
for checkpointing in [True, False]:
lowercase_ = GPTNeoXForCausalLM.from_pretrained('''EleutherAI/pythia-410m-deduped''' )
if checkpointing:
model.gradient_checkpointing_enable()
else:
model.gradient_checkpointing_disable()
model.to(_snake_case )
lowercase_ = tokenizer('''My favorite food is''' , return_tensors='''pt''' ).to(_snake_case )
# The hub repo. is updated on 2023-04-04, resulting in poor outputs.
# See: https://github.com/huggingface/transformers/pull/24193
lowercase_ = 'My favorite food is a good old-fashioned, old-fashioned, old-fashioned.\n\nI\'m not sure'
lowercase_ = model.generate(**_snake_case , do_sample=_snake_case , max_new_tokens=2_0 )
lowercase_ = tokenizer.batch_decode(_snake_case )[0]
self.assertEqual(_snake_case , _snake_case )
| 30 |
from ...configuration_utils import PretrainedConfig
from ...utils import logging
a_ :Optional[int] = logging.get_logger(__name__)
a_ :Dict = {"openai-gpt": "https://huggingface.co/openai-gpt/resolve/main/config.json"}
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """openai-gpt"""
_SCREAMING_SNAKE_CASE = {
"""max_position_embeddings""": """n_positions""",
"""hidden_size""": """n_embd""",
"""num_attention_heads""": """n_head""",
"""num_hidden_layers""": """n_layer""",
}
def __init__( self : Optional[int], _snake_case : Dict=4_0_4_7_8, _snake_case : str=5_1_2, _snake_case : int=7_6_8, _snake_case : Tuple=1_2, _snake_case : Any=1_2, _snake_case : str="gelu", _snake_case : List[str]=0.1, _snake_case : Any=0.1, _snake_case : Dict=0.1, _snake_case : int=1e-5, _snake_case : Optional[Any]=0.0_2, _snake_case : List[Any]="cls_index", _snake_case : Any=True, _snake_case : Any=None, _snake_case : int=True, _snake_case : Optional[Any]=0.1, **_snake_case : List[Any], ) ->Optional[int]:
snake_case__ : int = vocab_size
snake_case__ : Dict = n_positions
snake_case__ : str = n_embd
snake_case__ : str = n_layer
snake_case__ : List[Any] = n_head
snake_case__ : List[Any] = afn
snake_case__ : Optional[Any] = resid_pdrop
snake_case__ : List[str] = embd_pdrop
snake_case__ : List[Any] = attn_pdrop
snake_case__ : Optional[int] = layer_norm_epsilon
snake_case__ : str = initializer_range
snake_case__ : List[str] = summary_type
snake_case__ : Optional[int] = summary_use_proj
snake_case__ : List[str] = summary_activation
snake_case__ : Optional[Any] = summary_first_dropout
snake_case__ : int = summary_proj_to_labels
super().__init__(**_snake_case )
| 277 | 0 |
"""simple docstring"""
import math
import time
from typing import Dict, List, Optional
from torch.utils.data import Dataset
from transformers import SeqaSeqTrainer, is_torch_tpu_available
from transformers.trainer_utils import PredictionOutput, speed_metrics
if is_torch_tpu_available(check_device=False):
import torch_xla.core.xla_model as xm
import torch_xla.debug.metrics as met
class __lowerCAmelCase ( lowerCAmelCase_ ):
'''simple docstring'''
def __init__( self , *_a , _a=None , _a=None , **_a ):
super().__init__(*_snake_case , **_snake_case )
__a = eval_examples
__a = post_process_function
def __UpperCAmelCase ( self , _a = None , _a=None , _a = None , _a = "eval" , **_a , ):
__a = gen_kwargs.copy()
__a = (
gen_kwargs['max_length'] if gen_kwargs.get('''max_length''' ) is not None else self.args.generation_max_length
)
__a = (
gen_kwargs['num_beams'] if gen_kwargs.get('''num_beams''' ) is not None else self.args.generation_num_beams
)
__a = gen_kwargs
__a = self.eval_dataset if eval_dataset is None else eval_dataset
__a = self.get_eval_dataloader(_snake_case )
__a = self.eval_examples if eval_examples is None else eval_examples
# Temporarily disable metric computation, we will do it in the loop here.
__a = self.compute_metrics
__a = None
__a = time.time()
__a = self.prediction_loop if self.args.use_legacy_prediction_loop else self.evaluation_loop
try:
__a = eval_loop(
_snake_case , description='''Evaluation''' , prediction_loss_only=True if compute_metrics is None else None , ignore_keys=_snake_case , metric_key_prefix=_snake_case , )
finally:
__a = compute_metrics
__a = self.args.eval_batch_size * self.args.world_size
if f'''{metric_key_prefix}_jit_compilation_time''' in output.metrics:
start_time += output.metrics[f'''{metric_key_prefix}_jit_compilation_time''']
output.metrics.update(
speed_metrics(
_snake_case , _snake_case , num_samples=output.num_samples , num_steps=math.ceil(output.num_samples / total_batch_size ) , ) )
if self.post_process_function is not None and self.compute_metrics is not None and self.args.should_save:
# Only the main node write the results by default
__a = self.post_process_function(_snake_case , _snake_case , _snake_case )
__a = self.compute_metrics(_snake_case )
# Prefix all keys with metric_key_prefix + '_'
for key in list(metrics.keys() ):
if not key.startswith(f'''{metric_key_prefix}_''' ):
__a = metrics.pop(_snake_case )
metrics.update(output.metrics )
else:
__a = output.metrics
if self.args.should_log:
# Only the main node log the results by default
self.log(_snake_case )
if self.args.tpu_metrics_debug or self.args.debug:
# tpu-comment: Logging debug metrics for PyTorch/XLA (compile, execute times, ops, etc.)
xm.master_print(met.metrics_report() )
__a = self.callback_handler.on_evaluate(self.args , self.state , self.control , _snake_case )
return metrics
def __UpperCAmelCase ( self , _a , _a , _a=None , _a = "test" , **_a ):
__a = gen_kwargs.copy()
__a = self.get_test_dataloader(_snake_case )
# Temporarily disable metric computation, we will do it in the loop here.
__a = self.compute_metrics
__a = None
__a = time.time()
__a = self.prediction_loop if self.args.use_legacy_prediction_loop else self.evaluation_loop
try:
__a = eval_loop(
_snake_case , description='''Prediction''' , prediction_loss_only=True if compute_metrics is None else None , ignore_keys=_snake_case , metric_key_prefix=_snake_case , )
finally:
__a = compute_metrics
__a = self.args.eval_batch_size * self.args.world_size
if f'''{metric_key_prefix}_jit_compilation_time''' in output.metrics:
start_time += output.metrics[f'''{metric_key_prefix}_jit_compilation_time''']
output.metrics.update(
speed_metrics(
_snake_case , _snake_case , num_samples=output.num_samples , num_steps=math.ceil(output.num_samples / total_batch_size ) , ) )
if self.post_process_function is None or self.compute_metrics is None:
return output
__a = self.post_process_function(_snake_case , _snake_case , _snake_case , '''predict''' )
__a = self.compute_metrics(_snake_case )
# Prefix all keys with metric_key_prefix + '_'
for key in list(metrics.keys() ):
if not key.startswith(f'''{metric_key_prefix}_''' ):
__a = metrics.pop(_snake_case )
metrics.update(output.metrics )
return PredictionOutput(predictions=predictions.predictions , label_ids=predictions.label_ids , metrics=_snake_case )
| 45 |
import argparse
import logging
import os
from datetime import datetime
import numpy as np
import torch
from torch import nn
from torch.utils.data import DataLoader, RandomSampler, TensorDataset
from tqdm import tqdm
from transformers import GPTaLMHeadModel
a_ :Optional[Any] = logging.getLogger(__name__)
def lowercase_ (A : List[Any] , A : List[Any] ):
# save results
if os.path.exists(A ):
if os.path.exists(os.path.join(A , 'config.json' ) ) and os.path.isfile(
os.path.join(A , 'config.json' ) ):
os.remove(os.path.join(A , 'config.json' ) )
if os.path.exists(os.path.join(A , 'pytorch_model.bin' ) ) and os.path.isfile(
os.path.join(A , 'pytorch_model.bin' ) ):
os.remove(os.path.join(A , 'pytorch_model.bin' ) )
else:
os.makedirs(A )
model.save_pretrained(A )
def lowercase_ (A : Any , A : Optional[Any]=False ):
snake_case__ : str = 2
if unlogit:
snake_case__ : Dict = torch.pow(A , A )
snake_case__ : Any = p * torch.log(A )
snake_case__ : Tuple = 0
return -plogp.sum(dim=-1 )
def lowercase_ (A : List[str] ):
logger.info('lv, h >\t' + '\t'.join(F'''{x + 1}''' for x in range(len(A ) ) ) )
for row in range(len(A ) ):
if tensor.dtype != torch.long:
logger.info(F'''layer {row + 1}:\t''' + '\t'.join(F'''{x:.5f}''' for x in tensor[row].cpu().data ) )
else:
logger.info(F'''layer {row + 1}:\t''' + '\t'.join(F'''{x:d}''' for x in tensor[row].cpu().data ) )
def lowercase_ (A : Tuple , A : Optional[Any] , A : str , A : int=True , A : Optional[int]=True , A : Any=None , A : int=False ):
snake_case__ , snake_case__ : Optional[Any] = model.config.num_hidden_layers, model.config.num_attention_heads
snake_case__ : int = torch.zeros(A , A ).to(args.device )
snake_case__ : Any = torch.zeros(A , A ).to(args.device )
if head_mask is None:
snake_case__ : Dict = torch.ones(A , A ).to(args.device )
head_mask.requires_grad_(requires_grad=A )
# If actually pruned attention multi-head, set head mask to None to avoid shape mismatch
if actually_pruned:
snake_case__ : Optional[int] = None
snake_case__ : List[Any] = 0.0
snake_case__ : str = 0.0
for step, inputs in enumerate(tqdm(A , desc='Iteration' , disable=args.local_rank not in [-1, 0] ) ):
snake_case__ : Union[str, Any] = tuple(t.to(args.device ) for t in inputs )
((snake_case__) , ) : Optional[Any] = inputs
# Do a forward pass (not with torch.no_grad() since we need gradients for importance score - see below)
snake_case__ : Union[str, Any] = model(A , labels=A , head_mask=A )
# (loss), lm_logits, presents, (all hidden_states), (attentions)
snake_case__ , snake_case__ , snake_case__ : Dict = (
outputs[0],
outputs[1],
outputs[-1],
) # Loss and logits are the first, attention the last
loss.backward() # Backpropagate to populate the gradients in the head mask
total_loss += loss.detach().cpu().numpy()
if compute_entropy:
for layer, attn in enumerate(A ):
snake_case__ : Optional[Any] = entropy(attn.detach() , A )
attn_entropy[layer] += masked_entropy.sum(-1 ).sum(0 ).sum(0 ).detach()
if compute_importance:
head_importance += head_mask.grad.abs().detach()
tot_tokens += torch.ones_like(A ).float().detach().sum().data
# Normalize
attn_entropy /= tot_tokens
head_importance /= tot_tokens
# Layerwise importance normalization
if not args.dont_normalize_importance_by_layer:
snake_case__ : Union[str, Any] = 2
snake_case__ : List[Any] = torch.pow(torch.pow(A , A ).sum(-1 ) , 1 / exponent )
head_importance /= norm_by_layer.unsqueeze(-1 ) + 1e-20
if not args.dont_normalize_global_importance:
snake_case__ : Tuple = (head_importance - head_importance.min()) / (head_importance.max() - head_importance.min())
# Print matrices
if compute_entropy:
logger.info('Attention entropies' )
print_ad_tensor(A )
if compute_importance:
logger.info('Head importance scores' )
print_ad_tensor(A )
logger.info('Head ranked by importance scores' )
snake_case__ : Tuple = torch.zeros(head_importance.numel() , dtype=torch.long , device=args.device )
snake_case__ : Union[str, Any] = torch.arange(
head_importance.numel() , device=args.device )
snake_case__ : str = head_ranks.view_as(A )
print_ad_tensor(A )
return attn_entropy, head_importance, total_loss
def lowercase_ (A : Optional[int] , A : Dict , A : Optional[int] ):
snake_case__ , snake_case__ , snake_case__ : Any = compute_heads_importance(A , A , A , compute_entropy=A )
snake_case__ : Tuple = 1 / loss # instead of downsteam score use the LM loss
logger.info('Pruning: original score: %f, threshold: %f' , A , original_score * args.masking_threshold )
snake_case__ : Optional[Any] = torch.ones_like(A )
snake_case__ : Union[str, Any] = max(1 , int(new_head_mask.numel() * args.masking_amount ) )
snake_case__ : Dict = original_score
while current_score >= original_score * args.masking_threshold:
snake_case__ : int = new_head_mask.clone().detach() # save current head mask
# heads from least important to most - keep only not-masked heads
snake_case__ : List[Any] = float('Inf' )
snake_case__ : Union[str, Any] = head_importance.view(-1 ).sort()[1]
if len(A ) <= num_to_mask:
print('BREAK BY num_to_mask' )
break
# mask heads
snake_case__ : int = current_heads_to_mask[:num_to_mask]
logger.info('Heads to mask: %s' , str(current_heads_to_mask.tolist() ) )
snake_case__ : int = new_head_mask.view(-1 )
snake_case__ : int = 0.0
snake_case__ : Union[str, Any] = new_head_mask.view_as(A )
snake_case__ : List[str] = new_head_mask.clone().detach()
print_ad_tensor(A )
# Compute metric and head importance again
snake_case__ , snake_case__ , snake_case__ : Any = compute_heads_importance(
A , A , A , compute_entropy=A , head_mask=A )
snake_case__ : Dict = 1 / loss
logger.info(
'Masking: current score: %f, remaining heads %d (%.1f percents)' , A , new_head_mask.sum() , new_head_mask.sum() / new_head_mask.numel() * 1_0_0 , )
logger.info('Final head mask' )
print_ad_tensor(A )
np.save(os.path.join(args.output_dir , 'head_mask.npy' ) , head_mask.detach().cpu().numpy() )
return head_mask
def lowercase_ (A : List[str] , A : Tuple , A : Optional[Any] , A : int ):
snake_case__ : Any = datetime.now()
snake_case__ , snake_case__ , snake_case__ : str = compute_heads_importance(
A , A , A , compute_entropy=A , compute_importance=A , head_mask=A )
snake_case__ : Tuple = 1 / loss
snake_case__ : Dict = datetime.now() - before_time
snake_case__ : Union[str, Any] = sum(p.numel() for p in model.parameters() )
snake_case__ : Optional[Any] = {
layer: (1 - head_mask[layer].long()).nonzero().squeeze().tolist() for layer in range(len(A ) )
}
for k, v in heads_to_prune.items():
if isinstance(A , A ):
snake_case__ : Any = [
v,
]
assert sum(len(A ) for h in heads_to_prune.values() ) == (1 - head_mask.long()).sum().item()
model.prune_heads(A )
snake_case__ : Dict = sum(p.numel() for p in model.parameters() )
snake_case__ : Tuple = datetime.now()
snake_case__ , snake_case__ , snake_case__ : Dict = compute_heads_importance(
A , A , A , compute_entropy=A , compute_importance=A , head_mask=A , actually_pruned=A , )
snake_case__ : Any = 1 / loss
snake_case__ : int = datetime.now() - before_time
logger.info(
'Pruning: original num of params: %.2e, after pruning %.2e (%.1f percents)' , A , A , pruned_num_params / original_num_params * 1_0_0 , )
logger.info('Pruning: score with masking: %f score with pruning: %f' , A , A )
logger.info('Pruning: speed ratio (original timing / new timing): %f percents' , original_time / new_time * 1_0_0 )
save_model(A , args.output_dir )
def lowercase_ ():
snake_case__ : str = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
'--data_dir' , default=A , type=A , required=A , help='The input data dir. Should contain the .tsv files (or other data files) for the task.' , )
parser.add_argument(
'--model_name_or_path' , default=A , type=A , required=A , help='Path to pretrained model or model identifier from huggingface.co/models' , )
parser.add_argument(
'--output_dir' , default=A , type=A , required=A , help='The output directory where the model predictions and checkpoints will be written.' , )
# Other parameters
parser.add_argument(
'--config_name' , default='' , type=A , help='Pretrained config name or path if not the same as model_name_or_path' , )
parser.add_argument(
'--tokenizer_name' , default='' , type=A , help='Pretrained tokenizer name or path if not the same as model_name_or_path' , )
parser.add_argument(
'--cache_dir' , default=A , type=A , help='Where do you want to store the pre-trained models downloaded from s3' , )
parser.add_argument(
'--data_subset' , type=A , default=-1 , help='If > 0: limit the data to a subset of data_subset instances.' )
parser.add_argument(
'--overwrite_output_dir' , action='store_true' , help='Whether to overwrite data in output directory' )
parser.add_argument(
'--overwrite_cache' , action='store_true' , help='Overwrite the cached training and evaluation sets' )
parser.add_argument(
'--dont_normalize_importance_by_layer' , action='store_true' , help='Don\'t normalize importance score by layers' )
parser.add_argument(
'--dont_normalize_global_importance' , action='store_true' , help='Don\'t normalize all importance scores between 0 and 1' , )
parser.add_argument(
'--try_masking' , action='store_true' , help='Whether to try to mask head until a threshold of accuracy.' )
parser.add_argument(
'--masking_threshold' , default=0.9 , type=A , help='masking threshold in term of metrics (stop masking when metric < threshold * original metric value).' , )
parser.add_argument(
'--masking_amount' , default=0.1 , type=A , help='Amount to heads to masking at each masking step.' )
parser.add_argument('--metric_name' , default='acc' , type=A , help='Metric to use for head masking.' )
parser.add_argument(
'--max_seq_length' , default=1_2_8 , type=A , help=(
'The maximum total input sequence length after WordPiece tokenization. \n'
'Sequences longer than this will be truncated, sequences shorter padded.'
) , )
parser.add_argument('--batch_size' , default=1 , type=A , help='Batch size.' )
parser.add_argument('--seed' , type=A , default=4_2 )
parser.add_argument('--local_rank' , type=A , default=-1 , help='local_rank for distributed training on gpus' )
parser.add_argument('--no_cuda' , action='store_true' , help='Whether not to use CUDA when available' )
parser.add_argument('--server_ip' , type=A , default='' , help='Can be used for distant debugging.' )
parser.add_argument('--server_port' , type=A , default='' , help='Can be used for distant debugging.' )
snake_case__ : Optional[int] = parser.parse_args()
if args.server_ip and args.server_port:
# Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script
import ptvsd
print('Waiting for debugger attach' )
ptvsd.enable_attach(address=(args.server_ip, args.server_port) , redirect_output=A )
ptvsd.wait_for_attach()
# Setup devices and distributed training
if args.local_rank == -1 or args.no_cuda:
snake_case__ : List[Any] = torch.device('cuda' if torch.cuda.is_available() and not args.no_cuda else 'cpu' )
snake_case__ : Optional[Any] = 0 if args.no_cuda else torch.cuda.device_count()
else:
torch.cuda.set_device(args.local_rank )
snake_case__ : int = torch.device('cuda' , args.local_rank )
snake_case__ : List[str] = 1
torch.distributed.init_process_group(backend='nccl' ) # Initializes the distributed backend
# Setup logging
logging.basicConfig(level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN )
logger.info('device: {} n_gpu: {}, distributed: {}'.format(args.device , args.n_gpu , bool(args.local_rank != -1 ) ) )
snake_case__ : Any = GPTaLMHeadModel.from_pretrained(args.model_name_or_path )
# Distributed and parallel training
model.to(args.device )
if args.local_rank != -1:
snake_case__ : List[str] = nn.parallel.DistributedDataParallel(
A , device_ids=[args.local_rank] , output_device=args.local_rank , find_unused_parameters=A )
elif args.n_gpu > 1:
snake_case__ : Optional[int] = nn.DataParallel(A )
# Print/save training arguments
os.makedirs(args.output_dir , exist_ok=A )
torch.save(A , os.path.join(args.output_dir , 'run_args.bin' ) )
logger.info('Training/evaluation parameters %s' , A )
# Prepare dataset
snake_case__ : Optional[Any] = np.concatenate(
[
np.loadtxt(args.data_dir , dtype=np.intaa ),
] )
snake_case__ : List[str] = (torch.from_numpy(A ),)
snake_case__ : int = TensorDataset(*A )
snake_case__ : Union[str, Any] = RandomSampler(A )
snake_case__ : Any = DataLoader(A , sampler=A , batch_size=args.batch_size )
# Compute head entropy and importance score
compute_heads_importance(A , A , A )
# Try head masking (set heads to zero until the score goes under a threshole)
# and head pruning (remove masked heads and see the effect on the network)
if args.try_masking and args.masking_threshold > 0.0 and args.masking_threshold < 1.0:
snake_case__ : Dict = mask_heads(A , A , A )
prune_heads(A , A , A , A )
if __name__ == "__main__":
main()
| 277 | 0 |
'''simple docstring'''
from ..utils import DummyObject, requires_backends
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Any:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Any:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> int:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> int:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(cls , ["torch"] )
def __lowerCAmelCase ( *snake_case__ , **snake_case__ ):
requires_backends(snake_case__ , ["torch"] )
def __lowerCAmelCase ( *snake_case__ , **snake_case__ ):
requires_backends(snake_case__ , ["torch"] )
def __lowerCAmelCase ( *snake_case__ , **snake_case__ ):
requires_backends(snake_case__ , ["torch"] )
def __lowerCAmelCase ( *snake_case__ , **snake_case__ ):
requires_backends(snake_case__ , ["torch"] )
def __lowerCAmelCase ( *snake_case__ , **snake_case__ ):
requires_backends(snake_case__ , ["torch"] )
def __lowerCAmelCase ( *snake_case__ , **snake_case__ ):
requires_backends(snake_case__ , ["torch"] )
def __lowerCAmelCase ( *snake_case__ , **snake_case__ ):
requires_backends(snake_case__ , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Any:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Tuple:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Any:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Tuple:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Any:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> int:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> int:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> int:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> int:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> int:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> int:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Tuple:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Tuple:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> int:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Tuple:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> int:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Tuple:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> int:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Tuple:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Tuple:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Any:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> int:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Tuple:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Any:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Tuple:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Tuple:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> int:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Tuple:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> List[str]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[Any]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Any:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Optional[int]:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> str:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> int:
requires_backends(cls , ["torch"] )
class A ( metaclass=lowerCAmelCase_ ):
'''simple docstring'''
A = ["torch"]
def __init__(self , *_UpperCAmelCase , **_UpperCAmelCase ) -> Dict:
requires_backends(self , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Tuple:
requires_backends(cls , ["torch"] )
@classmethod
def a_ (cls , *_UpperCAmelCase , **_UpperCAmelCase ) -> Union[str, Any]:
requires_backends(cls , ["torch"] )
| 298 |
import argparse
import json
from collections import OrderedDict
from pathlib import Path
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import (
SegformerConfig,
SegformerForImageClassification,
SegformerForSemanticSegmentation,
SegformerImageProcessor,
)
from transformers.utils import logging
logging.set_verbosity_info()
a_ :Dict = logging.get_logger(__name__)
def lowercase_ (A : Optional[Any] , A : Any=False ):
snake_case__ : List[Any] = OrderedDict()
for key, value in state_dict.items():
if encoder_only and not key.startswith('head' ):
snake_case__ : str = 'segformer.encoder.' + key
if key.startswith('backbone' ):
snake_case__ : str = key.replace('backbone' , 'segformer.encoder' )
if "patch_embed" in key:
# replace for example patch_embed1 by patch_embeddings.0
snake_case__ : Optional[int] = key[key.find('patch_embed' ) + len('patch_embed' )]
snake_case__ : int = key.replace(F'''patch_embed{idx}''' , F'''patch_embeddings.{int(A )-1}''' )
if "norm" in key:
snake_case__ : Optional[int] = key.replace('norm' , 'layer_norm' )
if "segformer.encoder.layer_norm" in key:
# replace for example layer_norm1 by layer_norm.0
snake_case__ : Tuple = key[key.find('segformer.encoder.layer_norm' ) + len('segformer.encoder.layer_norm' )]
snake_case__ : Union[str, Any] = key.replace(F'''layer_norm{idx}''' , F'''layer_norm.{int(A )-1}''' )
if "layer_norm1" in key:
snake_case__ : List[Any] = key.replace('layer_norm1' , 'layer_norm_1' )
if "layer_norm2" in key:
snake_case__ : List[Any] = key.replace('layer_norm2' , 'layer_norm_2' )
if "block" in key:
# replace for example block1 by block.0
snake_case__ : List[Any] = key[key.find('block' ) + len('block' )]
snake_case__ : List[Any] = key.replace(F'''block{idx}''' , F'''block.{int(A )-1}''' )
if "attn.q" in key:
snake_case__ : int = key.replace('attn.q' , 'attention.self.query' )
if "attn.proj" in key:
snake_case__ : str = key.replace('attn.proj' , 'attention.output.dense' )
if "attn" in key:
snake_case__ : Optional[int] = key.replace('attn' , 'attention.self' )
if "fc1" in key:
snake_case__ : str = key.replace('fc1' , 'dense1' )
if "fc2" in key:
snake_case__ : Dict = key.replace('fc2' , 'dense2' )
if "linear_pred" in key:
snake_case__ : Union[str, Any] = key.replace('linear_pred' , 'classifier' )
if "linear_fuse" in key:
snake_case__ : List[str] = key.replace('linear_fuse.conv' , 'linear_fuse' )
snake_case__ : List[Any] = key.replace('linear_fuse.bn' , 'batch_norm' )
if "linear_c" in key:
# replace for example linear_c4 by linear_c.3
snake_case__ : Optional[int] = key[key.find('linear_c' ) + len('linear_c' )]
snake_case__ : Tuple = key.replace(F'''linear_c{idx}''' , F'''linear_c.{int(A )-1}''' )
if key.startswith('head' ):
snake_case__ : Tuple = key.replace('head' , 'classifier' )
snake_case__ : Optional[int] = value
return new_state_dict
def lowercase_ (A : Tuple , A : Optional[int] ):
# for each of the encoder blocks:
for i in range(config.num_encoder_blocks ):
for j in range(config.depths[i] ):
# read in weights + bias of keys and values (which is a single matrix in the original implementation)
snake_case__ : List[str] = state_dict.pop(F'''segformer.encoder.block.{i}.{j}.attention.self.kv.weight''' )
snake_case__ : Optional[Any] = state_dict.pop(F'''segformer.encoder.block.{i}.{j}.attention.self.kv.bias''' )
# next, add keys and values (in that order) to the state dict
snake_case__ : str = kv_weight[
: config.hidden_sizes[i], :
]
snake_case__ : Dict = kv_bias[: config.hidden_sizes[i]]
snake_case__ : List[str] = kv_weight[
config.hidden_sizes[i] :, :
]
snake_case__ : List[Any] = kv_bias[
config.hidden_sizes[i] :
]
def lowercase_ ():
snake_case__ : Union[str, Any] = 'http://images.cocodataset.org/val2017/000000039769.jpg'
snake_case__ : Dict = Image.open(requests.get(A , stream=A ).raw )
return image
@torch.no_grad()
def lowercase_ (A : Any , A : Union[str, Any] , A : Optional[Any] ):
snake_case__ : List[str] = SegformerConfig()
snake_case__ : Dict = False
# set attributes based on model_name
snake_case__ : Optional[int] = 'huggingface/label-files'
if "segformer" in model_name:
snake_case__ : str = model_name[len('segformer.' ) : len('segformer.' ) + 2]
if "ade" in model_name:
snake_case__ : Optional[int] = 1_5_0
snake_case__ : int = 'ade20k-id2label.json'
snake_case__ : List[Any] = (1, 1_5_0, 1_2_8, 1_2_8)
elif "city" in model_name:
snake_case__ : str = 1_9
snake_case__ : List[str] = 'cityscapes-id2label.json'
snake_case__ : Optional[Any] = (1, 1_9, 1_2_8, 1_2_8)
else:
raise ValueError(F'''Model {model_name} not supported''' )
elif "mit" in model_name:
snake_case__ : str = True
snake_case__ : Union[str, Any] = model_name[4:6]
snake_case__ : Optional[Any] = 1_0_0_0
snake_case__ : Optional[int] = 'imagenet-1k-id2label.json'
snake_case__ : List[Any] = (1, 1_0_0_0)
else:
raise ValueError(F'''Model {model_name} not supported''' )
# set config attributes
snake_case__ : str = json.load(open(hf_hub_download(A , A , repo_type='dataset' ) , 'r' ) )
snake_case__ : List[Any] = {int(A ): v for k, v in idalabel.items()}
snake_case__ : Union[str, Any] = idalabel
snake_case__ : Tuple = {v: k for k, v in idalabel.items()}
if size == "b0":
pass
elif size == "b1":
snake_case__ : List[Any] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : Tuple = 2_5_6
elif size == "b2":
snake_case__ : List[str] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : int = 7_6_8
snake_case__ : List[Any] = [3, 4, 6, 3]
elif size == "b3":
snake_case__ : Optional[Any] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : int = 7_6_8
snake_case__ : Optional[Any] = [3, 4, 1_8, 3]
elif size == "b4":
snake_case__ : str = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : Optional[Any] = 7_6_8
snake_case__ : Union[str, Any] = [3, 8, 2_7, 3]
elif size == "b5":
snake_case__ : List[str] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : Optional[Any] = 7_6_8
snake_case__ : Any = [3, 6, 4_0, 3]
else:
raise ValueError(F'''Size {size} not supported''' )
# load image processor (only resize + normalize)
snake_case__ : Dict = SegformerImageProcessor(
image_scale=(5_1_2, 5_1_2) , keep_ratio=A , align=A , do_random_crop=A )
# prepare image
snake_case__ : List[str] = prepare_img()
snake_case__ : Dict = image_processor(images=A , return_tensors='pt' ).pixel_values
logger.info(F'''Converting model {model_name}...''' )
# load original state dict
if encoder_only:
snake_case__ : Tuple = torch.load(A , map_location=torch.device('cpu' ) )
else:
snake_case__ : int = torch.load(A , map_location=torch.device('cpu' ) )['state_dict']
# rename keys
snake_case__ : List[Any] = rename_keys(A , encoder_only=A )
if not encoder_only:
del state_dict["decode_head.conv_seg.weight"]
del state_dict["decode_head.conv_seg.bias"]
# key and value matrices need special treatment
read_in_k_v(A , A )
# create HuggingFace model and load state dict
if encoder_only:
snake_case__ : str = False
snake_case__ : List[Any] = SegformerForImageClassification(A )
else:
snake_case__ : Dict = SegformerForSemanticSegmentation(A )
model.load_state_dict(A )
model.eval()
# forward pass
snake_case__ : int = model(A )
snake_case__ : Any = outputs.logits
# set expected_slice based on model name
# ADE20k checkpoints
if model_name == "segformer.b0.512x512.ade.160k":
snake_case__ : Dict = torch.tensor(
[
[[-4.6310, -5.5232, -6.2356], [-5.1921, -6.1444, -6.5996], [-5.4424, -6.2790, -6.7574]],
[[-12.1391, -13.3122, -13.9554], [-12.8732, -13.9352, -14.3563], [-12.9438, -13.8226, -14.2513]],
[[-12.5134, -13.4686, -14.4915], [-12.8669, -14.4343, -14.7758], [-13.2523, -14.5819, -15.0694]],
] )
elif model_name == "segformer.b1.512x512.ade.160k":
snake_case__ : Optional[int] = torch.tensor(
[
[[-7.5820, -8.7231, -8.3215], [-8.0600, -10.3529, -10.0304], [-7.5208, -9.4103, -9.6239]],
[[-12.6918, -13.8994, -13.7137], [-13.3196, -15.7523, -15.4789], [-12.9343, -14.8757, -14.9689]],
[[-11.1911, -11.9421, -11.3243], [-11.3342, -13.6839, -13.3581], [-10.3909, -12.1832, -12.4858]],
] )
elif model_name == "segformer.b2.512x512.ade.160k":
snake_case__ : List[Any] = torch.tensor(
[
[[-11.8173, -14.3850, -16.3128], [-14.5648, -16.5804, -18.6568], [-14.7223, -15.7387, -18.4218]],
[[-15.7290, -17.9171, -19.4423], [-18.3105, -19.9448, -21.4661], [-17.9296, -18.6497, -20.7910]],
[[-15.0783, -17.0336, -18.2789], [-16.8771, -18.6870, -20.1612], [-16.2454, -17.1426, -19.5055]],
] )
elif model_name == "segformer.b3.512x512.ade.160k":
snake_case__ : Union[str, Any] = torch.tensor(
[
[[-9.0878, -10.2081, -10.1891], [-9.3144, -10.7941, -10.9843], [-9.2294, -10.3855, -10.5704]],
[[-12.2316, -13.9068, -13.6102], [-12.9161, -14.3702, -14.3235], [-12.5233, -13.7174, -13.7932]],
[[-14.6275, -15.2490, -14.9727], [-14.3400, -15.9687, -16.2827], [-14.1484, -15.4033, -15.8937]],
] )
elif model_name == "segformer.b4.512x512.ade.160k":
snake_case__ : Dict = torch.tensor(
[
[[-12.3144, -13.2447, -14.0802], [-13.3614, -14.5816, -15.6117], [-13.3340, -14.4433, -16.2219]],
[[-19.2781, -20.4128, -20.7506], [-20.6153, -21.6566, -22.0998], [-19.9800, -21.0430, -22.1494]],
[[-18.8739, -19.7804, -21.1834], [-20.1233, -21.6765, -23.2944], [-20.0315, -21.2641, -23.6944]],
] )
elif model_name == "segformer.b5.640x640.ade.160k":
snake_case__ : List[Any] = torch.tensor(
[
[[-9.5524, -12.0835, -11.7348], [-10.5229, -13.6446, -14.5662], [-9.5842, -12.8851, -13.9414]],
[[-15.3432, -17.5323, -17.0818], [-16.3330, -18.9255, -19.2101], [-15.1340, -17.7848, -18.3971]],
[[-12.6072, -14.9486, -14.6631], [-13.7629, -17.0907, -17.7745], [-12.7899, -16.1695, -17.1671]],
] )
# Cityscapes checkpoints
elif model_name == "segformer.b0.1024x1024.city.160k":
snake_case__ : str = torch.tensor(
[
[[-11.9295, -13.4057, -14.8106], [-13.3431, -14.8179, -15.3781], [-14.2836, -15.5942, -16.1588]],
[[-11.4906, -12.8067, -13.6564], [-13.1189, -14.0500, -14.1543], [-13.8748, -14.5136, -14.8789]],
[[0.5374, 0.1067, -0.4742], [0.1141, -0.2255, -0.7099], [-0.3000, -0.5924, -1.3105]],
] )
elif model_name == "segformer.b0.512x1024.city.160k":
snake_case__ : Tuple = torch.tensor(
[
[[-7.8217, -9.8767, -10.1717], [-9.4438, -10.9058, -11.4047], [-9.7939, -12.3495, -12.1079]],
[[-7.1514, -9.5336, -10.0860], [-9.7776, -11.6822, -11.8439], [-10.1411, -12.7655, -12.8972]],
[[0.3021, 0.0805, -0.2310], [-0.0328, -0.1605, -0.2714], [-0.1408, -0.5477, -0.6976]],
] )
elif model_name == "segformer.b0.640x1280.city.160k":
snake_case__ : Any = torch.tensor(
[
[
[-1.1_372e01, -1.2_787e01, -1.3_477e01],
[-1.2_536e01, -1.4_194e01, -1.4_409e01],
[-1.3_217e01, -1.4_888e01, -1.5_327e01],
],
[
[-1.4_791e01, -1.7_122e01, -1.8_277e01],
[-1.7_163e01, -1.9_192e01, -1.9_533e01],
[-1.7_897e01, -1.9_991e01, -2.0_315e01],
],
[
[7.6_723e-01, 4.1_921e-01, -7.7_878e-02],
[4.7_772e-01, 9.5_557e-03, -2.8_082e-01],
[3.6_032e-01, -2.4_826e-01, -5.1_168e-01],
],
] )
elif model_name == "segformer.b0.768x768.city.160k":
snake_case__ : Optional[int] = torch.tensor(
[
[[-9.4959, -11.3087, -11.7479], [-11.0025, -12.6540, -12.3319], [-11.4064, -13.0487, -12.9905]],
[[-9.8905, -11.3084, -12.0854], [-11.1726, -12.7698, -12.9583], [-11.5985, -13.3278, -14.1774]],
[[0.2213, 0.0192, -0.2466], [-0.1731, -0.4213, -0.4874], [-0.3126, -0.6541, -1.1389]],
] )
elif model_name == "segformer.b1.1024x1024.city.160k":
snake_case__ : Union[str, Any] = torch.tensor(
[
[[-13.5748, -13.9111, -12.6500], [-14.3500, -15.3683, -14.2328], [-14.7532, -16.0424, -15.6087]],
[[-17.1651, -15.8725, -12.9653], [-17.2580, -17.3718, -14.8223], [-16.6058, -16.8783, -16.7452]],
[[-3.6456, -3.0209, -1.4203], [-3.0797, -3.1959, -2.0000], [-1.8757, -1.9217, -1.6997]],
] )
elif model_name == "segformer.b2.1024x1024.city.160k":
snake_case__ : List[str] = torch.tensor(
[
[[-16.0976, -16.4856, -17.3962], [-16.6234, -19.0342, -19.7685], [-16.0900, -18.0661, -19.1180]],
[[-18.4750, -18.8488, -19.5074], [-19.4030, -22.1570, -22.5977], [-19.1191, -20.8486, -22.3783]],
[[-4.5178, -5.5037, -6.5109], [-5.0884, -7.2174, -8.0334], [-4.4156, -5.8117, -7.2970]],
] )
elif model_name == "segformer.b3.1024x1024.city.160k":
snake_case__ : List[Any] = torch.tensor(
[
[[-14.2081, -14.4732, -14.1977], [-14.5867, -16.4423, -16.6356], [-13.4441, -14.9685, -16.8696]],
[[-14.4576, -14.7073, -15.0451], [-15.0816, -17.6237, -17.9873], [-14.4213, -16.0199, -18.5992]],
[[-4.7349, -4.9588, -5.0966], [-4.3210, -6.9325, -7.2591], [-3.4312, -4.7484, -7.1917]],
] )
elif model_name == "segformer.b4.1024x1024.city.160k":
snake_case__ : str = torch.tensor(
[
[[-11.7737, -11.9526, -11.3273], [-13.6692, -14.4574, -13.8878], [-13.8937, -14.6924, -15.9345]],
[[-14.6706, -14.5330, -14.1306], [-16.1502, -16.8180, -16.4269], [-16.8338, -17.8939, -20.1746]],
[[1.0491, 0.8289, 1.0310], [1.1044, 0.5219, 0.8055], [1.0899, 0.6926, 0.5590]],
] )
elif model_name == "segformer.b5.1024x1024.city.160k":
snake_case__ : List[str] = torch.tensor(
[
[[-12.5641, -13.4777, -13.0684], [-13.9587, -15.8983, -16.6557], [-13.3109, -15.7350, -16.3141]],
[[-14.7074, -15.4352, -14.5944], [-16.6353, -18.1663, -18.6120], [-15.1702, -18.0329, -18.1547]],
[[-1.7990, -2.0951, -1.7784], [-2.6397, -3.8245, -3.9686], [-1.5264, -2.8126, -2.9316]],
] )
else:
snake_case__ : Tuple = logits.argmax(-1 ).item()
print('Predicted class:' , model.config.idalabel[predicted_class_idx] )
# verify logits
if not encoder_only:
assert logits.shape == expected_shape
assert torch.allclose(logits[0, :3, :3, :3] , A , atol=1e-2 )
# finally, save model and image processor
logger.info(F'''Saving PyTorch model and image processor to {pytorch_dump_folder_path}...''' )
Path(A ).mkdir(exist_ok=A )
model.save_pretrained(A )
image_processor.save_pretrained(A )
if __name__ == "__main__":
a_ :Optional[int] = argparse.ArgumentParser()
parser.add_argument(
"--model_name",
default="segformer.b0.512x512.ade.160k",
type=str,
help="Name of the model you'd like to convert.",
)
parser.add_argument(
"--checkpoint_path", default=None, type=str, help="Path to the original PyTorch checkpoint (.pth file)."
)
parser.add_argument(
"--pytorch_dump_folder_path", default=None, type=str, help="Path to the folder to output PyTorch model."
)
a_ :Union[str, Any] = parser.parse_args()
convert_segformer_checkpoint(args.model_name, args.checkpoint_path, args.pytorch_dump_folder_path)
| 277 | 0 |
_snake_case = {
"a": "AAAAA",
"b": "AAAAB",
"c": "AAABA",
"d": "AAABB",
"e": "AABAA",
"f": "AABAB",
"g": "AABBA",
"h": "AABBB",
"i": "ABAAA",
"j": "BBBAA",
"k": "ABAAB",
"l": "ABABA",
"m": "ABABB",
"n": "ABBAA",
"o": "ABBAB",
"p": "ABBBA",
"q": "ABBBB",
"r": "BAAAA",
"s": "BAAAB",
"t": "BAABA",
"u": "BAABB",
"v": "BBBAB",
"w": "BABAA",
"x": "BABAB",
"y": "BABBA",
"z": "BABBB",
" ": " ",
}
_snake_case = {value: key for key, value in encode_dict.items()}
def lowerCAmelCase_ ( snake_case_ ):
_A : List[Any] = ''
for letter in word.lower():
if letter.isalpha() or letter == " ":
encoded += encode_dict[letter]
else:
raise Exception("""encode() accepts only letters of the alphabet and spaces""" )
return encoded
def lowerCAmelCase_ ( snake_case_ ):
if set(snake_case_ ) - {"A", "B", " "} != set():
raise Exception("""decode() accepts only \'A\', \'B\' and spaces""" )
_A : str = ''
for word in coded.split():
while len(snake_case_ ) != 0:
decoded += decode_dict[word[:5]]
_A : List[str] = word[5:]
decoded += " "
return decoded.strip()
if __name__ == "__main__":
from doctest import testmod
testmod()
| 26 |
import argparse
import json
import os
import fairseq
import torch
from fairseq.data import Dictionary
from transformers import (
WavaVecaConfig,
WavaVecaCTCTokenizer,
WavaVecaFeatureExtractor,
WavaVecaForCTC,
WavaVecaForPreTraining,
WavaVecaProcessor,
logging,
)
from transformers.models.wavaveca.modeling_wavaveca import WavaVecaForSequenceClassification
logging.set_verbosity_info()
a_ :List[Any] = logging.get_logger(__name__)
a_ :List[Any] = {
"post_extract_proj": "feature_projection.projection",
"encoder.pos_conv.0": "encoder.pos_conv_embed.conv",
"self_attn.k_proj": "encoder.layers.*.attention.k_proj",
"self_attn.v_proj": "encoder.layers.*.attention.v_proj",
"self_attn.q_proj": "encoder.layers.*.attention.q_proj",
"self_attn.out_proj": "encoder.layers.*.attention.out_proj",
"self_attn_layer_norm": "encoder.layers.*.layer_norm",
"fc1": "encoder.layers.*.feed_forward.intermediate_dense",
"fc2": "encoder.layers.*.feed_forward.output_dense",
"final_layer_norm": "encoder.layers.*.final_layer_norm",
"encoder.layer_norm": "encoder.layer_norm",
"adapter_layer": "encoder.layers.*.adapter_layer",
"w2v_model.layer_norm": "feature_projection.layer_norm",
"quantizer.weight_proj": "quantizer.weight_proj",
"quantizer.vars": "quantizer.codevectors",
"project_q": "project_q",
"final_proj": "project_hid",
"w2v_encoder.proj": "lm_head",
"mask_emb": "masked_spec_embed",
"pooling_layer.linear": "projector",
"pooling_layer.projection": "classifier",
}
a_ :List[Any] = [
"lm_head",
"quantizer.weight_proj",
"quantizer.codevectors",
"project_q",
"project_hid",
"projector",
"classifier",
]
def lowercase_ (A : Dict ):
snake_case__ : Optional[Any] = {}
with open(A , 'r' ) as file:
for line_number, line in enumerate(A ):
snake_case__ : Dict = line.strip()
if line:
snake_case__ : int = line.split()
snake_case__ : List[str] = line_number
snake_case__ : Dict = words[0]
snake_case__ : Optional[Any] = value
return result
def lowercase_ (A : int , A : int , A : Optional[int] , A : Optional[Any] , A : Tuple ):
for attribute in key.split('.' ):
snake_case__ : Optional[int] = getattr(A , A )
snake_case__ : Union[str, Any] = None
for param_key in PARAM_MAPPING.keys():
if full_name.endswith(A ):
snake_case__ : List[str] = PARAM_MAPPING[full_name.split('.' )[-1]]
snake_case__ : Dict = 'param'
if weight_type is not None and weight_type != "param":
snake_case__ : Union[str, Any] = getattr(A , A ).shape
elif weight_type is not None and weight_type == "param":
snake_case__ : Optional[int] = hf_pointer
for attribute in hf_param_name.split('.' ):
snake_case__ : Optional[Any] = getattr(A , A )
snake_case__ : Dict = shape_pointer.shape
# let's reduce dimension
snake_case__ : List[Any] = value[0]
else:
snake_case__ : Union[str, Any] = hf_pointer.shape
if hf_shape != value.shape:
raise ValueError(
F'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
F''' {value.shape} for {full_name}''' )
if weight_type == "weight":
snake_case__ : Any = value
elif weight_type == "weight_g":
snake_case__ : List[Any] = value
elif weight_type == "weight_v":
snake_case__ : Any = value
elif weight_type == "bias":
snake_case__ : List[Any] = value
elif weight_type == "param":
for attribute in hf_param_name.split('.' ):
snake_case__ : int = getattr(A , A )
snake_case__ : Optional[int] = value
else:
snake_case__ : Optional[Any] = value
logger.info(F'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def lowercase_ (A : Tuple , A : List[Any] , A : int , A : str , A : Tuple ):
snake_case__ : Optional[int] = None
for param_key in PARAM_MAPPING.keys():
if full_name.endswith(A ):
snake_case__ : List[str] = PARAM_MAPPING[full_name.split('.' )[-1]]
snake_case__ : str = 'param'
if weight_type is not None and weight_type != "param":
snake_case__ : int = '.'.join([key, weight_type] )
elif weight_type is not None and weight_type == "param":
snake_case__ : Any = '.'.join([key, hf_param_name] )
else:
snake_case__ : Dict = key
snake_case__ : List[str] = value if 'lm_head' in full_key else value[0]
a_ :List[str] = {
"W_a": "linear_1.weight",
"W_b": "linear_2.weight",
"b_a": "linear_1.bias",
"b_b": "linear_2.bias",
"ln_W": "norm.weight",
"ln_b": "norm.bias",
}
def lowercase_ (A : str , A : Optional[Any] , A : Optional[Any]=None , A : List[str]=None ):
snake_case__ : Optional[int] = False
for key, mapped_key in MAPPING.items():
snake_case__ : Tuple = 'wav2vec2.' + mapped_key if mapped_key not in TOP_LEVEL_KEYS else mapped_key
if key in name or key.split('w2v_model.' )[-1] == name.split('.' )[0]:
snake_case__ : Optional[int] = True
if "*" in mapped_key:
snake_case__ : List[Any] = name.split(A )[0].split('.' )[-2]
snake_case__ : Union[str, Any] = mapped_key.replace('*' , A )
if "weight_g" in name:
snake_case__ : Tuple = 'weight_g'
elif "weight_v" in name:
snake_case__ : List[str] = 'weight_v'
elif "bias" in name:
snake_case__ : Dict = 'bias'
elif "weight" in name:
# TODO: don't match quantizer.weight_proj
snake_case__ : Optional[int] = 'weight'
else:
snake_case__ : str = None
if hf_dict is not None:
rename_dict(A , A , A , A , A )
else:
set_recursively(A , A , A , A , A )
return is_used
return is_used
def lowercase_ (A : Optional[Any] , A : Dict , A : Optional[int] ):
snake_case__ : Dict = []
snake_case__ : Tuple = fairseq_model.state_dict()
snake_case__ : str = hf_model.wavaveca.feature_extractor
for name, value in fairseq_dict.items():
snake_case__ : str = False
if "conv_layers" in name:
load_conv_layer(
A , A , A , A , hf_model.config.feat_extract_norm == 'group' , )
snake_case__ : Any = True
else:
snake_case__ : Dict = load_wavaveca_layer(A , A , A )
if not is_used:
unused_weights.append(A )
logger.warning(F'''Unused weights: {unused_weights}''' )
def lowercase_ (A : Dict , A : Optional[Any] , A : Tuple , A : str , A : List[str] ):
snake_case__ : List[Any] = full_name.split('conv_layers.' )[-1]
snake_case__ : List[str] = name.split('.' )
snake_case__ : List[Any] = int(items[0] )
snake_case__ : str = int(items[1] )
if type_id == 0:
if "bias" in name:
if value.shape != feature_extractor.conv_layers[layer_id].conv.bias.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.''' )
snake_case__ : Any = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
if value.shape != feature_extractor.conv_layers[layer_id].conv.weight.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.''' )
snake_case__ : str = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
if value.shape != feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape} was found.''' )
snake_case__ : str = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
if value.shape != feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape} was found.''' )
snake_case__ : int = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(A )
@torch.no_grad()
def lowercase_ (A : Union[str, Any] , A : str , A : Tuple=None , A : List[str]=None , A : Any=True , A : Optional[int]=False ):
if config_path is not None:
snake_case__ : List[Any] = WavaVecaConfig.from_pretrained(A )
else:
snake_case__ : List[Any] = WavaVecaConfig()
if is_seq_class:
snake_case__ : Dict = read_txt_into_dict(A )
snake_case__ : Any = idalabel
snake_case__ : Union[str, Any] = WavaVecaForSequenceClassification(A )
snake_case__ : Any = WavaVecaFeatureExtractor(
feature_size=1 , sampling_rate=1_6_0_0_0 , padding_value=0 , do_normalize=A , return_attention_mask=A , )
feature_extractor.save_pretrained(A )
elif is_finetuned:
if dict_path:
snake_case__ : str = Dictionary.load(A )
# important change bos & pad token id since CTC symbol is <pad> and
# not <s> as in fairseq
snake_case__ : List[str] = target_dict.pad_index
snake_case__ : Optional[int] = target_dict.bos_index
snake_case__ : Optional[int] = target_dict.eos_index
snake_case__ : List[Any] = len(target_dict.symbols )
snake_case__ : str = os.path.join(A , 'vocab.json' )
if not os.path.isdir(A ):
logger.error('--pytorch_dump_folder_path ({}) should be a directory'.format(A ) )
return
os.makedirs(A , exist_ok=A )
snake_case__ : Optional[Any] = target_dict.indices
# fairseq has the <pad> and <s> switched
snake_case__ : Optional[Any] = 0
snake_case__ : Union[str, Any] = 1
with open(A , 'w' , encoding='utf-8' ) as vocab_handle:
json.dump(A , A )
snake_case__ : List[Any] = WavaVecaCTCTokenizer(
A , unk_token=target_dict.unk_word , pad_token=target_dict.pad_word , bos_token=target_dict.bos_word , eos_token=target_dict.eos_word , word_delimiter_token='|' , do_lower_case=A , )
snake_case__ : str = True if config.feat_extract_norm == 'layer' else False
snake_case__ : Optional[Any] = WavaVecaFeatureExtractor(
feature_size=1 , sampling_rate=1_6_0_0_0 , padding_value=0 , do_normalize=A , return_attention_mask=A , )
snake_case__ : Union[str, Any] = WavaVecaProcessor(feature_extractor=A , tokenizer=A )
processor.save_pretrained(A )
snake_case__ : str = WavaVecaForCTC(A )
else:
snake_case__ : int = WavaVecaForPreTraining(A )
if is_finetuned or is_seq_class:
snake_case__ , snake_case__ , snake_case__ : str = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] , arg_overrides={'data': '/'.join(dict_path.split('/' )[:-1] )} )
else:
snake_case__ : Tuple = argparse.Namespace(task='audio_pretraining' )
snake_case__ : str = fairseq.tasks.setup_task(A )
snake_case__ , snake_case__ , snake_case__ : Any = fairseq.checkpoint_utils.load_model_ensemble_and_task([checkpoint_path] , task=A )
snake_case__ : List[Any] = model[0].eval()
recursively_load_weights(A , A , not is_finetuned )
hf_wavavec.save_pretrained(A )
if __name__ == "__main__":
a_ :List[Any] = argparse.ArgumentParser()
parser.add_argument("--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model.")
parser.add_argument("--checkpoint_path", default=None, type=str, help="Path to fairseq checkpoint")
parser.add_argument("--dict_path", default=None, type=str, help="Path to dict of fine-tuned model")
parser.add_argument("--config_path", default=None, type=str, help="Path to hf config.json of model to convert")
parser.add_argument(
"--not_finetuned", action="store_true", help="Whether the model to convert is a fine-tuned model or not"
)
parser.add_argument(
"--is_seq_class",
action="store_true",
help="Whether the model to convert is a fine-tuned sequence classification model or not",
)
a_ :str = parser.parse_args()
a_ :Tuple = not args.not_finetuned and not args.is_seq_class
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.config_path,
args.dict_path,
is_finetuned,
args.is_seq_class,
)
| 277 | 0 |
import argparse
import json
from collections import OrderedDict
from pathlib import Path
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import (
SegformerConfig,
SegformerForImageClassification,
SegformerForSemanticSegmentation,
SegformerImageProcessor,
)
from transformers.utils import logging
logging.set_verbosity_info()
UpperCAmelCase__ = logging.get_logger(__name__)
def A ( _UpperCAmelCase : Optional[Any] , _UpperCAmelCase : Any=False ) -> str:
'''simple docstring'''
_UpperCAmelCase = OrderedDict()
for key, value in state_dict.items():
if encoder_only and not key.startswith('head' ):
_UpperCAmelCase = 'segformer.encoder.' + key
if key.startswith('backbone' ):
_UpperCAmelCase = key.replace('backbone' , 'segformer.encoder' )
if "patch_embed" in key:
# replace for example patch_embed1 by patch_embeddings.0
_UpperCAmelCase = key[key.find('patch_embed' ) + len('patch_embed' )]
_UpperCAmelCase = key.replace(F"patch_embed{idx}" , F"patch_embeddings.{int(_UpperCAmelCase )-1}" )
if "norm" in key:
_UpperCAmelCase = key.replace('norm' , 'layer_norm' )
if "segformer.encoder.layer_norm" in key:
# replace for example layer_norm1 by layer_norm.0
_UpperCAmelCase = key[key.find('segformer.encoder.layer_norm' ) + len('segformer.encoder.layer_norm' )]
_UpperCAmelCase = key.replace(F"layer_norm{idx}" , F"layer_norm.{int(_UpperCAmelCase )-1}" )
if "layer_norm1" in key:
_UpperCAmelCase = key.replace('layer_norm1' , 'layer_norm_1' )
if "layer_norm2" in key:
_UpperCAmelCase = key.replace('layer_norm2' , 'layer_norm_2' )
if "block" in key:
# replace for example block1 by block.0
_UpperCAmelCase = key[key.find('block' ) + len('block' )]
_UpperCAmelCase = key.replace(F"block{idx}" , F"block.{int(_UpperCAmelCase )-1}" )
if "attn.q" in key:
_UpperCAmelCase = key.replace('attn.q' , 'attention.self.query' )
if "attn.proj" in key:
_UpperCAmelCase = key.replace('attn.proj' , 'attention.output.dense' )
if "attn" in key:
_UpperCAmelCase = key.replace('attn' , 'attention.self' )
if "fc1" in key:
_UpperCAmelCase = key.replace('fc1' , 'dense1' )
if "fc2" in key:
_UpperCAmelCase = key.replace('fc2' , 'dense2' )
if "linear_pred" in key:
_UpperCAmelCase = key.replace('linear_pred' , 'classifier' )
if "linear_fuse" in key:
_UpperCAmelCase = key.replace('linear_fuse.conv' , 'linear_fuse' )
_UpperCAmelCase = key.replace('linear_fuse.bn' , 'batch_norm' )
if "linear_c" in key:
# replace for example linear_c4 by linear_c.3
_UpperCAmelCase = key[key.find('linear_c' ) + len('linear_c' )]
_UpperCAmelCase = key.replace(F"linear_c{idx}" , F"linear_c.{int(_UpperCAmelCase )-1}" )
if key.startswith('head' ):
_UpperCAmelCase = key.replace('head' , 'classifier' )
_UpperCAmelCase = value
return new_state_dict
def A ( _UpperCAmelCase : Tuple , _UpperCAmelCase : Optional[int] ) -> Optional[Any]:
'''simple docstring'''
# for each of the encoder blocks:
for i in range(config.num_encoder_blocks ):
for j in range(config.depths[i] ):
# read in weights + bias of keys and values (which is a single matrix in the original implementation)
_UpperCAmelCase = state_dict.pop(F"segformer.encoder.block.{i}.{j}.attention.self.kv.weight" )
_UpperCAmelCase = state_dict.pop(F"segformer.encoder.block.{i}.{j}.attention.self.kv.bias" )
# next, add keys and values (in that order) to the state dict
_UpperCAmelCase = kv_weight[
: config.hidden_sizes[i], :
]
_UpperCAmelCase = kv_bias[: config.hidden_sizes[i]]
_UpperCAmelCase = kv_weight[
config.hidden_sizes[i] :, :
]
_UpperCAmelCase = kv_bias[
config.hidden_sizes[i] :
]
def A ( ) -> Tuple:
'''simple docstring'''
_UpperCAmelCase = 'http://images.cocodataset.org/val2017/000000039769.jpg'
_UpperCAmelCase = Image.open(requests.get(_UpperCAmelCase , stream=_UpperCAmelCase ).raw )
return image
@torch.no_grad()
def A ( _UpperCAmelCase : Any , _UpperCAmelCase : Union[str, Any] , _UpperCAmelCase : Optional[Any] ) -> int:
'''simple docstring'''
_UpperCAmelCase = SegformerConfig()
_UpperCAmelCase = False
# set attributes based on model_name
_UpperCAmelCase = 'huggingface/label-files'
if "segformer" in model_name:
_UpperCAmelCase = model_name[len('segformer.' ) : len('segformer.' ) + 2]
if "ade" in model_name:
_UpperCAmelCase = 150
_UpperCAmelCase = 'ade20k-id2label.json'
_UpperCAmelCase = (1, 150, 128, 128)
elif "city" in model_name:
_UpperCAmelCase = 19
_UpperCAmelCase = 'cityscapes-id2label.json'
_UpperCAmelCase = (1, 19, 128, 128)
else:
raise ValueError(F"Model {model_name} not supported" )
elif "mit" in model_name:
_UpperCAmelCase = True
_UpperCAmelCase = model_name[4:6]
_UpperCAmelCase = 1_000
_UpperCAmelCase = 'imagenet-1k-id2label.json'
_UpperCAmelCase = (1, 1_000)
else:
raise ValueError(F"Model {model_name} not supported" )
# set config attributes
_UpperCAmelCase = json.load(open(hf_hub_download(_UpperCAmelCase , _UpperCAmelCase , repo_type='dataset' ) , 'r' ) )
_UpperCAmelCase = {int(_UpperCAmelCase ): v for k, v in idalabel.items()}
_UpperCAmelCase = idalabel
_UpperCAmelCase = {v: k for k, v in idalabel.items()}
if size == "b0":
pass
elif size == "b1":
_UpperCAmelCase = [64, 128, 320, 512]
_UpperCAmelCase = 256
elif size == "b2":
_UpperCAmelCase = [64, 128, 320, 512]
_UpperCAmelCase = 768
_UpperCAmelCase = [3, 4, 6, 3]
elif size == "b3":
_UpperCAmelCase = [64, 128, 320, 512]
_UpperCAmelCase = 768
_UpperCAmelCase = [3, 4, 18, 3]
elif size == "b4":
_UpperCAmelCase = [64, 128, 320, 512]
_UpperCAmelCase = 768
_UpperCAmelCase = [3, 8, 27, 3]
elif size == "b5":
_UpperCAmelCase = [64, 128, 320, 512]
_UpperCAmelCase = 768
_UpperCAmelCase = [3, 6, 40, 3]
else:
raise ValueError(F"Size {size} not supported" )
# load image processor (only resize + normalize)
_UpperCAmelCase = SegformerImageProcessor(
image_scale=(512, 512) , keep_ratio=_UpperCAmelCase , align=_UpperCAmelCase , do_random_crop=_UpperCAmelCase )
# prepare image
_UpperCAmelCase = prepare_img()
_UpperCAmelCase = image_processor(images=_UpperCAmelCase , return_tensors='pt' ).pixel_values
logger.info(F"Converting model {model_name}..." )
# load original state dict
if encoder_only:
_UpperCAmelCase = torch.load(_UpperCAmelCase , map_location=torch.device('cpu' ) )
else:
_UpperCAmelCase = torch.load(_UpperCAmelCase , map_location=torch.device('cpu' ) )['state_dict']
# rename keys
_UpperCAmelCase = rename_keys(_UpperCAmelCase , encoder_only=_UpperCAmelCase )
if not encoder_only:
del state_dict["decode_head.conv_seg.weight"]
del state_dict["decode_head.conv_seg.bias"]
# key and value matrices need special treatment
read_in_k_v(_UpperCAmelCase , _UpperCAmelCase )
# create HuggingFace model and load state dict
if encoder_only:
_UpperCAmelCase = False
_UpperCAmelCase = SegformerForImageClassification(_UpperCAmelCase )
else:
_UpperCAmelCase = SegformerForSemanticSegmentation(_UpperCAmelCase )
model.load_state_dict(_UpperCAmelCase )
model.eval()
# forward pass
_UpperCAmelCase = model(_UpperCAmelCase )
_UpperCAmelCase = outputs.logits
# set expected_slice based on model name
# ADE20k checkpoints
if model_name == "segformer.b0.512x512.ade.160k":
_UpperCAmelCase = torch.tensor(
[
[[-4.6310, -5.5232, -6.2356], [-5.1921, -6.1444, -6.5996], [-5.4424, -6.2790, -6.7574]],
[[-12.1391, -13.3122, -13.9554], [-12.8732, -13.9352, -14.3563], [-12.9438, -13.8226, -14.2513]],
[[-12.5134, -13.4686, -14.4915], [-12.8669, -14.4343, -14.7758], [-13.2523, -14.5819, -15.0694]],
] )
elif model_name == "segformer.b1.512x512.ade.160k":
_UpperCAmelCase = torch.tensor(
[
[[-7.5820, -8.7231, -8.3215], [-8.0600, -10.3529, -10.0304], [-7.5208, -9.4103, -9.6239]],
[[-12.6918, -13.8994, -13.7137], [-13.3196, -15.7523, -15.4789], [-12.9343, -14.8757, -14.9689]],
[[-11.1911, -11.9421, -11.3243], [-11.3342, -13.6839, -13.3581], [-10.3909, -12.1832, -12.4858]],
] )
elif model_name == "segformer.b2.512x512.ade.160k":
_UpperCAmelCase = torch.tensor(
[
[[-11.8173, -14.3850, -16.3128], [-14.5648, -16.5804, -18.6568], [-14.7223, -15.7387, -18.4218]],
[[-15.7290, -17.9171, -19.4423], [-18.3105, -19.9448, -21.4661], [-17.9296, -18.6497, -20.7910]],
[[-15.0783, -17.0336, -18.2789], [-16.8771, -18.6870, -20.1612], [-16.2454, -17.1426, -19.5055]],
] )
elif model_name == "segformer.b3.512x512.ade.160k":
_UpperCAmelCase = torch.tensor(
[
[[-9.0878, -10.2081, -10.1891], [-9.3144, -10.7941, -10.9843], [-9.2294, -10.3855, -10.5704]],
[[-12.2316, -13.9068, -13.6102], [-12.9161, -14.3702, -14.3235], [-12.5233, -13.7174, -13.7932]],
[[-14.6275, -15.2490, -14.9727], [-14.3400, -15.9687, -16.2827], [-14.1484, -15.4033, -15.8937]],
] )
elif model_name == "segformer.b4.512x512.ade.160k":
_UpperCAmelCase = torch.tensor(
[
[[-12.3144, -13.2447, -14.0802], [-13.3614, -14.5816, -15.6117], [-13.3340, -14.4433, -16.2219]],
[[-19.2781, -20.4128, -20.7506], [-20.6153, -21.6566, -22.0998], [-19.9800, -21.0430, -22.1494]],
[[-18.8739, -19.7804, -21.1834], [-20.1233, -21.6765, -23.2944], [-20.0315, -21.2641, -23.6944]],
] )
elif model_name == "segformer.b5.640x640.ade.160k":
_UpperCAmelCase = torch.tensor(
[
[[-9.5524, -12.0835, -11.7348], [-10.5229, -13.6446, -14.5662], [-9.5842, -12.8851, -13.9414]],
[[-15.3432, -17.5323, -17.0818], [-16.3330, -18.9255, -19.2101], [-15.1340, -17.7848, -18.3971]],
[[-12.6072, -14.9486, -14.6631], [-13.7629, -17.0907, -17.7745], [-12.7899, -16.1695, -17.1671]],
] )
# Cityscapes checkpoints
elif model_name == "segformer.b0.1024x1024.city.160k":
_UpperCAmelCase = torch.tensor(
[
[[-11.9295, -13.4057, -14.8106], [-13.3431, -14.8179, -15.3781], [-14.2836, -15.5942, -16.1588]],
[[-11.4906, -12.8067, -13.6564], [-13.1189, -14.0500, -14.1543], [-13.8748, -14.5136, -14.8789]],
[[0.5374, 0.1067, -0.4742], [0.1141, -0.2255, -0.7099], [-0.3000, -0.5924, -1.3105]],
] )
elif model_name == "segformer.b0.512x1024.city.160k":
_UpperCAmelCase = torch.tensor(
[
[[-7.8217, -9.8767, -10.1717], [-9.4438, -10.9058, -11.4047], [-9.7939, -12.3495, -12.1079]],
[[-7.1514, -9.5336, -10.0860], [-9.7776, -11.6822, -11.8439], [-10.1411, -12.7655, -12.8972]],
[[0.3021, 0.0805, -0.2310], [-0.0328, -0.1605, -0.2714], [-0.1408, -0.5477, -0.6976]],
] )
elif model_name == "segformer.b0.640x1280.city.160k":
_UpperCAmelCase = torch.tensor(
[
[
[-1.13_72E01, -1.27_87E01, -1.34_77E01],
[-1.25_36E01, -1.41_94E01, -1.44_09E01],
[-1.32_17E01, -1.48_88E01, -1.53_27E01],
],
[
[-1.47_91E01, -1.71_22E01, -1.82_77E01],
[-1.71_63E01, -1.91_92E01, -1.95_33E01],
[-1.78_97E01, -1.99_91E01, -2.03_15E01],
],
[
[7.67_23E-01, 4.19_21E-01, -7.78_78E-02],
[4.77_72E-01, 9.55_57E-03, -2.80_82E-01],
[3.60_32E-01, -2.48_26E-01, -5.11_68E-01],
],
] )
elif model_name == "segformer.b0.768x768.city.160k":
_UpperCAmelCase = torch.tensor(
[
[[-9.4959, -11.3087, -11.7479], [-11.0025, -12.6540, -12.3319], [-11.4064, -13.0487, -12.9905]],
[[-9.8905, -11.3084, -12.0854], [-11.1726, -12.7698, -12.9583], [-11.5985, -13.3278, -14.1774]],
[[0.2213, 0.0192, -0.2466], [-0.1731, -0.4213, -0.4874], [-0.3126, -0.6541, -1.1389]],
] )
elif model_name == "segformer.b1.1024x1024.city.160k":
_UpperCAmelCase = torch.tensor(
[
[[-13.5748, -13.9111, -12.6500], [-14.3500, -15.3683, -14.2328], [-14.7532, -16.0424, -15.6087]],
[[-17.1651, -15.8725, -12.9653], [-17.2580, -17.3718, -14.8223], [-16.6058, -16.8783, -16.7452]],
[[-3.6456, -3.0209, -1.4203], [-3.0797, -3.1959, -2.0000], [-1.8757, -1.9217, -1.6997]],
] )
elif model_name == "segformer.b2.1024x1024.city.160k":
_UpperCAmelCase = torch.tensor(
[
[[-16.0976, -16.4856, -17.3962], [-16.6234, -19.0342, -19.7685], [-16.0900, -18.0661, -19.1180]],
[[-18.4750, -18.8488, -19.5074], [-19.4030, -22.1570, -22.5977], [-19.1191, -20.8486, -22.3783]],
[[-4.5178, -5.5037, -6.5109], [-5.0884, -7.2174, -8.0334], [-4.4156, -5.8117, -7.2970]],
] )
elif model_name == "segformer.b3.1024x1024.city.160k":
_UpperCAmelCase = torch.tensor(
[
[[-14.2081, -14.4732, -14.1977], [-14.5867, -16.4423, -16.6356], [-13.4441, -14.9685, -16.8696]],
[[-14.4576, -14.7073, -15.0451], [-15.0816, -17.6237, -17.9873], [-14.4213, -16.0199, -18.5992]],
[[-4.7349, -4.9588, -5.0966], [-4.3210, -6.9325, -7.2591], [-3.4312, -4.7484, -7.1917]],
] )
elif model_name == "segformer.b4.1024x1024.city.160k":
_UpperCAmelCase = torch.tensor(
[
[[-11.7737, -11.9526, -11.3273], [-13.6692, -14.4574, -13.8878], [-13.8937, -14.6924, -15.9345]],
[[-14.6706, -14.5330, -14.1306], [-16.1502, -16.8180, -16.4269], [-16.8338, -17.8939, -20.1746]],
[[1.0491, 0.8289, 1.0310], [1.1044, 0.5219, 0.8055], [1.0899, 0.6926, 0.5590]],
] )
elif model_name == "segformer.b5.1024x1024.city.160k":
_UpperCAmelCase = torch.tensor(
[
[[-12.5641, -13.4777, -13.0684], [-13.9587, -15.8983, -16.6557], [-13.3109, -15.7350, -16.3141]],
[[-14.7074, -15.4352, -14.5944], [-16.6353, -18.1663, -18.6120], [-15.1702, -18.0329, -18.1547]],
[[-1.7990, -2.0951, -1.7784], [-2.6397, -3.8245, -3.9686], [-1.5264, -2.8126, -2.9316]],
] )
else:
_UpperCAmelCase = logits.argmax(-1 ).item()
print('Predicted class:' , model.config.idalabel[predicted_class_idx] )
# verify logits
if not encoder_only:
assert logits.shape == expected_shape
assert torch.allclose(logits[0, :3, :3, :3] , _UpperCAmelCase , atol=1E-2 )
# finally, save model and image processor
logger.info(F"Saving PyTorch model and image processor to {pytorch_dump_folder_path}..." )
Path(_UpperCAmelCase ).mkdir(exist_ok=_UpperCAmelCase )
model.save_pretrained(_UpperCAmelCase )
image_processor.save_pretrained(_UpperCAmelCase )
if __name__ == "__main__":
UpperCAmelCase__ = argparse.ArgumentParser()
parser.add_argument(
"--model_name",
default="segformer.b0.512x512.ade.160k",
type=str,
help="Name of the model you'd like to convert.",
)
parser.add_argument(
"--checkpoint_path", default=None, type=str, help="Path to the original PyTorch checkpoint (.pth file)."
)
parser.add_argument(
"--pytorch_dump_folder_path", default=None, type=str, help="Path to the folder to output PyTorch model."
)
UpperCAmelCase__ = parser.parse_args()
convert_segformer_checkpoint(args.model_name, args.checkpoint_path, args.pytorch_dump_folder_path)
| 339 |
from typing import Dict, List
from nltk.translate import gleu_score
import datasets
from datasets import MetricInfo
a_ :Any = "\\n@misc{wu2016googles,\n title={Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation},\n author={Yonghui Wu and Mike Schuster and Zhifeng Chen and Quoc V. Le and Mohammad Norouzi and Wolfgang Macherey\n and Maxim Krikun and Yuan Cao and Qin Gao and Klaus Macherey and Jeff Klingner and Apurva Shah and Melvin\n Johnson and Xiaobing Liu and Łukasz Kaiser and Stephan Gouws and Yoshikiyo Kato and Taku Kudo and Hideto\n Kazawa and Keith Stevens and George Kurian and Nishant Patil and Wei Wang and Cliff Young and\n Jason Smith and Jason Riesa and Alex Rudnick and Oriol Vinyals and Greg Corrado and Macduff Hughes\n and Jeffrey Dean},\n year={2016},\n eprint={1609.08144},\n archivePrefix={arXiv},\n primaryClass={cs.CL}\n}\n"
a_ :List[str] = "\\nThe BLEU score has some undesirable properties when used for single\nsentences, as it was designed to be a corpus measure. We therefore\nuse a slightly different score for our RL experiments which we call\nthe 'GLEU score'. For the GLEU score, we record all sub-sequences of\n1, 2, 3 or 4 tokens in output and target sequence (n-grams). We then\ncompute a recall, which is the ratio of the number of matching n-grams\nto the number of total n-grams in the target (ground truth) sequence,\nand a precision, which is the ratio of the number of matching n-grams\nto the number of total n-grams in the generated output sequence. Then\nGLEU score is simply the minimum of recall and precision. This GLEU\nscore's range is always between 0 (no matches) and 1 (all match) and\nit is symmetrical when switching output and target. According to\nour experiments, GLEU score correlates quite well with the BLEU\nmetric on a corpus level but does not have its drawbacks for our per\nsentence reward objective.\n"
a_ :List[str] = "\\nComputes corpus-level Google BLEU (GLEU) score of translated segments against one or more references.\nInstead of averaging the sentence level GLEU scores (i.e. macro-average precision), Wu et al. (2016) sum up the matching\ntokens and the max of hypothesis and reference tokens for each sentence, then compute using the aggregate values.\n\nArgs:\n predictions (list of str): list of translations to score.\n Each translation should be tokenized into a list of tokens.\n references (list of list of str): list of lists of references for each translation.\n Each reference should be tokenized into a list of tokens.\n min_len (int): The minimum order of n-gram this function should extract. Defaults to 1.\n max_len (int): The maximum order of n-gram this function should extract. Defaults to 4.\n\nReturns:\n 'google_bleu': google_bleu score\n\nExamples:\n Example 1:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses, references=list_of_references)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.44\n\n Example 2:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n >>> ref1b = ['It', 'is', 'a', 'guide', 'to', 'action', 'that',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'will', 'never',\n ... 'heed', 'the', 'cat', 'commands']\n >>> ref1c = ['It', 'is', 'the', 'practical', 'guide', 'for', 'the',\n ... 'rubber', 'duck', 'army', 'never', 'to', 'heed', 'the', 'directions',\n ... 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a, ref1b, ref1c], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses, references=list_of_references)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.61\n\n Example 3:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n >>> ref1b = ['It', 'is', 'a', 'guide', 'to', 'action', 'that',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'will', 'never',\n ... 'heed', 'the', 'cat', 'commands']\n >>> ref1c = ['It', 'is', 'the', 'practical', 'guide', 'for', 'the',\n ... 'rubber', 'duck', 'army', 'never', 'to', 'heed', 'the', 'directions',\n ... 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a, ref1b, ref1c], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses, references=list_of_references, min_len=2)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.53\n\n Example 4:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n >>> ref1b = ['It', 'is', 'a', 'guide', 'to', 'action', 'that',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'will', 'never',\n ... 'heed', 'the', 'cat', 'commands']\n >>> ref1c = ['It', 'is', 'the', 'practical', 'guide', 'for', 'the',\n ... 'rubber', 'duck', 'army', 'never', 'to', 'heed', 'the', 'directions',\n ... 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a, ref1b, ref1c], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses,references=list_of_references, min_len=2, max_len=6)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.4\n"
@datasets.utils.file_utils.add_start_docstrings(_DESCRIPTION , _KWARGS_DESCRIPTION )
class snake_case__ ( datasets.Metric ):
"""simple docstring"""
def lowercase_ ( self : str ) ->MetricInfo:
return datasets.MetricInfo(
description=_DESCRIPTION, citation=_CITATION, inputs_description=_KWARGS_DESCRIPTION, features=datasets.Features(
{
'predictions': datasets.Sequence(datasets.Value('string', id='token' ), id='sequence' ),
'references': datasets.Sequence(
datasets.Sequence(datasets.Value('string', id='token' ), id='sequence' ), id='references' ),
} ), )
def lowercase_ ( self : str, _snake_case : List[List[List[str]]], _snake_case : List[List[str]], _snake_case : int = 1, _snake_case : int = 4, ) ->Dict[str, float]:
return {
"google_bleu": gleu_score.corpus_gleu(
list_of_references=_snake_case, hypotheses=_snake_case, min_len=_snake_case, max_len=_snake_case )
}
| 277 | 0 |
import unittest
import numpy as np
from transformers.testing_utils import require_torch, require_vision
from transformers.utils import is_torch_available, is_vision_available
from ...test_image_processing_common import ImageProcessingSavingTestMixin, prepare_image_inputs
if is_torch_available():
import torch
if is_vision_available():
from PIL import Image
from transformers import LevitImageProcessor
class UpperCamelCase_ ( unittest.TestCase ):
'''simple docstring'''
def __init__( self , a , a=7 , a=3 , a=18 , a=30 , a=4_00 , a=True , a=None , a=True , a=None , a=True , a=[0.5, 0.5, 0.5] , a=[0.5, 0.5, 0.5] , ) -> str:
snake_case_ = size if size is not None else {'shortest_edge': 18}
snake_case_ = crop_size if crop_size is not None else {'height': 18, 'width': 18}
snake_case_ = parent
snake_case_ = batch_size
snake_case_ = num_channels
snake_case_ = image_size
snake_case_ = min_resolution
snake_case_ = max_resolution
snake_case_ = do_resize
snake_case_ = size
snake_case_ = do_center_crop
snake_case_ = crop_size
snake_case_ = do_normalize
snake_case_ = image_mean
snake_case_ = image_std
def _UpperCamelCase ( self ) -> List[str]:
return {
"image_mean": self.image_mean,
"image_std": self.image_std,
"do_normalize": self.do_normalize,
"do_resize": self.do_resize,
"do_center_crop": self.do_center_crop,
"size": self.size,
"crop_size": self.crop_size,
}
@require_torch
@require_vision
class UpperCamelCase_ ( lowerCAmelCase_ , unittest.TestCase ):
'''simple docstring'''
lowerCAmelCase = LevitImageProcessor if is_vision_available() else None
def _UpperCamelCase ( self ) -> Any:
snake_case_ = LevitImageProcessingTester(self )
@property
def _UpperCamelCase ( self ) -> Dict:
return self.image_processor_tester.prepare_image_processor_dict()
def _UpperCamelCase ( self ) -> List[str]:
snake_case_ = self.image_processing_class(**self.image_processor_dict )
self.assertTrue(hasattr(_snake_case , 'image_mean' ) )
self.assertTrue(hasattr(_snake_case , 'image_std' ) )
self.assertTrue(hasattr(_snake_case , 'do_normalize' ) )
self.assertTrue(hasattr(_snake_case , 'do_resize' ) )
self.assertTrue(hasattr(_snake_case , 'do_center_crop' ) )
self.assertTrue(hasattr(_snake_case , 'size' ) )
def _UpperCamelCase ( self ) -> str:
snake_case_ = self.image_processing_class.from_dict(self.image_processor_dict )
self.assertEqual(image_processor.size , {'shortest_edge': 18} )
self.assertEqual(image_processor.crop_size , {'height': 18, 'width': 18} )
snake_case_ = self.image_processing_class.from_dict(self.image_processor_dict , size=42 , crop_size=84 )
self.assertEqual(image_processor.size , {'shortest_edge': 42} )
self.assertEqual(image_processor.crop_size , {'height': 84, 'width': 84} )
def _UpperCamelCase ( self ) -> Optional[int]:
pass
def _UpperCamelCase ( self ) -> Optional[Any]:
# Initialize image_processing
snake_case_ = self.image_processing_class(**self.image_processor_dict )
# create random PIL images
snake_case_ = prepare_image_inputs(self.image_processor_tester , equal_resolution=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case , Image.Image )
# Test not batched input
snake_case_ = image_processing(image_inputs[0] , return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape , (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
) , )
# Test batched
snake_case_ = image_processing(_snake_case , return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape , (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
) , )
def _UpperCamelCase ( self ) -> Tuple:
# Initialize image_processing
snake_case_ = self.image_processing_class(**self.image_processor_dict )
# create random numpy tensors
snake_case_ = prepare_image_inputs(self.image_processor_tester , equal_resolution=_snake_case , numpify=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case , np.ndarray )
# Test not batched input
snake_case_ = image_processing(image_inputs[0] , return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape , (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
) , )
# Test batched
snake_case_ = image_processing(_snake_case , return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape , (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
) , )
def _UpperCamelCase ( self ) -> List[Any]:
# Initialize image_processing
snake_case_ = self.image_processing_class(**self.image_processor_dict )
# create random PyTorch tensors
snake_case_ = prepare_image_inputs(self.image_processor_tester , equal_resolution=_snake_case , torchify=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case , torch.Tensor )
# Test not batched input
snake_case_ = image_processing(image_inputs[0] , return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape , (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
) , )
# Test batched
snake_case_ = image_processing(_snake_case , return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape , (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
) , )
| 178 |
from math import factorial
def lowercase_ (A : int , A : int , A : float ):
if successes > trials:
raise ValueError('successes must be lower or equal to trials' )
if trials < 0 or successes < 0:
raise ValueError('the function is defined for non-negative integers' )
if not isinstance(A , A ) or not isinstance(A , A ):
raise ValueError('the function is defined for non-negative integers' )
if not 0 < prob < 1:
raise ValueError('prob has to be in range of 1 - 0' )
snake_case__ : List[Any] = (prob**successes) * ((1 - prob) ** (trials - successes))
# Calculate the binomial coefficient: n! / k!(n-k)!
snake_case__ : List[str] = float(factorial(A ) )
coefficient /= factorial(A ) * factorial(trials - successes )
return probability * coefficient
if __name__ == "__main__":
from doctest import testmod
testmod()
print("Probability of 2 successes out of 4 trails")
print("with probability of 0.75 is:", end=" ")
print(binomial_distribution(2, 4, 0.75))
| 277 | 0 |
"""simple docstring"""
import json
import os
import tempfile
from unittest.mock import patch
import torch
from torch.utils.data import DataLoader, TensorDataset
from accelerate import DistributedType, infer_auto_device_map, init_empty_weights
from accelerate.accelerator import Accelerator
from accelerate.state import GradientState, PartialState
from accelerate.test_utils import require_bnb, require_multi_gpu, slow
from accelerate.test_utils.testing import AccelerateTestCase, require_cuda
from accelerate.utils import patch_environment
def lowercase () -> Optional[int]:
SCREAMING_SNAKE_CASE = torch.nn.Linear(2 , 4 )
SCREAMING_SNAKE_CASE = torch.optim.AdamW(model.parameters() , lr=1.0 )
SCREAMING_SNAKE_CASE = torch.optim.lr_scheduler.OneCycleLR(SCREAMING_SNAKE_CASE_ , max_lr=0.01 , steps_per_epoch=2 , epochs=1 )
SCREAMING_SNAKE_CASE = DataLoader(TensorDataset(torch.tensor([1, 2, 3] ) ) )
SCREAMING_SNAKE_CASE = DataLoader(TensorDataset(torch.tensor([4, 5, 6] ) ) )
return model, optimizer, scheduler, train_dl, valid_dl
def lowercase (SCREAMING_SNAKE_CASE_ : Optional[Any] ) -> int:
return (model.weight.abs().sum() + model.bias.abs().sum()).item()
def lowercase (SCREAMING_SNAKE_CASE_ : str ) -> List[str]:
SCREAMING_SNAKE_CASE = torch.nn.Linear(*tuple(model.weight.T.shape ) ).state_dict()
model.load_state_dict(SCREAMING_SNAKE_CASE_ )
class lowerCAmelCase ( lowerCAmelCase_ ):
'''simple docstring'''
@require_cuda
def __A ( self ) -> Optional[int]:
SCREAMING_SNAKE_CASE = Accelerator()
assert PartialState._shared_state["_cpu"] is False
assert PartialState._shared_state["device"].type == "cuda"
with self.assertRaises(_snake_case ):
SCREAMING_SNAKE_CASE = Accelerator(cpu=_snake_case )
def __A ( self ) -> Dict:
SCREAMING_SNAKE_CASE = Accelerator()
SCREAMING_SNAKE_CASE = GradientState()
assert state.num_steps == 1
SCREAMING_SNAKE_CASE = 4
assert state.num_steps == 4
assert state.sync_gradients is True
SCREAMING_SNAKE_CASE = False
assert state.sync_gradients is False
GradientState._reset_state()
def __A ( self ) -> Optional[Any]:
SCREAMING_SNAKE_CASE = Accelerator()
SCREAMING_SNAKE_CASE = create_components()
(
SCREAMING_SNAKE_CASE
) = accelerator.prepare(_snake_case , _snake_case , _snake_case , _snake_case , _snake_case )
self.assertTrue(prepared_model in accelerator._models )
self.assertTrue(prepared_optimizer in accelerator._optimizers )
self.assertTrue(prepared_scheduler in accelerator._schedulers )
self.assertTrue(prepared_train_dl in accelerator._dataloaders )
self.assertTrue(prepared_valid_dl in accelerator._dataloaders )
def __A ( self ) -> int:
SCREAMING_SNAKE_CASE = Accelerator()
SCREAMING_SNAKE_CASE = create_components()
accelerator.prepare(_snake_case , _snake_case , _snake_case , _snake_case , _snake_case )
accelerator.free_memory()
self.assertTrue(len(accelerator._models ) == 0 )
self.assertTrue(len(accelerator._optimizers ) == 0 )
self.assertTrue(len(accelerator._schedulers ) == 0 )
self.assertTrue(len(accelerator._dataloaders ) == 0 )
def __A ( self ) -> List[Any]:
PartialState._reset_state()
# Mock torch.cuda.set_device to avoid an exception as the device doesn't exist
def noop(*lowerCAmelCase__ , **lowerCAmelCase__ ):
pass
with patch('torch.cuda.set_device' , _snake_case ), patch_environment(ACCELERATE_TORCH_DEVICE='cuda:64' ):
SCREAMING_SNAKE_CASE = Accelerator()
self.assertEqual(str(accelerator.state.device ) , 'cuda:64' )
def __A ( self ) -> List[Any]:
SCREAMING_SNAKE_CASE = Accelerator()
SCREAMING_SNAKE_CASE = create_components()
accelerator.prepare(_snake_case , _snake_case , _snake_case , _snake_case , _snake_case )
SCREAMING_SNAKE_CASE = get_signature(_snake_case )
with tempfile.TemporaryDirectory() as tmpdirname:
accelerator.save_state(_snake_case )
# make sure random weights don't match
load_random_weights(_snake_case )
self.assertTrue(abs(model_signature - get_signature(_snake_case ) ) > 1e-3 )
# make sure loaded weights match
accelerator.load_state(_snake_case )
self.assertTrue(abs(model_signature - get_signature(_snake_case ) ) < 1e-3 )
def __A ( self ) -> int:
SCREAMING_SNAKE_CASE = Accelerator()
SCREAMING_SNAKE_CASE = create_components()
accelerator.prepare(_snake_case , _snake_case , _snake_case , _snake_case , _snake_case )
SCREAMING_SNAKE_CASE = get_signature(_snake_case )
# saving hook
def save_config(lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ ):
SCREAMING_SNAKE_CASE = {'class_name': models[0].__class__.__name__}
with open(os.path.join(_snake_case , 'data.json' ) , 'w' ) as f:
json.dump(_snake_case , _snake_case )
# loading hook
def load_config(lowerCAmelCase__ , lowerCAmelCase__ ):
with open(os.path.join(_snake_case , 'data.json' ) , 'r' ) as f:
SCREAMING_SNAKE_CASE = json.load(_snake_case )
SCREAMING_SNAKE_CASE = config['class_name']
SCREAMING_SNAKE_CASE = accelerator.register_save_state_pre_hook(_snake_case )
SCREAMING_SNAKE_CASE = accelerator.register_load_state_pre_hook(_snake_case )
with tempfile.TemporaryDirectory() as tmpdirname:
accelerator.save_state(_snake_case )
# make sure random weights don't match with hooks
load_random_weights(_snake_case )
self.assertTrue(abs(model_signature - get_signature(_snake_case ) ) > 1e-3 )
# random class name to verify correct one is loaded
SCREAMING_SNAKE_CASE = 'random'
# make sure loaded weights match with hooks
accelerator.load_state(_snake_case )
self.assertTrue(abs(model_signature - get_signature(_snake_case ) ) < 1e-3 )
# mode.class_name is loaded from config
self.assertTrue(model.class_name == model.__class__.__name__ )
# remove hooks
save_hook.remove()
load_hook.remove()
with tempfile.TemporaryDirectory() as tmpdirname:
accelerator.save_state(_snake_case )
# make sure random weights don't match with hooks removed
load_random_weights(_snake_case )
self.assertTrue(abs(model_signature - get_signature(_snake_case ) ) > 1e-3 )
# random class name to verify correct one is loaded
SCREAMING_SNAKE_CASE = 'random'
# make sure loaded weights match with hooks removed
accelerator.load_state(_snake_case )
self.assertTrue(abs(model_signature - get_signature(_snake_case ) ) < 1e-3 )
# mode.class_name is NOT loaded from config
self.assertTrue(model.class_name != model.__class__.__name__ )
def __A ( self ) -> Any:
SCREAMING_SNAKE_CASE = Accelerator()
SCREAMING_SNAKE_CASE = create_components()
SCREAMING_SNAKE_CASE = None
# This should work
SCREAMING_SNAKE_CASE = accelerator.prepare(
_snake_case , _snake_case , _snake_case , _snake_case , _snake_case , _snake_case )
self.assertTrue(dummy_obj is None )
def __A ( self ) -> Optional[int]:
SCREAMING_SNAKE_CASE = Accelerator()
SCREAMING_SNAKE_CASE = create_components()
SCREAMING_SNAKE_CASE = [1, 2, 3]
# This should work
SCREAMING_SNAKE_CASE = accelerator.prepare(
_snake_case , _snake_case , _snake_case , _snake_case , _snake_case , _snake_case )
self.assertEqual(
getattr(_snake_case , '_is_accelerate_prepared' , _snake_case ) , _snake_case , 'Dummy object should have `_is_accelerate_prepared` set to `True`' , )
self.assertEqual(
getattr(_snake_case , '_is_accelerate_prepared' , _snake_case ) , _snake_case , 'Model is missing `_is_accelerator_prepared` or is set to `False`' , )
self.assertEqual(
getattr(_snake_case , '_is_accelerate_prepared' , _snake_case ) , _snake_case , 'Optimizer is missing `_is_accelerator_prepared` or is set to `False`' , )
self.assertEqual(
getattr(_snake_case , '_is_accelerate_prepared' , _snake_case ) , _snake_case , 'Scheduler is missing `_is_accelerator_prepared` or is set to `False`' , )
self.assertEqual(
getattr(_snake_case , '_is_accelerate_prepared' , _snake_case ) , _snake_case , 'Train Dataloader is missing `_is_accelerator_prepared` or is set to `False`' , )
self.assertEqual(
getattr(_snake_case , '_is_accelerate_prepared' , _snake_case ) , _snake_case , 'Valid Dataloader is missing `_is_accelerator_prepared` or is set to `False`' , )
@slow
@require_bnb
def __A ( self ) -> Any:
from transformers import AutoModelForCausalLM
SCREAMING_SNAKE_CASE = AutoModelForCausalLM.from_pretrained(
'EleutherAI/gpt-neo-125m' , load_in_abit=_snake_case , device_map={'': 0} , )
SCREAMING_SNAKE_CASE = Accelerator()
# This should work
SCREAMING_SNAKE_CASE = accelerator.prepare(_snake_case )
@slow
@require_bnb
def __A ( self ) -> str:
from transformers import AutoModelForCausalLM
SCREAMING_SNAKE_CASE = Accelerator()
with init_empty_weights():
SCREAMING_SNAKE_CASE = AutoModelForCausalLM.from_pretrained(
'EleutherAI/gpt-neo-125m' , )
model.tie_weights()
SCREAMING_SNAKE_CASE = infer_auto_device_map(_snake_case )
SCREAMING_SNAKE_CASE = 'cpu'
SCREAMING_SNAKE_CASE = AutoModelForCausalLM.from_pretrained(
'EleutherAI/gpt-neo-125m' , device_map=_snake_case , load_in_abit=_snake_case , llm_inta_enable_fpaa_cpu_offload=_snake_case )
# This should not work and get value error
with self.assertRaises(_snake_case ):
SCREAMING_SNAKE_CASE = accelerator.prepare(_snake_case )
@slow
@require_bnb
@require_multi_gpu
def __A ( self ) -> Optional[Any]:
from transformers import AutoModelForCausalLM
SCREAMING_SNAKE_CASE = {'distributed_type': DistributedType.MULTI_GPU}
with init_empty_weights():
SCREAMING_SNAKE_CASE = AutoModelForCausalLM.from_pretrained(
'EleutherAI/gpt-neo-125m' , )
model.tie_weights()
SCREAMING_SNAKE_CASE = infer_auto_device_map(_snake_case )
SCREAMING_SNAKE_CASE = 1
SCREAMING_SNAKE_CASE = AutoModelForCausalLM.from_pretrained(
'EleutherAI/gpt-neo-125m' , load_in_abit=_snake_case , device_map=_snake_case , )
SCREAMING_SNAKE_CASE = Accelerator()
# This should not work and get value error
with self.assertRaises(_snake_case ):
SCREAMING_SNAKE_CASE = accelerator.prepare(_snake_case )
PartialState._reset_state()
@slow
@require_bnb
@require_multi_gpu
def __A ( self ) -> Tuple:
from transformers import AutoModelForCausalLM
with init_empty_weights():
SCREAMING_SNAKE_CASE = AutoModelForCausalLM.from_pretrained(
'EleutherAI/gpt-neo-125m' , )
SCREAMING_SNAKE_CASE = infer_auto_device_map(_snake_case )
SCREAMING_SNAKE_CASE = 1
SCREAMING_SNAKE_CASE = AutoModelForCausalLM.from_pretrained(
'EleutherAI/gpt-neo-125m' , load_in_abit=_snake_case , device_map=_snake_case , )
SCREAMING_SNAKE_CASE = Accelerator()
# This should work
SCREAMING_SNAKE_CASE = accelerator.prepare(_snake_case )
@require_cuda
def __A ( self ) -> List[Any]:
SCREAMING_SNAKE_CASE = torch.nn.Linear(10 , 10 )
SCREAMING_SNAKE_CASE = torch.optim.SGD(model.parameters() , lr=0.01 )
SCREAMING_SNAKE_CASE = Accelerator(cpu=_snake_case )
SCREAMING_SNAKE_CASE = accelerator.prepare(_snake_case )
| 113 |
from collections import UserDict
from typing import Union
import numpy as np
import requests
from ..utils import (
add_end_docstrings,
logging,
)
from .audio_classification import ffmpeg_read
from .base import PIPELINE_INIT_ARGS, Pipeline
a_ :List[Any] = logging.get_logger(__name__)
@add_end_docstrings(lowerCAmelCase_ )
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
def __init__( self : Optional[Any], **_snake_case : str ) ->Dict:
super().__init__(**_snake_case )
if self.framework != "pt":
raise ValueError(F'''The {self.__class__} is only available in PyTorch.''' )
# No specific FOR_XXX available yet
def __call__( self : Union[str, Any], _snake_case : Union[np.ndarray, bytes, str], **_snake_case : Tuple ) ->Dict:
return super().__call__(_snake_case, **_snake_case )
def lowercase_ ( self : Tuple, **_snake_case : Any ) ->Union[str, Any]:
snake_case__ : str = {}
if "candidate_labels" in kwargs:
snake_case__ : str = kwargs['candidate_labels']
if "hypothesis_template" in kwargs:
snake_case__ : str = kwargs['hypothesis_template']
return preprocess_params, {}, {}
def lowercase_ ( self : Dict, _snake_case : str, _snake_case : Optional[int]=None, _snake_case : List[str]="This is a sound of {}." ) ->int:
if isinstance(_snake_case, _snake_case ):
if audio.startswith('http://' ) or audio.startswith('https://' ):
# We need to actually check for a real protocol, otherwise it's impossible to use a local file
# like http_huggingface_co.png
snake_case__ : List[Any] = requests.get(_snake_case ).content
else:
with open(_snake_case, 'rb' ) as f:
snake_case__ : Union[str, Any] = f.read()
if isinstance(_snake_case, _snake_case ):
snake_case__ : List[Any] = ffmpeg_read(_snake_case, self.feature_extractor.sampling_rate )
if not isinstance(_snake_case, np.ndarray ):
raise ValueError('We expect a numpy ndarray as input' )
if len(audio.shape ) != 1:
raise ValueError('We expect a single channel audio input for ZeroShotAudioClassificationPipeline' )
snake_case__ : Tuple = self.feature_extractor(
[audio], sampling_rate=self.feature_extractor.sampling_rate, return_tensors='pt' )
snake_case__ : int = candidate_labels
snake_case__ : int = [hypothesis_template.format(_snake_case ) for x in candidate_labels]
snake_case__ : Optional[int] = self.tokenizer(_snake_case, return_tensors=self.framework, padding=_snake_case )
snake_case__ : List[Any] = [text_inputs]
return inputs
def lowercase_ ( self : Optional[int], _snake_case : Optional[Any] ) ->int:
snake_case__ : Optional[int] = model_inputs.pop('candidate_labels' )
snake_case__ : str = model_inputs.pop('text_inputs' )
if isinstance(text_inputs[0], _snake_case ):
snake_case__ : Optional[Any] = text_inputs[0]
else:
# Batching case.
snake_case__ : int = text_inputs[0][0]
snake_case__ : Any = self.model(**_snake_case, **_snake_case )
snake_case__ : List[Any] = {
'candidate_labels': candidate_labels,
'logits': outputs.logits_per_audio,
}
return model_outputs
def lowercase_ ( self : Union[str, Any], _snake_case : str ) ->List[str]:
snake_case__ : int = model_outputs.pop('candidate_labels' )
snake_case__ : List[Any] = model_outputs['logits'][0]
if self.framework == "pt":
snake_case__ : Tuple = logits.softmax(dim=0 )
snake_case__ : Union[str, Any] = probs.tolist()
else:
raise ValueError('`tf` framework not supported.' )
snake_case__ : Union[str, Any] = [
{'score': score, 'label': candidate_label}
for score, candidate_label in sorted(zip(_snake_case, _snake_case ), key=lambda _snake_case : -x[0] )
]
return result
| 277 | 0 |
'''simple docstring'''
import unittest
from transformers import MODEL_FOR_ZERO_SHOT_OBJECT_DETECTION_MAPPING, is_vision_available, pipeline
from transformers.testing_utils import (
is_pipeline_test,
nested_simplify,
require_tf,
require_torch,
require_vision,
slow,
)
from .test_pipelines_common import ANY
if is_vision_available():
from PIL import Image
else:
class _lowercase :
'''simple docstring'''
@staticmethod
def a ( *SCREAMING_SNAKE_CASE__ : Optional[Any] , **SCREAMING_SNAKE_CASE__ : int ) -> Optional[int]:
pass
@is_pipeline_test
@require_vision
@require_torch
class _lowercase ( unittest.TestCase ):
'''simple docstring'''
_SCREAMING_SNAKE_CASE : Dict = MODEL_FOR_ZERO_SHOT_OBJECT_DETECTION_MAPPING
def a ( self : Optional[int] , SCREAMING_SNAKE_CASE__ : Dict , SCREAMING_SNAKE_CASE__ : Dict , SCREAMING_SNAKE_CASE__ : Any ) -> Dict:
__lowerCAmelCase = pipeline(
"""zero-shot-object-detection""" , model="""hf-internal-testing/tiny-random-owlvit-object-detection""" )
__lowerCAmelCase = [
{
'image': './tests/fixtures/tests_samples/COCO/000000039769.png',
'candidate_labels': ['cat', 'remote', 'couch'],
}
]
return object_detector, examples
def a ( self : Union[str, Any] , SCREAMING_SNAKE_CASE__ : Optional[int] , SCREAMING_SNAKE_CASE__ : Optional[int] ) -> Any:
__lowerCAmelCase = object_detector(examples[0] , threshold=0.0 )
__lowerCAmelCase = len(_snake_case )
self.assertGreater(_snake_case , 0 )
self.assertEqual(
_snake_case , [
{
"""score""": ANY(_snake_case ),
"""label""": ANY(_snake_case ),
"""box""": {"""xmin""": ANY(_snake_case ), """ymin""": ANY(_snake_case ), """xmax""": ANY(_snake_case ), """ymax""": ANY(_snake_case )},
}
for i in range(_snake_case )
] , )
@require_tf
@unittest.skip("""Zero Shot Object Detection not implemented in TF""" )
def a ( self : int ) -> Optional[int]:
pass
@require_torch
def a ( self : Any ) -> Tuple:
__lowerCAmelCase = pipeline(
"""zero-shot-object-detection""" , model="""hf-internal-testing/tiny-random-owlvit-object-detection""" )
__lowerCAmelCase = object_detector(
"""./tests/fixtures/tests_samples/COCO/000000039769.png""" , candidate_labels=["""cat""", """remote""", """couch"""] , threshold=0.6_4 , )
self.assertEqual(
nested_simplify(_snake_case , decimals=4 ) , [
{"""score""": 0.7_2_3_5, """label""": """cat""", """box""": {"""xmin""": 2_04, """ymin""": 1_67, """xmax""": 2_32, """ymax""": 1_90}},
{"""score""": 0.7_2_1_8, """label""": """remote""", """box""": {"""xmin""": 2_04, """ymin""": 1_67, """xmax""": 2_32, """ymax""": 1_90}},
{"""score""": 0.7_1_8_4, """label""": """couch""", """box""": {"""xmin""": 2_04, """ymin""": 1_67, """xmax""": 2_32, """ymax""": 1_90}},
{"""score""": 0.6_7_4_8, """label""": """remote""", """box""": {"""xmin""": 5_71, """ymin""": 83, """xmax""": 5_98, """ymax""": 1_03}},
{"""score""": 0.6_6_5_6, """label""": """cat""", """box""": {"""xmin""": 5_71, """ymin""": 83, """xmax""": 5_98, """ymax""": 1_03}},
{"""score""": 0.6_6_1_4, """label""": """couch""", """box""": {"""xmin""": 5_71, """ymin""": 83, """xmax""": 5_98, """ymax""": 1_03}},
{"""score""": 0.6_4_5_6, """label""": """remote""", """box""": {"""xmin""": 4_94, """ymin""": 1_05, """xmax""": 5_21, """ymax""": 1_27}},
{"""score""": 0.6_4_2, """label""": """remote""", """box""": {"""xmin""": 67, """ymin""": 2_74, """xmax""": 93, """ymax""": 2_97}},
{"""score""": 0.6_4_1_9, """label""": """cat""", """box""": {"""xmin""": 4_94, """ymin""": 1_05, """xmax""": 5_21, """ymax""": 1_27}},
] , )
__lowerCAmelCase = object_detector(
[
{
"""image""": """./tests/fixtures/tests_samples/COCO/000000039769.png""",
"""candidate_labels""": ["""cat""", """remote""", """couch"""],
}
] , threshold=0.6_4 , )
self.assertEqual(
nested_simplify(_snake_case , decimals=4 ) , [
[
{"""score""": 0.7_2_3_5, """label""": """cat""", """box""": {"""xmin""": 2_04, """ymin""": 1_67, """xmax""": 2_32, """ymax""": 1_90}},
{"""score""": 0.7_2_1_8, """label""": """remote""", """box""": {"""xmin""": 2_04, """ymin""": 1_67, """xmax""": 2_32, """ymax""": 1_90}},
{"""score""": 0.7_1_8_4, """label""": """couch""", """box""": {"""xmin""": 2_04, """ymin""": 1_67, """xmax""": 2_32, """ymax""": 1_90}},
{"""score""": 0.6_7_4_8, """label""": """remote""", """box""": {"""xmin""": 5_71, """ymin""": 83, """xmax""": 5_98, """ymax""": 1_03}},
{"""score""": 0.6_6_5_6, """label""": """cat""", """box""": {"""xmin""": 5_71, """ymin""": 83, """xmax""": 5_98, """ymax""": 1_03}},
{"""score""": 0.6_6_1_4, """label""": """couch""", """box""": {"""xmin""": 5_71, """ymin""": 83, """xmax""": 5_98, """ymax""": 1_03}},
{"""score""": 0.6_4_5_6, """label""": """remote""", """box""": {"""xmin""": 4_94, """ymin""": 1_05, """xmax""": 5_21, """ymax""": 1_27}},
{"""score""": 0.6_4_2, """label""": """remote""", """box""": {"""xmin""": 67, """ymin""": 2_74, """xmax""": 93, """ymax""": 2_97}},
{"""score""": 0.6_4_1_9, """label""": """cat""", """box""": {"""xmin""": 4_94, """ymin""": 1_05, """xmax""": 5_21, """ymax""": 1_27}},
]
] , )
@require_torch
@slow
def a ( self : List[Any] ) -> Dict:
__lowerCAmelCase = pipeline("""zero-shot-object-detection""" )
__lowerCAmelCase = object_detector(
"""http://images.cocodataset.org/val2017/000000039769.jpg""" , candidate_labels=["""cat""", """remote""", """couch"""] , )
self.assertEqual(
nested_simplify(_snake_case , decimals=4 ) , [
{"""score""": 0.2_8_6_8, """label""": """cat""", """box""": {"""xmin""": 3_24, """ymin""": 20, """xmax""": 6_40, """ymax""": 3_73}},
{"""score""": 0.2_7_7, """label""": """remote""", """box""": {"""xmin""": 40, """ymin""": 72, """xmax""": 1_77, """ymax""": 1_15}},
{"""score""": 0.2_5_3_7, """label""": """cat""", """box""": {"""xmin""": 1, """ymin""": 55, """xmax""": 3_15, """ymax""": 4_72}},
{"""score""": 0.1_4_7_4, """label""": """remote""", """box""": {"""xmin""": 3_35, """ymin""": 74, """xmax""": 3_71, """ymax""": 1_87}},
{"""score""": 0.1_2_0_8, """label""": """couch""", """box""": {"""xmin""": 4, """ymin""": 0, """xmax""": 6_42, """ymax""": 4_76}},
] , )
__lowerCAmelCase = object_detector(
[
{
"""image""": """http://images.cocodataset.org/val2017/000000039769.jpg""",
"""candidate_labels""": ["""cat""", """remote""", """couch"""],
},
{
"""image""": """http://images.cocodataset.org/val2017/000000039769.jpg""",
"""candidate_labels""": ["""cat""", """remote""", """couch"""],
},
] , )
self.assertEqual(
nested_simplify(_snake_case , decimals=4 ) , [
[
{"""score""": 0.2_8_6_8, """label""": """cat""", """box""": {"""xmin""": 3_24, """ymin""": 20, """xmax""": 6_40, """ymax""": 3_73}},
{"""score""": 0.2_7_7, """label""": """remote""", """box""": {"""xmin""": 40, """ymin""": 72, """xmax""": 1_77, """ymax""": 1_15}},
{"""score""": 0.2_5_3_7, """label""": """cat""", """box""": {"""xmin""": 1, """ymin""": 55, """xmax""": 3_15, """ymax""": 4_72}},
{"""score""": 0.1_4_7_4, """label""": """remote""", """box""": {"""xmin""": 3_35, """ymin""": 74, """xmax""": 3_71, """ymax""": 1_87}},
{"""score""": 0.1_2_0_8, """label""": """couch""", """box""": {"""xmin""": 4, """ymin""": 0, """xmax""": 6_42, """ymax""": 4_76}},
],
[
{"""score""": 0.2_8_6_8, """label""": """cat""", """box""": {"""xmin""": 3_24, """ymin""": 20, """xmax""": 6_40, """ymax""": 3_73}},
{"""score""": 0.2_7_7, """label""": """remote""", """box""": {"""xmin""": 40, """ymin""": 72, """xmax""": 1_77, """ymax""": 1_15}},
{"""score""": 0.2_5_3_7, """label""": """cat""", """box""": {"""xmin""": 1, """ymin""": 55, """xmax""": 3_15, """ymax""": 4_72}},
{"""score""": 0.1_4_7_4, """label""": """remote""", """box""": {"""xmin""": 3_35, """ymin""": 74, """xmax""": 3_71, """ymax""": 1_87}},
{"""score""": 0.1_2_0_8, """label""": """couch""", """box""": {"""xmin""": 4, """ymin""": 0, """xmax""": 6_42, """ymax""": 4_76}},
],
] , )
@require_tf
@unittest.skip("""Zero Shot Object Detection not implemented in TF""" )
def a ( self : List[str] ) -> Any:
pass
@require_torch
@slow
def a ( self : Optional[int] ) -> Tuple:
__lowerCAmelCase = 0.2
__lowerCAmelCase = pipeline("""zero-shot-object-detection""" )
__lowerCAmelCase = object_detector(
"""http://images.cocodataset.org/val2017/000000039769.jpg""" , candidate_labels=["""cat""", """remote""", """couch"""] , threshold=_snake_case , )
self.assertEqual(
nested_simplify(_snake_case , decimals=4 ) , [
{"""score""": 0.2_8_6_8, """label""": """cat""", """box""": {"""xmin""": 3_24, """ymin""": 20, """xmax""": 6_40, """ymax""": 3_73}},
{"""score""": 0.2_7_7, """label""": """remote""", """box""": {"""xmin""": 40, """ymin""": 72, """xmax""": 1_77, """ymax""": 1_15}},
{"""score""": 0.2_5_3_7, """label""": """cat""", """box""": {"""xmin""": 1, """ymin""": 55, """xmax""": 3_15, """ymax""": 4_72}},
] , )
@require_torch
@slow
def a ( self : List[Any] ) -> str:
__lowerCAmelCase = 2
__lowerCAmelCase = pipeline("""zero-shot-object-detection""" )
__lowerCAmelCase = object_detector(
"""http://images.cocodataset.org/val2017/000000039769.jpg""" , candidate_labels=["""cat""", """remote""", """couch"""] , top_k=_snake_case , )
self.assertEqual(
nested_simplify(_snake_case , decimals=4 ) , [
{"""score""": 0.2_8_6_8, """label""": """cat""", """box""": {"""xmin""": 3_24, """ymin""": 20, """xmax""": 6_40, """ymax""": 3_73}},
{"""score""": 0.2_7_7, """label""": """remote""", """box""": {"""xmin""": 40, """ymin""": 72, """xmax""": 1_77, """ymax""": 1_15}},
] , )
| 229 |
import collections
import inspect
import unittest
from transformers import SwinvaConfig
from transformers.testing_utils import require_torch, require_vision, slow, torch_device
from transformers.utils import cached_property, is_torch_available, is_vision_available
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, _config_zero_init, floats_tensor, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from torch import nn
from transformers import SwinvaForImageClassification, SwinvaForMaskedImageModeling, SwinvaModel
from transformers.models.swinva.modeling_swinva import SWINV2_PRETRAINED_MODEL_ARCHIVE_LIST
if is_vision_available():
from PIL import Image
from transformers import AutoImageProcessor
class snake_case__ :
"""simple docstring"""
def __init__( self : Tuple, _snake_case : Any, _snake_case : int=1_3, _snake_case : Optional[int]=3_2, _snake_case : Tuple=2, _snake_case : Any=3, _snake_case : Tuple=1_6, _snake_case : Tuple=[1, 2, 1], _snake_case : Dict=[2, 2, 4], _snake_case : str=2, _snake_case : Union[str, Any]=2.0, _snake_case : Dict=True, _snake_case : Dict=0.0, _snake_case : str=0.0, _snake_case : str=0.1, _snake_case : List[str]="gelu", _snake_case : int=False, _snake_case : Optional[Any]=True, _snake_case : List[Any]=0.0_2, _snake_case : Union[str, Any]=1e-5, _snake_case : Union[str, Any]=True, _snake_case : List[Any]=None, _snake_case : Any=True, _snake_case : List[Any]=1_0, _snake_case : str=8, ) ->Union[str, Any]:
snake_case__ : Any = parent
snake_case__ : Tuple = batch_size
snake_case__ : Tuple = image_size
snake_case__ : Any = patch_size
snake_case__ : Optional[int] = num_channels
snake_case__ : Tuple = embed_dim
snake_case__ : Any = depths
snake_case__ : Any = num_heads
snake_case__ : List[str] = window_size
snake_case__ : Dict = mlp_ratio
snake_case__ : Optional[int] = qkv_bias
snake_case__ : Optional[Any] = hidden_dropout_prob
snake_case__ : List[str] = attention_probs_dropout_prob
snake_case__ : Union[str, Any] = drop_path_rate
snake_case__ : str = hidden_act
snake_case__ : Union[str, Any] = use_absolute_embeddings
snake_case__ : Union[str, Any] = patch_norm
snake_case__ : Any = layer_norm_eps
snake_case__ : Tuple = initializer_range
snake_case__ : Dict = is_training
snake_case__ : Any = scope
snake_case__ : Optional[Any] = use_labels
snake_case__ : str = type_sequence_label_size
snake_case__ : List[Any] = encoder_stride
def lowercase_ ( self : Tuple ) ->str:
snake_case__ : Tuple = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size] )
snake_case__ : List[Any] = None
if self.use_labels:
snake_case__ : Optional[Any] = ids_tensor([self.batch_size], self.type_sequence_label_size )
snake_case__ : Any = self.get_config()
return config, pixel_values, labels
def lowercase_ ( self : Optional[int] ) ->Optional[int]:
return SwinvaConfig(
image_size=self.image_size, patch_size=self.patch_size, num_channels=self.num_channels, embed_dim=self.embed_dim, depths=self.depths, num_heads=self.num_heads, window_size=self.window_size, mlp_ratio=self.mlp_ratio, qkv_bias=self.qkv_bias, hidden_dropout_prob=self.hidden_dropout_prob, attention_probs_dropout_prob=self.attention_probs_dropout_prob, drop_path_rate=self.drop_path_rate, hidden_act=self.hidden_act, use_absolute_embeddings=self.use_absolute_embeddings, path_norm=self.patch_norm, layer_norm_eps=self.layer_norm_eps, initializer_range=self.initializer_range, encoder_stride=self.encoder_stride, )
def lowercase_ ( self : Optional[int], _snake_case : str, _snake_case : List[str], _snake_case : int ) ->Dict:
snake_case__ : List[Any] = SwinvaModel(config=_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Optional[int] = model(_snake_case )
snake_case__ : List[Any] = ((config.image_size // config.patch_size) ** 2) // (4 ** (len(config.depths ) - 1))
snake_case__ : List[Any] = int(config.embed_dim * 2 ** (len(config.depths ) - 1) )
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, expected_seq_len, expected_dim) )
def lowercase_ ( self : Optional[Any], _snake_case : Any, _snake_case : List[str], _snake_case : Dict ) ->List[Any]:
snake_case__ : List[str] = SwinvaForMaskedImageModeling(config=_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Union[str, Any] = model(_snake_case )
self.parent.assertEqual(
result.logits.shape, (self.batch_size, self.num_channels, self.image_size, self.image_size) )
# test greyscale images
snake_case__ : Optional[Any] = 1
snake_case__ : Optional[int] = SwinvaForMaskedImageModeling(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Tuple = floats_tensor([self.batch_size, 1, self.image_size, self.image_size] )
snake_case__ : Any = model(_snake_case )
self.parent.assertEqual(result.logits.shape, (self.batch_size, 1, self.image_size, self.image_size) )
def lowercase_ ( self : List[str], _snake_case : int, _snake_case : List[Any], _snake_case : Optional[int] ) ->Any:
snake_case__ : Tuple = self.type_sequence_label_size
snake_case__ : int = SwinvaForImageClassification(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Tuple = model(_snake_case, labels=_snake_case )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.type_sequence_label_size) )
def lowercase_ ( self : Any ) ->Dict:
snake_case__ : str = self.prepare_config_and_inputs()
snake_case__ , snake_case__ , snake_case__ : List[str] = config_and_inputs
snake_case__ : Union[str, Any] = {'pixel_values': pixel_values}
return config, inputs_dict
@require_torch
class snake_case__ ( lowerCAmelCase_ , lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = (
(SwinvaModel, SwinvaForImageClassification, SwinvaForMaskedImageModeling) if is_torch_available() else ()
)
_SCREAMING_SNAKE_CASE = (
{"""feature-extraction""": SwinvaModel, """image-classification""": SwinvaForImageClassification}
if is_torch_available()
else {}
)
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
def lowercase_ ( self : Union[str, Any] ) ->Dict:
snake_case__ : Optional[int] = SwinvaModelTester(self )
snake_case__ : int = ConfigTester(self, config_class=_snake_case, embed_dim=3_7 )
def lowercase_ ( self : Tuple ) ->int:
self.config_tester.create_and_test_config_to_json_string()
self.config_tester.create_and_test_config_to_json_file()
self.config_tester.create_and_test_config_from_and_save_pretrained()
self.config_tester.create_and_test_config_with_num_labels()
self.config_tester.check_config_can_be_init_without_params()
self.config_tester.check_config_arguments_init()
def lowercase_ ( self : Any ) ->str:
snake_case__ : int = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*_snake_case )
@unittest.skip(reason='Got `CUDA error: misaligned address` with PyTorch 2.0.0.' )
def lowercase_ ( self : Any ) ->Union[str, Any]:
pass
@unittest.skip(reason='Swinv2 does not use inputs_embeds' )
def lowercase_ ( self : str ) ->Union[str, Any]:
pass
def lowercase_ ( self : Optional[Any] ) ->Union[str, Any]:
snake_case__ , snake_case__ : Optional[int] = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
snake_case__ : Union[str, Any] = model_class(_snake_case )
self.assertIsInstance(model.get_input_embeddings(), (nn.Module) )
snake_case__ : Union[str, Any] = model.get_output_embeddings()
self.assertTrue(x is None or isinstance(_snake_case, nn.Linear ) )
def lowercase_ ( self : List[str] ) ->Optional[int]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
snake_case__ : Any = model_class(_snake_case )
snake_case__ : Dict = inspect.signature(model.forward )
# signature.parameters is an OrderedDict => so arg_names order is deterministic
snake_case__ : Optional[Any] = [*signature.parameters.keys()]
snake_case__ : List[Any] = ['pixel_values']
self.assertListEqual(arg_names[:1], _snake_case )
def lowercase_ ( self : str ) ->Union[str, Any]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : int = True
for model_class in self.all_model_classes:
snake_case__ : str = True
snake_case__ : Union[str, Any] = False
snake_case__ : Tuple = True
snake_case__ : int = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : Optional[int] = model(**self._prepare_for_class(_snake_case, _snake_case ) )
snake_case__ : List[str] = outputs.attentions
snake_case__ : List[Any] = len(self.model_tester.depths )
self.assertEqual(len(_snake_case ), _snake_case )
# check that output_attentions also work using config
del inputs_dict["output_attentions"]
snake_case__ : str = True
snake_case__ : Tuple = config.window_size**2
snake_case__ : Optional[int] = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : str = model(**self._prepare_for_class(_snake_case, _snake_case ) )
snake_case__ : Tuple = outputs.attentions
self.assertEqual(len(_snake_case ), _snake_case )
self.assertListEqual(
list(attentions[0].shape[-3:] ), [self.model_tester.num_heads[0], window_size_squared, window_size_squared], )
snake_case__ : Optional[Any] = len(_snake_case )
# Check attention is always last and order is fine
snake_case__ : Optional[int] = True
snake_case__ : Dict = True
snake_case__ : List[Any] = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : Optional[int] = model(**self._prepare_for_class(_snake_case, _snake_case ) )
if hasattr(self.model_tester, 'num_hidden_states_types' ):
snake_case__ : str = self.model_tester.num_hidden_states_types
else:
# also another +1 for reshaped_hidden_states
snake_case__ : Dict = 2
self.assertEqual(out_len + added_hidden_states, len(_snake_case ) )
snake_case__ : Any = outputs.attentions
self.assertEqual(len(_snake_case ), _snake_case )
self.assertListEqual(
list(self_attentions[0].shape[-3:] ), [self.model_tester.num_heads[0], window_size_squared, window_size_squared], )
def lowercase_ ( self : Dict, _snake_case : Tuple, _snake_case : Any, _snake_case : int, _snake_case : Optional[int] ) ->str:
snake_case__ : Dict = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : List[Any] = model(**self._prepare_for_class(_snake_case, _snake_case ) )
snake_case__ : Dict = outputs.hidden_states
snake_case__ : int = getattr(
self.model_tester, 'expected_num_hidden_layers', len(self.model_tester.depths ) + 1 )
self.assertEqual(len(_snake_case ), _snake_case )
# Swinv2 has a different seq_length
snake_case__ : int = (
config.patch_size
if isinstance(config.patch_size, collections.abc.Iterable )
else (config.patch_size, config.patch_size)
)
snake_case__ : Optional[Any] = (image_size[1] // patch_size[1]) * (image_size[0] // patch_size[0])
self.assertListEqual(
list(hidden_states[0].shape[-2:] ), [num_patches, self.model_tester.embed_dim], )
snake_case__ : Union[str, Any] = outputs.reshaped_hidden_states
self.assertEqual(len(_snake_case ), _snake_case )
snake_case__ , snake_case__ , snake_case__ , snake_case__ : str = reshaped_hidden_states[0].shape
snake_case__ : Any = (
reshaped_hidden_states[0].view(_snake_case, _snake_case, height * width ).permute(0, 2, 1 )
)
self.assertListEqual(
list(reshaped_hidden_states.shape[-2:] ), [num_patches, self.model_tester.embed_dim], )
def lowercase_ ( self : str ) ->List[Any]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : List[Any] = (
self.model_tester.image_size
if isinstance(self.model_tester.image_size, collections.abc.Iterable )
else (self.model_tester.image_size, self.model_tester.image_size)
)
for model_class in self.all_model_classes:
snake_case__ : Optional[int] = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, _snake_case )
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
snake_case__ : Dict = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, _snake_case )
def lowercase_ ( self : List[str] ) ->str:
snake_case__ , snake_case__ : Optional[Any] = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : List[str] = 3
snake_case__ : Union[str, Any] = (
self.model_tester.image_size
if isinstance(self.model_tester.image_size, collections.abc.Iterable )
else (self.model_tester.image_size, self.model_tester.image_size)
)
snake_case__ : str = (
config.patch_size
if isinstance(config.patch_size, collections.abc.Iterable )
else (config.patch_size, config.patch_size)
)
snake_case__ : Tuple = image_size[0] + patch_size[0] - (image_size[0] % patch_size[0])
snake_case__ : Optional[Any] = image_size[1] + patch_size[1] - (image_size[1] % patch_size[1])
for model_class in self.all_model_classes:
snake_case__ : int = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, (padded_height, padded_width) )
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
snake_case__ : List[str] = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, (padded_height, padded_width) )
def lowercase_ ( self : List[str] ) ->Optional[int]:
snake_case__ : Any = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_masked_image_modeling(*_snake_case )
def lowercase_ ( self : List[Any] ) ->str:
snake_case__ : Any = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_image_classification(*_snake_case )
@slow
def lowercase_ ( self : str ) ->Union[str, Any]:
for model_name in SWINV2_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
snake_case__ : Dict = SwinvaModel.from_pretrained(_snake_case )
self.assertIsNotNone(_snake_case )
def lowercase_ ( self : Optional[int] ) ->List[str]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : List[Any] = _config_zero_init(_snake_case )
for model_class in self.all_model_classes:
snake_case__ : List[str] = model_class(config=_snake_case )
for name, param in model.named_parameters():
if "embeddings" not in name and "logit_scale" not in name and param.requires_grad:
self.assertIn(
((param.data.mean() * 1e9).round() / 1e9).item(), [0.0, 1.0], msg=F'''Parameter {name} of model {model_class} seems not properly initialized''', )
@require_vision
@require_torch
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
@cached_property
def lowercase_ ( self : Union[str, Any] ) ->List[str]:
return (
AutoImageProcessor.from_pretrained('microsoft/swinv2-tiny-patch4-window8-256' )
if is_vision_available()
else None
)
@slow
def lowercase_ ( self : int ) ->List[Any]:
snake_case__ : Any = SwinvaForImageClassification.from_pretrained('microsoft/swinv2-tiny-patch4-window8-256' ).to(
_snake_case )
snake_case__ : int = self.default_image_processor
snake_case__ : Union[str, Any] = Image.open('./tests/fixtures/tests_samples/COCO/000000039769.png' )
snake_case__ : Optional[Any] = image_processor(images=_snake_case, return_tensors='pt' ).to(_snake_case )
# forward pass
with torch.no_grad():
snake_case__ : List[str] = model(**_snake_case )
# verify the logits
snake_case__ : int = torch.Size((1, 1_0_0_0) )
self.assertEqual(outputs.logits.shape, _snake_case )
snake_case__ : Optional[int] = torch.tensor([-0.3_9_4_7, -0.4_3_0_6, 0.0_0_2_6] ).to(_snake_case )
self.assertTrue(torch.allclose(outputs.logits[0, :3], _snake_case, atol=1e-4 ) )
| 277 | 0 |
"""simple docstring"""
import argparse
import json
import pickle
from pathlib import Path
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import MaskFormerConfig, MaskFormerForInstanceSegmentation, MaskFormerImageProcessor, SwinConfig
from transformers.utils import logging
logging.set_verbosity_info()
UpperCAmelCase = logging.get_logger(__name__)
def lowercase ( a__ : str ) -> List[Any]:
_UpperCamelCase = SwinConfig.from_pretrained(
'''microsoft/swin-tiny-patch4-window7-224''' , out_features=['''stage1''', '''stage2''', '''stage3''', '''stage4'''] )
_UpperCamelCase = MaskFormerConfig(backbone_config=a__ )
_UpperCamelCase = 'huggingface/label-files'
if "ade20k-full" in model_name:
# this should be ok
_UpperCamelCase = 847
_UpperCamelCase = 'maskformer-ade20k-full-id2label.json'
elif "ade" in model_name:
# this should be ok
_UpperCamelCase = 150
_UpperCamelCase = 'ade20k-id2label.json'
elif "coco-stuff" in model_name:
# this should be ok
_UpperCamelCase = 171
_UpperCamelCase = 'maskformer-coco-stuff-id2label.json'
elif "coco" in model_name:
# TODO
_UpperCamelCase = 133
_UpperCamelCase = 'coco-panoptic-id2label.json'
elif "cityscapes" in model_name:
# this should be ok
_UpperCamelCase = 19
_UpperCamelCase = 'cityscapes-id2label.json'
elif "vistas" in model_name:
# this should be ok
_UpperCamelCase = 65
_UpperCamelCase = 'mapillary-vistas-id2label.json'
_UpperCamelCase = json.load(open(hf_hub_download(a__ , a__ , repo_type='''dataset''' ) , '''r''' ) )
_UpperCamelCase = {int(a__ ): v for k, v in idalabel.items()}
return config
def lowercase ( a__ : Any ) -> List[Any]:
_UpperCamelCase = []
# stem
# fmt: off
rename_keys.append(('''backbone.patch_embed.proj.weight''', '''model.pixel_level_module.encoder.model.embeddings.patch_embeddings.projection.weight''') )
rename_keys.append(('''backbone.patch_embed.proj.bias''', '''model.pixel_level_module.encoder.model.embeddings.patch_embeddings.projection.bias''') )
rename_keys.append(('''backbone.patch_embed.norm.weight''', '''model.pixel_level_module.encoder.model.embeddings.norm.weight''') )
rename_keys.append(('''backbone.patch_embed.norm.bias''', '''model.pixel_level_module.encoder.model.embeddings.norm.bias''') )
# stages
for i in range(len(config.backbone_config.depths ) ):
for j in range(config.backbone_config.depths[i] ):
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm1.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_before.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm1.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_before.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.relative_position_bias_table''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.self.relative_position_bias_table''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.relative_position_index''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.self.relative_position_index''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.proj.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.output.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.proj.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.output.dense.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm2.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_after.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm2.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_after.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc1.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.intermediate.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc1.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.intermediate.dense.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc2.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.output.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc2.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.output.dense.bias''') )
if i < 3:
rename_keys.append((F'''backbone.layers.{i}.downsample.reduction.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.reduction.weight''') )
rename_keys.append((F'''backbone.layers.{i}.downsample.norm.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.norm.weight''') )
rename_keys.append((F'''backbone.layers.{i}.downsample.norm.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.norm.bias''') )
rename_keys.append((F'''backbone.norm{i}.weight''', F'''model.pixel_level_module.encoder.hidden_states_norms.{i}.weight''') )
rename_keys.append((F'''backbone.norm{i}.bias''', F'''model.pixel_level_module.encoder.hidden_states_norms.{i}.bias''') )
# FPN
rename_keys.append(('''sem_seg_head.layer_4.weight''', '''model.pixel_level_module.decoder.fpn.stem.0.weight''') )
rename_keys.append(('''sem_seg_head.layer_4.norm.weight''', '''model.pixel_level_module.decoder.fpn.stem.1.weight''') )
rename_keys.append(('''sem_seg_head.layer_4.norm.bias''', '''model.pixel_level_module.decoder.fpn.stem.1.bias''') )
for source_index, target_index in zip(range(3 , 0 , -1 ) , range(0 , 3 ) ):
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.0.weight''') )
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.norm.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.1.weight''') )
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.norm.bias''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.1.bias''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.0.weight''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.norm.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.1.weight''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.norm.bias''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.1.bias''') )
rename_keys.append(('''sem_seg_head.mask_features.weight''', '''model.pixel_level_module.decoder.mask_projection.weight''') )
rename_keys.append(('''sem_seg_head.mask_features.bias''', '''model.pixel_level_module.decoder.mask_projection.bias''') )
# Transformer decoder
for idx in range(config.decoder_config.decoder_layers ):
# self-attention out projection
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.out_proj.weight''', F'''model.transformer_module.decoder.layers.{idx}.self_attn.out_proj.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.out_proj.bias''', F'''model.transformer_module.decoder.layers.{idx}.self_attn.out_proj.bias''') )
# cross-attention out projection
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.out_proj.weight''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn.out_proj.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.out_proj.bias''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn.out_proj.bias''') )
# MLP 1
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear1.weight''', F'''model.transformer_module.decoder.layers.{idx}.fc1.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear1.bias''', F'''model.transformer_module.decoder.layers.{idx}.fc1.bias''') )
# MLP 2
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear2.weight''', F'''model.transformer_module.decoder.layers.{idx}.fc2.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear2.bias''', F'''model.transformer_module.decoder.layers.{idx}.fc2.bias''') )
# layernorm 1 (self-attention layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm1.weight''', F'''model.transformer_module.decoder.layers.{idx}.self_attn_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm1.bias''', F'''model.transformer_module.decoder.layers.{idx}.self_attn_layer_norm.bias''') )
# layernorm 2 (cross-attention layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm2.weight''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm2.bias''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn_layer_norm.bias''') )
# layernorm 3 (final layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm3.weight''', F'''model.transformer_module.decoder.layers.{idx}.final_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm3.bias''', F'''model.transformer_module.decoder.layers.{idx}.final_layer_norm.bias''') )
rename_keys.append(('''sem_seg_head.predictor.transformer.decoder.norm.weight''', '''model.transformer_module.decoder.layernorm.weight''') )
rename_keys.append(('''sem_seg_head.predictor.transformer.decoder.norm.bias''', '''model.transformer_module.decoder.layernorm.bias''') )
# heads on top
rename_keys.append(('''sem_seg_head.predictor.query_embed.weight''', '''model.transformer_module.queries_embedder.weight''') )
rename_keys.append(('''sem_seg_head.predictor.input_proj.weight''', '''model.transformer_module.input_projection.weight''') )
rename_keys.append(('''sem_seg_head.predictor.input_proj.bias''', '''model.transformer_module.input_projection.bias''') )
rename_keys.append(('''sem_seg_head.predictor.class_embed.weight''', '''class_predictor.weight''') )
rename_keys.append(('''sem_seg_head.predictor.class_embed.bias''', '''class_predictor.bias''') )
for i in range(3 ):
rename_keys.append((F'''sem_seg_head.predictor.mask_embed.layers.{i}.weight''', F'''mask_embedder.{i}.0.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.mask_embed.layers.{i}.bias''', F'''mask_embedder.{i}.0.bias''') )
# fmt: on
return rename_keys
def lowercase ( a__ : Tuple , a__ : Tuple , a__ : Optional[Any] ) -> Any:
_UpperCamelCase = dct.pop(a__ )
_UpperCamelCase = val
def lowercase ( a__ : Optional[Any] , a__ : Tuple ) -> Union[str, Any]:
_UpperCamelCase = [int(backbone_config.embed_dim * 2**i ) for i in range(len(backbone_config.depths ) )]
for i in range(len(backbone_config.depths ) ):
_UpperCamelCase = num_features[i]
for j in range(backbone_config.depths[i] ):
# fmt: off
# read in weights + bias of input projection layer (in original implementation, this is a single matrix + bias)
_UpperCamelCase = state_dict.pop(F'''backbone.layers.{i}.blocks.{j}.attn.qkv.weight''' )
_UpperCamelCase = state_dict.pop(F'''backbone.layers.{i}.blocks.{j}.attn.qkv.bias''' )
# next, add query, keys and values (in that order) to the state dict
_UpperCamelCase = in_proj_weight[:dim, :]
_UpperCamelCase = in_proj_bias[: dim]
_UpperCamelCase = in_proj_weight[
dim : dim * 2, :
]
_UpperCamelCase = in_proj_bias[
dim : dim * 2
]
_UpperCamelCase = in_proj_weight[
-dim :, :
]
_UpperCamelCase = in_proj_bias[-dim :]
# fmt: on
def lowercase ( a__ : List[str] , a__ : List[Any] ) -> Union[str, Any]:
# fmt: off
_UpperCamelCase = config.decoder_config.hidden_size
for idx in range(config.decoder_config.decoder_layers ):
# read in weights + bias of self-attention input projection layer (in the original implementation, this is a single matrix + bias)
_UpperCamelCase = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.in_proj_weight''' )
_UpperCamelCase = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.in_proj_bias''' )
# next, add query, keys and values (in that order) to the state dict
_UpperCamelCase = in_proj_weight[: hidden_size, :]
_UpperCamelCase = in_proj_bias[:config.hidden_size]
_UpperCamelCase = in_proj_weight[hidden_size : hidden_size * 2, :]
_UpperCamelCase = in_proj_bias[hidden_size : hidden_size * 2]
_UpperCamelCase = in_proj_weight[-hidden_size :, :]
_UpperCamelCase = in_proj_bias[-hidden_size :]
# read in weights + bias of cross-attention input projection layer (in the original implementation, this is a single matrix + bias)
_UpperCamelCase = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.in_proj_weight''' )
_UpperCamelCase = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.in_proj_bias''' )
# next, add query, keys and values (in that order) to the state dict
_UpperCamelCase = in_proj_weight[: hidden_size, :]
_UpperCamelCase = in_proj_bias[:config.hidden_size]
_UpperCamelCase = in_proj_weight[hidden_size : hidden_size * 2, :]
_UpperCamelCase = in_proj_bias[hidden_size : hidden_size * 2]
_UpperCamelCase = in_proj_weight[-hidden_size :, :]
_UpperCamelCase = in_proj_bias[-hidden_size :]
# fmt: on
def lowercase ( ) -> Tuple:
_UpperCamelCase = 'http://images.cocodataset.org/val2017/000000039769.jpg'
_UpperCamelCase = Image.open(requests.get(a__ , stream=a__ ).raw )
return im
@torch.no_grad()
def lowercase ( a__ : str , a__ : str , a__ : str , a__ : bool = False ) -> List[Any]:
_UpperCamelCase = get_maskformer_config(a__ )
# load original state_dict
with open(a__ , '''rb''' ) as f:
_UpperCamelCase = pickle.load(a__ )
_UpperCamelCase = data['model']
# for name, param in state_dict.items():
# print(name, param.shape)
# rename keys
_UpperCamelCase = create_rename_keys(a__ )
for src, dest in rename_keys:
rename_key(a__ , a__ , a__ )
read_in_swin_q_k_v(a__ , config.backbone_config )
read_in_decoder_q_k_v(a__ , a__ )
# update to torch tensors
for key, value in state_dict.items():
_UpperCamelCase = torch.from_numpy(a__ )
# load 🤗 model
_UpperCamelCase = MaskFormerForInstanceSegmentation(a__ )
model.eval()
for name, param in model.named_parameters():
print(a__ , param.shape )
_UpperCamelCase = model.load_state_dict(a__ , strict=a__ )
assert missing_keys == [
"model.pixel_level_module.encoder.model.layernorm.weight",
"model.pixel_level_module.encoder.model.layernorm.bias",
]
assert len(a__ ) == 0, F'''Unexpected keys: {unexpected_keys}'''
# verify results
_UpperCamelCase = prepare_img()
if "vistas" in model_name:
_UpperCamelCase = 65
elif "cityscapes" in model_name:
_UpperCamelCase = 65535
else:
_UpperCamelCase = 255
_UpperCamelCase = True if 'ade' in model_name else False
_UpperCamelCase = MaskFormerImageProcessor(ignore_index=a__ , reduce_labels=a__ )
_UpperCamelCase = image_processor(a__ , return_tensors='''pt''' )
_UpperCamelCase = model(**a__ )
print('''Logits:''' , outputs.class_queries_logits[0, :3, :3] )
if model_name == "maskformer-swin-tiny-ade":
_UpperCamelCase = torch.tensor(
[[3.6353, -4.4770, -2.6065], [0.5081, -4.2394, -3.5343], [2.1909, -5.0353, -1.9323]] )
assert torch.allclose(outputs.class_queries_logits[0, :3, :3] , a__ , atol=1e-4 )
print('''Looks ok!''' )
if pytorch_dump_folder_path is not None:
print(F'''Saving model and image processor to {pytorch_dump_folder_path}''' )
Path(a__ ).mkdir(exist_ok=a__ )
model.save_pretrained(a__ )
image_processor.save_pretrained(a__ )
if push_to_hub:
print('''Pushing model and image processor to the hub...''' )
model.push_to_hub(F'''nielsr/{model_name}''' )
image_processor.push_to_hub(F'''nielsr/{model_name}''' )
if __name__ == "__main__":
UpperCAmelCase = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"""--model_name""",
default="""maskformer-swin-tiny-ade""",
type=str,
help=("""Name of the MaskFormer model you'd like to convert""",),
)
parser.add_argument(
"""--checkpoint_path""",
default="""/Users/nielsrogge/Documents/MaskFormer_checkpoints/MaskFormer-Swin-tiny-ADE20k/model.pkl""",
type=str,
help="""Path to the original state dict (.pth file).""",
)
parser.add_argument(
"""--pytorch_dump_folder_path""", default=None, type=str, help="""Path to the output PyTorch model directory."""
)
parser.add_argument(
"""--push_to_hub""", action="""store_true""", help="""Whether or not to push the converted model to the 🤗 hub."""
)
UpperCAmelCase = parser.parse_args()
convert_maskformer_checkpoint(
args.model_name, args.checkpoint_path, args.pytorch_dump_folder_path, args.push_to_hub
)
| 256 |
import unittest
import numpy as np
from transformers.testing_utils import require_torch, require_vision
from transformers.utils import is_torch_available, is_vision_available
from ...test_image_processing_common import ImageProcessingSavingTestMixin, prepare_image_inputs
if is_torch_available():
import torch
if is_vision_available():
from PIL import Image
from transformers import PoolFormerImageProcessor
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
def __init__( self : Optional[int], _snake_case : List[Any], _snake_case : str=7, _snake_case : Tuple=3, _snake_case : List[str]=3_0, _snake_case : Tuple=4_0_0, _snake_case : Any=True, _snake_case : List[Any]=None, _snake_case : int=0.9, _snake_case : Optional[Any]=None, _snake_case : str=True, _snake_case : Union[str, Any]=[0.5, 0.5, 0.5], _snake_case : Union[str, Any]=[0.5, 0.5, 0.5], ) ->List[Any]:
snake_case__ : int = size if size is not None else {'shortest_edge': 3_0}
snake_case__ : Tuple = crop_size if crop_size is not None else {'height': 3_0, 'width': 3_0}
snake_case__ : Union[str, Any] = parent
snake_case__ : Dict = batch_size
snake_case__ : int = num_channels
snake_case__ : Tuple = min_resolution
snake_case__ : Any = max_resolution
snake_case__ : List[Any] = do_resize_and_center_crop
snake_case__ : str = size
snake_case__ : str = crop_pct
snake_case__ : List[str] = crop_size
snake_case__ : Optional[int] = do_normalize
snake_case__ : Tuple = image_mean
snake_case__ : Tuple = image_std
def lowercase_ ( self : Optional[int] ) ->int:
return {
"size": self.size,
"do_resize_and_center_crop": self.do_resize_and_center_crop,
"crop_pct": self.crop_pct,
"crop_size": self.crop_size,
"do_normalize": self.do_normalize,
"image_mean": self.image_mean,
"image_std": self.image_std,
}
@require_torch
@require_vision
class snake_case__ ( lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = PoolFormerImageProcessor if is_vision_available() else None
def lowercase_ ( self : Union[str, Any] ) ->Dict:
snake_case__ : Union[str, Any] = PoolFormerImageProcessingTester(self )
@property
def lowercase_ ( self : int ) ->Dict:
return self.image_processor_tester.prepare_image_processor_dict()
def lowercase_ ( self : Union[str, Any] ) ->Optional[int]:
snake_case__ : List[str] = self.image_processing_class(**self.image_processor_dict )
self.assertTrue(hasattr(_snake_case, 'do_resize_and_center_crop' ) )
self.assertTrue(hasattr(_snake_case, 'size' ) )
self.assertTrue(hasattr(_snake_case, 'crop_pct' ) )
self.assertTrue(hasattr(_snake_case, 'do_normalize' ) )
self.assertTrue(hasattr(_snake_case, 'image_mean' ) )
self.assertTrue(hasattr(_snake_case, 'image_std' ) )
def lowercase_ ( self : List[str] ) ->List[str]:
snake_case__ : Union[str, Any] = self.image_processing_class.from_dict(self.image_processor_dict )
self.assertEqual(image_processor.size, {'shortest_edge': 3_0} )
self.assertEqual(image_processor.crop_size, {'height': 3_0, 'width': 3_0} )
snake_case__ : int = self.image_processing_class.from_dict(self.image_processor_dict, size=4_2, crop_size=8_4 )
self.assertEqual(image_processor.size, {'shortest_edge': 4_2} )
self.assertEqual(image_processor.crop_size, {'height': 8_4, 'width': 8_4} )
def lowercase_ ( self : List[Any] ) ->List[Any]:
pass
def lowercase_ ( self : List[str] ) ->str:
# Initialize image_processing
snake_case__ : Union[str, Any] = self.image_processing_class(**self.image_processor_dict )
# create random PIL images
snake_case__ : List[str] = prepare_image_inputs(self.image_processor_tester, equal_resolution=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case, Image.Image )
# Test not batched input
snake_case__ : Optional[int] = image_processing(image_inputs[0], return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
# Test batched
snake_case__ : str = image_processing(_snake_case, return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
def lowercase_ ( self : int ) ->List[Any]:
# Initialize image_processing
snake_case__ : Optional[int] = self.image_processing_class(**self.image_processor_dict )
# create random numpy tensors
snake_case__ : Dict = prepare_image_inputs(self.image_processor_tester, equal_resolution=_snake_case, numpify=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case, np.ndarray )
# Test not batched input
snake_case__ : Dict = image_processing(image_inputs[0], return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
# Test batched
snake_case__ : List[Any] = image_processing(_snake_case, return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
def lowercase_ ( self : List[str] ) ->List[str]:
# Initialize image_processing
snake_case__ : Tuple = self.image_processing_class(**self.image_processor_dict )
# create random PyTorch tensors
snake_case__ : List[str] = prepare_image_inputs(self.image_processor_tester, equal_resolution=_snake_case, torchify=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case, torch.Tensor )
# Test not batched input
snake_case__ : Tuple = image_processing(image_inputs[0], return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
# Test batched
snake_case__ : Optional[Any] = image_processing(_snake_case, return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
| 277 | 0 |
import gc
import random
import unittest
import numpy as np
import torch
from PIL import Image
from transformers import XLMRobertaTokenizerFast
from diffusers import DDIMScheduler, KandinskyInpaintPipeline, KandinskyPriorPipeline, UNetaDConditionModel, VQModel
from diffusers.pipelines.kandinsky.text_encoder import MCLIPConfig, MultilingualCLIP
from diffusers.utils import floats_tensor, load_image, load_numpy, slow, torch_device
from diffusers.utils.testing_utils import enable_full_determinism, require_torch_gpu
from ..test_pipelines_common import PipelineTesterMixin, assert_mean_pixel_difference
enable_full_determinism()
class A_ ( lowerCAmelCase_ , unittest.TestCase ):
_lowercase : Optional[Any] = KandinskyInpaintPipeline
_lowercase : str = ['prompt', 'image_embeds', 'negative_image_embeds', 'image', 'mask_image']
_lowercase : str = [
'prompt',
'negative_prompt',
'image_embeds',
'negative_image_embeds',
'image',
'mask_image',
]
_lowercase : List[Any] = [
'generator',
'height',
'width',
'latents',
'guidance_scale',
'negative_prompt',
'num_inference_steps',
'return_dict',
'guidance_scale',
'num_images_per_prompt',
'output_type',
'return_dict',
]
_lowercase : str = False
@property
def UpperCAmelCase ( self : Optional[int] ) -> Optional[Any]:
return 3_2
@property
def UpperCAmelCase ( self : Dict ) -> Tuple:
return 3_2
@property
def UpperCAmelCase ( self : List[str] ) -> int:
return self.time_input_dim
@property
def UpperCAmelCase ( self : List[str] ) -> Optional[int]:
return self.time_input_dim * 4
@property
def UpperCAmelCase ( self : str ) -> List[Any]:
return 1_0_0
@property
def UpperCAmelCase ( self : Union[str, Any] ) -> Tuple:
__lowerCAmelCase: List[Any] = XLMRobertaTokenizerFast.from_pretrained('YiYiXu/tiny-random-mclip-base' )
return tokenizer
@property
def UpperCAmelCase ( self : List[Any] ) -> str:
torch.manual_seed(0 )
__lowerCAmelCase: Tuple = MCLIPConfig(
numDims=self.cross_attention_dim , transformerDimensions=self.text_embedder_hidden_size , hidden_size=self.text_embedder_hidden_size , intermediate_size=3_7 , num_attention_heads=4 , num_hidden_layers=5 , vocab_size=1_0_0_5 , )
__lowerCAmelCase: Dict = MultilingualCLIP(_snake_case )
__lowerCAmelCase: Any = text_encoder.eval()
return text_encoder
@property
def UpperCAmelCase ( self : Optional[Any] ) -> Union[str, Any]:
torch.manual_seed(0 )
__lowerCAmelCase: Union[str, Any] = {
'in_channels': 9,
# Out channels is double in channels because predicts mean and variance
'out_channels': 8,
'addition_embed_type': 'text_image',
'down_block_types': ('ResnetDownsampleBlock2D', 'SimpleCrossAttnDownBlock2D'),
'up_block_types': ('SimpleCrossAttnUpBlock2D', 'ResnetUpsampleBlock2D'),
'mid_block_type': 'UNetMidBlock2DSimpleCrossAttn',
'block_out_channels': (self.block_out_channels_a, self.block_out_channels_a * 2),
'layers_per_block': 1,
'encoder_hid_dim': self.text_embedder_hidden_size,
'encoder_hid_dim_type': 'text_image_proj',
'cross_attention_dim': self.cross_attention_dim,
'attention_head_dim': 4,
'resnet_time_scale_shift': 'scale_shift',
'class_embed_type': None,
}
__lowerCAmelCase: Optional[int] = UNetaDConditionModel(**_snake_case )
return model
@property
def UpperCAmelCase ( self : Any ) -> Tuple:
return {
"block_out_channels": [3_2, 6_4],
"down_block_types": ["DownEncoderBlock2D", "AttnDownEncoderBlock2D"],
"in_channels": 3,
"latent_channels": 4,
"layers_per_block": 1,
"norm_num_groups": 8,
"norm_type": "spatial",
"num_vq_embeddings": 1_2,
"out_channels": 3,
"up_block_types": [
"AttnUpDecoderBlock2D",
"UpDecoderBlock2D",
],
"vq_embed_dim": 4,
}
@property
def UpperCAmelCase ( self : str ) -> Dict:
torch.manual_seed(0 )
__lowerCAmelCase: List[str] = VQModel(**self.dummy_movq_kwargs )
return model
def UpperCAmelCase ( self : str ) -> List[Any]:
__lowerCAmelCase: str = self.dummy_text_encoder
__lowerCAmelCase: Optional[Any] = self.dummy_tokenizer
__lowerCAmelCase: List[Any] = self.dummy_unet
__lowerCAmelCase: Optional[int] = self.dummy_movq
__lowerCAmelCase: Optional[Any] = DDIMScheduler(
num_train_timesteps=1_0_0_0 , beta_schedule='linear' , beta_start=0.00085 , beta_end=0.012 , clip_sample=_snake_case , set_alpha_to_one=_snake_case , steps_offset=1 , prediction_type='epsilon' , thresholding=_snake_case , )
__lowerCAmelCase: Union[str, Any] = {
'text_encoder': text_encoder,
'tokenizer': tokenizer,
'unet': unet,
'scheduler': scheduler,
'movq': movq,
}
return components
def UpperCAmelCase ( self : str , UpperCAmelCase : Any , UpperCAmelCase : Tuple=0 ) -> str:
__lowerCAmelCase: List[str] = floats_tensor((1, self.cross_attention_dim) , rng=random.Random(_snake_case ) ).to(_snake_case )
__lowerCAmelCase: List[str] = floats_tensor((1, self.cross_attention_dim) , rng=random.Random(seed + 1 ) ).to(_snake_case )
# create init_image
__lowerCAmelCase: Any = floats_tensor((1, 3, 6_4, 6_4) , rng=random.Random(_snake_case ) ).to(_snake_case )
__lowerCAmelCase: int = image.cpu().permute(0 , 2 , 3 , 1 )[0]
__lowerCAmelCase: Dict = Image.fromarray(np.uinta(_snake_case ) ).convert('RGB' ).resize((2_5_6, 2_5_6) )
# create mask
__lowerCAmelCase: Optional[Any] = np.ones((6_4, 6_4) , dtype=np.floataa )
__lowerCAmelCase: Optional[Any] = 0
if str(_snake_case ).startswith('mps' ):
__lowerCAmelCase: Any = torch.manual_seed(_snake_case )
else:
__lowerCAmelCase: int = torch.Generator(device=_snake_case ).manual_seed(_snake_case )
__lowerCAmelCase: int = {
'prompt': 'horse',
'image': init_image,
'mask_image': mask,
'image_embeds': image_embeds,
'negative_image_embeds': negative_image_embeds,
'generator': generator,
'height': 6_4,
'width': 6_4,
'num_inference_steps': 2,
'guidance_scale': 4.0,
'output_type': 'np',
}
return inputs
def UpperCAmelCase ( self : List[str] ) -> Any:
__lowerCAmelCase: Any = 'cpu'
__lowerCAmelCase: str = self.get_dummy_components()
__lowerCAmelCase: Any = self.pipeline_class(**_snake_case )
__lowerCAmelCase: Any = pipe.to(_snake_case )
pipe.set_progress_bar_config(disable=_snake_case )
__lowerCAmelCase: Optional[int] = pipe(**self.get_dummy_inputs(_snake_case ) )
__lowerCAmelCase: int = output.images
__lowerCAmelCase: str = pipe(
**self.get_dummy_inputs(_snake_case ) , return_dict=_snake_case , )[0]
__lowerCAmelCase: Optional[Any] = image[0, -3:, -3:, -1]
__lowerCAmelCase: Union[str, Any] = image_from_tuple[0, -3:, -3:, -1]
print(F'''image.shape {image.shape}''' )
assert image.shape == (1, 6_4, 6_4, 3)
__lowerCAmelCase: Any = np.array(
[0.8326919, 0.73790467, 0.20918581, 0.9309612, 0.5511791, 0.43713328, 0.5513321, 0.49922934, 0.59497786] )
assert (
np.abs(image_slice.flatten() - expected_slice ).max() < 1E-2
), F''' expected_slice {expected_slice}, but got {image_slice.flatten()}'''
assert (
np.abs(image_from_tuple_slice.flatten() - expected_slice ).max() < 1E-2
), F''' expected_slice {expected_slice}, but got {image_from_tuple_slice.flatten()}'''
def UpperCAmelCase ( self : Dict ) -> Optional[int]:
super().test_inference_batch_single_identical(expected_max_diff=3E-3 )
@slow
@require_torch_gpu
class A_ ( unittest.TestCase ):
def UpperCAmelCase ( self : int ) -> Dict:
# clean up the VRAM after each test
super().tearDown()
gc.collect()
torch.cuda.empty_cache()
def UpperCAmelCase ( self : Any ) -> Tuple:
__lowerCAmelCase: Optional[Any] = load_numpy(
'https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main'
'/kandinsky/kandinsky_inpaint_cat_with_hat_fp16.npy' )
__lowerCAmelCase: Optional[Any] = load_image(
'https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main' '/kandinsky/cat.png' )
__lowerCAmelCase: Dict = np.ones((7_6_8, 7_6_8) , dtype=np.floataa )
__lowerCAmelCase: Union[str, Any] = 0
__lowerCAmelCase: Optional[Any] = 'a hat'
__lowerCAmelCase: Optional[int] = KandinskyPriorPipeline.from_pretrained(
'kandinsky-community/kandinsky-2-1-prior' , torch_dtype=torch.floataa )
pipe_prior.to(_snake_case )
__lowerCAmelCase: List[str] = KandinskyInpaintPipeline.from_pretrained(
'kandinsky-community/kandinsky-2-1-inpaint' , torch_dtype=torch.floataa )
__lowerCAmelCase: Any = pipeline.to(_snake_case )
pipeline.set_progress_bar_config(disable=_snake_case )
__lowerCAmelCase: Dict = torch.Generator(device='cpu' ).manual_seed(0 )
__lowerCAmelCase: List[Any] = pipe_prior(
_snake_case , generator=_snake_case , num_inference_steps=5 , negative_prompt='' , ).to_tuple()
__lowerCAmelCase: Dict = pipeline(
_snake_case , image=_snake_case , mask_image=_snake_case , image_embeds=_snake_case , negative_image_embeds=_snake_case , generator=_snake_case , num_inference_steps=1_0_0 , height=7_6_8 , width=7_6_8 , output_type='np' , )
__lowerCAmelCase: List[Any] = output.images[0]
assert image.shape == (7_6_8, 7_6_8, 3)
assert_mean_pixel_difference(_snake_case , _snake_case )
| 322 |
from collections import deque
from .hash_table import HashTable
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
def __init__( self : Optional[Any], *_snake_case : Optional[Any], **_snake_case : List[Any] ) ->Optional[int]:
super().__init__(*_snake_case, **_snake_case )
def lowercase_ ( self : Optional[Any], _snake_case : Tuple, _snake_case : Dict ) ->Dict:
snake_case__ : int = deque([] ) if self.values[key] is None else self.values[key]
self.values[key].appendleft(_snake_case )
snake_case__ : Dict = self.values[key]
def lowercase_ ( self : Any ) ->Optional[Any]:
return (
sum(self.charge_factor - len(_snake_case ) for slot in self.values )
/ self.size_table
* self.charge_factor
)
def lowercase_ ( self : Union[str, Any], _snake_case : str, _snake_case : Optional[int]=None ) ->Optional[Any]:
if not (
len(self.values[key] ) == self.charge_factor and self.values.count(_snake_case ) == 0
):
return key
return super()._collision_resolution(_snake_case, _snake_case )
| 277 | 0 |
from dataclasses import dataclass
from typing import Optional, Tuple, Union
import flax
import jax.numpy as jnp
from jax import random
from ..configuration_utils import ConfigMixin, register_to_config
from ..utils import BaseOutput
from .scheduling_utils_flax import FlaxSchedulerMixin
@flax.struct.dataclass
class __lowerCAmelCase :
_lowercase : List[Any] = None
_lowercase : List[str] = None
_lowercase : Tuple = None # sigma(t_i)
@classmethod
def _lowercase ( cls ) -> Optional[Any]:
'''simple docstring'''
return cls()
@dataclass
class __lowerCAmelCase ( lowerCAmelCase_):
_lowercase : Optional[int] = 42
_lowercase : Union[str, Any] = 42
_lowercase : List[str] = 42
class __lowerCAmelCase ( lowerCAmelCase_ , lowerCAmelCase_):
@property
def _lowercase ( self ) -> List[str]:
'''simple docstring'''
return True
@register_to_config
def __init__( self , lowerCAmelCase__ = 0.02 , lowerCAmelCase__ = 1_0_0 , lowerCAmelCase__ = 1.0_07 , lowerCAmelCase__ = 8_0 , lowerCAmelCase__ = 0.05 , lowerCAmelCase__ = 5_0 , ) -> List[Any]:
'''simple docstring'''
pass
def _lowercase ( self ) -> str:
'''simple docstring'''
return KarrasVeSchedulerState.create()
def _lowercase ( self , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ = () ) -> KarrasVeSchedulerState:
'''simple docstring'''
a__ : Tuple =jnp.arange(0 , _snake_case )[::-1].copy()
a__ : List[str] =[
(
self.config.sigma_max**2
* (self.config.sigma_min**2 / self.config.sigma_max**2) ** (i / (num_inference_steps - 1))
)
for i in timesteps
]
return state.replace(
num_inference_steps=_snake_case , schedule=jnp.array(_snake_case , dtype=jnp.floataa ) , timesteps=_snake_case , )
def _lowercase ( self , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , ) -> Tuple[jnp.ndarray, float]:
'''simple docstring'''
if self.config.s_min <= sigma <= self.config.s_max:
a__ : Optional[int] =min(self.config.s_churn / state.num_inference_steps , 2**0.5 - 1 )
else:
a__ : Any =0
# sample eps ~ N(0, S_noise^2 * I)
a__ : List[Any] =random.split(_snake_case , num=1 )
a__ : List[str] =self.config.s_noise * random.normal(key=_snake_case , shape=sample.shape )
a__ : Optional[Any] =sigma + gamma * sigma
a__ : Optional[Any] =sample + ((sigma_hat**2 - sigma**2) ** 0.5 * eps)
return sample_hat, sigma_hat
def _lowercase ( self , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ = True , ) -> Union[FlaxKarrasVeOutput, Tuple]:
'''simple docstring'''
a__ : int =sample_hat + sigma_hat * model_output
a__ : List[str] =(sample_hat - pred_original_sample) / sigma_hat
a__ : int =sample_hat + (sigma_prev - sigma_hat) * derivative
if not return_dict:
return (sample_prev, derivative, state)
return FlaxKarrasVeOutput(prev_sample=_snake_case , derivative=_snake_case , state=_snake_case )
def _lowercase ( self , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ = True , ) -> Union[FlaxKarrasVeOutput, Tuple]:
'''simple docstring'''
a__ : Optional[Any] =sample_prev + sigma_prev * model_output
a__ : str =(sample_prev - pred_original_sample) / sigma_prev
a__ : int =sample_hat + (sigma_prev - sigma_hat) * (0.5 * derivative + 0.5 * derivative_corr)
if not return_dict:
return (sample_prev, derivative, state)
return FlaxKarrasVeOutput(prev_sample=_snake_case , derivative=_snake_case , state=_snake_case )
def _lowercase ( self , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ ) -> Union[str, Any]:
'''simple docstring'''
raise NotImplementedError()
| 95 |
def lowercase_ (A : Union[str, Any] , A : List[str] , A : int , A : Optional[int] ):
global f # a global dp table for knapsack
if f[i][j] < 0:
if j < wt[i - 1]:
snake_case__ : Union[str, Any] = mf_knapsack(i - 1 , A , A , A )
else:
snake_case__ : Any = max(
mf_knapsack(i - 1 , A , A , A ) , mf_knapsack(i - 1 , A , A , j - wt[i - 1] ) + val[i - 1] , )
snake_case__ : Optional[int] = val
return f[i][j]
def lowercase_ (A : Optional[int] , A : Union[str, Any] , A : str , A : Dict ):
snake_case__ : int = [[0] * (w + 1) for _ in range(n + 1 )]
for i in range(1 , n + 1 ):
for w_ in range(1 , w + 1 ):
if wt[i - 1] <= w_:
snake_case__ : Union[str, Any] = max(val[i - 1] + dp[i - 1][w_ - wt[i - 1]] , dp[i - 1][w_] )
else:
snake_case__ : str = dp[i - 1][w_]
return dp[n][w_], dp
def lowercase_ (A : int , A : list , A : list ):
if not (isinstance(A , (list, tuple) ) and isinstance(A , (list, tuple) )):
raise ValueError(
'Both the weights and values vectors must be either lists or tuples' )
snake_case__ : Dict = len(A )
if num_items != len(A ):
snake_case__ : str = (
'The number of weights must be the same as the number of values.\n'
F'''But got {num_items} weights and {len(A )} values'''
)
raise ValueError(A )
for i in range(A ):
if not isinstance(wt[i] , A ):
snake_case__ : Optional[int] = (
'All weights must be integers but got weight of '
F'''type {type(wt[i] )} at index {i}'''
)
raise TypeError(A )
snake_case__ , snake_case__ : Optional[int] = knapsack(A , A , A , A )
snake_case__ : set = set()
_construct_solution(A , A , A , A , A )
return optimal_val, example_optional_set
def lowercase_ (A : list , A : list , A : int , A : int , A : set ):
# for the current item i at a maximum weight j to be part of an optimal subset,
# the optimal value at (i, j) must be greater than the optimal value at (i-1, j).
# where i - 1 means considering only the previous items at the given maximum weight
if i > 0 and j > 0:
if dp[i - 1][j] == dp[i][j]:
_construct_solution(A , A , i - 1 , A , A )
else:
optimal_set.add(A )
_construct_solution(A , A , i - 1 , j - wt[i - 1] , A )
if __name__ == "__main__":
a_ :Any = [3, 2, 4, 4]
a_ :List[Any] = [4, 3, 2, 3]
a_ :Union[str, Any] = 4
a_ :List[str] = 6
a_ :Union[str, Any] = [[0] * (w + 1)] + [[0] + [-1] * (w + 1) for _ in range(n + 1)]
a_ , a_ :List[Any] = knapsack(w, wt, val, n)
print(optimal_solution)
print(mf_knapsack(n, wt, val, w)) # switched the n and w
# testing the dynamic programming problem with example
# the optimal subset for the above example are items 3 and 4
a_ , a_ :Any = knapsack_with_example_solution(w, wt, val)
assert optimal_solution == 8
assert optimal_subset == {3, 4}
print("optimal_value = ", optimal_solution)
print("An optimal subset corresponding to the optimal value", optimal_subset)
| 277 | 0 |
import math
def a ( snake_case__: int ):
'''simple docstring'''
assert isinstance(snake_case__ , snake_case__ ) and (
number >= 0
), "'number' must been an int and positive"
if 1 < number < 4:
# 2 and 3 are primes
return True
elif number < 2 or not number % 2:
# Negatives, 0, 1 and all even numbers are not primes
return False
lowercase_ = range(3 , int(math.sqrt(snake_case__ ) + 1 ) , 2 )
return not any(not number % i for i in odd_numbers )
def a ( snake_case__: Any , snake_case__: Optional[Any]=1 , **snake_case__: Any ):
'''simple docstring'''
lowercase_ = factor * value
lowercase_ = value
while not is_prime(snake_case__ ):
value += 1 if not ("desc" in kwargs and kwargs["desc"] is True) else -1
if value == first_value_val:
return next_prime(value + 1 , **snake_case__ )
return value
| 30 |
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_flax_available, is_torch_available
a_ :int = {
"configuration_longt5": ["LONGT5_PRETRAINED_CONFIG_ARCHIVE_MAP", "LongT5Config", "LongT5OnnxConfig"],
}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :List[str] = [
"LONGT5_PRETRAINED_MODEL_ARCHIVE_LIST",
"LongT5EncoderModel",
"LongT5ForConditionalGeneration",
"LongT5Model",
"LongT5PreTrainedModel",
]
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :int = [
"FlaxLongT5ForConditionalGeneration",
"FlaxLongT5Model",
"FlaxLongT5PreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_longta import LONGT5_PRETRAINED_CONFIG_ARCHIVE_MAP, LongTaConfig, LongTaOnnxConfig
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_longta import (
LONGT5_PRETRAINED_MODEL_ARCHIVE_LIST,
LongTaEncoderModel,
LongTaForConditionalGeneration,
LongTaModel,
LongTaPreTrainedModel,
)
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_flax_longta import (
FlaxLongTaForConditionalGeneration,
FlaxLongTaModel,
FlaxLongTaPreTrainedModel,
)
else:
import sys
a_ :Optional[int] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)
| 277 | 0 |
"""simple docstring"""
def lowercase ( lowerCAmelCase__ : int , lowerCAmelCase__ : int ) -> str:
while b:
__a = b, a % b
return a
def lowercase ( lowerCAmelCase__ : int , lowerCAmelCase__ : int ) -> List[str]:
return a if b == 0 else euclidean_gcd_recursive(lowerCAmelCase__ , a % b )
def lowercase ( ) -> Optional[Any]:
print(f'''euclidean_gcd(3, 5) = {euclidean_gcd(3 , 5 )}''' )
print(f'''euclidean_gcd(5, 3) = {euclidean_gcd(5 , 3 )}''' )
print(f'''euclidean_gcd(1, 3) = {euclidean_gcd(1 , 3 )}''' )
print(f'''euclidean_gcd(3, 6) = {euclidean_gcd(3 , 6 )}''' )
print(f'''euclidean_gcd(6, 3) = {euclidean_gcd(6 , 3 )}''' )
print(f'''euclidean_gcd_recursive(3, 5) = {euclidean_gcd_recursive(3 , 5 )}''' )
print(f'''euclidean_gcd_recursive(5, 3) = {euclidean_gcd_recursive(5 , 3 )}''' )
print(f'''euclidean_gcd_recursive(1, 3) = {euclidean_gcd_recursive(1 , 3 )}''' )
print(f'''euclidean_gcd_recursive(3, 6) = {euclidean_gcd_recursive(3 , 6 )}''' )
print(f'''euclidean_gcd_recursive(6, 3) = {euclidean_gcd_recursive(6 , 3 )}''' )
if __name__ == "__main__":
main()
| 45 |
import argparse
import json
import os
import torch
from torch import nn
from transformers import NllbMoeConfig, NllbMoeModel
from transformers.modeling_utils import dtype_byte_size
from transformers.utils import WEIGHTS_INDEX_NAME, WEIGHTS_NAME
def lowercase_ (A : List[str] ):
snake_case__ : Tuple = [
'encoder.version',
'decoder.version',
'model.encoder.version',
'model.decoder.version',
'decoder.output_projection.weight',
'_float_tensor',
'encoder.embed_positions._float_tensor',
'decoder.embed_positions._float_tensor',
]
for k in ignore_keys:
state_dict.pop(A , A )
def lowercase_ (A : str ):
snake_case__ , snake_case__ : Union[str, Any] = emb.weight.shape
snake_case__ : str = nn.Linear(A , A , bias=A )
snake_case__ : str = emb.weight.data
return lin_layer
def lowercase_ (A : Optional[int] , A : Union[str, Any]=None ):
snake_case__ : Any = {}
for old_key in state_dict.keys():
snake_case__ : Tuple = old_key
if "moe_layer.experts." in key:
if expert_idx is not None:
snake_case__ : int = key.replace('moe_layer.experts.0' , F'''ffn.experts.expert_{expert_idx}''' )
else:
snake_case__ : Any = key.replace('moe_layer.experts.' , 'ffn.experts.expert_' )
if "gate" in key:
snake_case__ : Dict = key.replace('.moe_layer.gate.wg' , '.ffn.router.classifier' )
if "fc2" and "experts" not in key:
snake_case__ : str = key.replace('.fc2.' , '.ffn.fc2.' )
if "fc1" and "experts" not in key:
snake_case__ : str = key.replace('.fc1.' , '.ffn.fc1.' )
if ".encoder_attn." in key:
snake_case__ : Tuple = key.replace('.encoder_attn.' , '.cross_attention.' )
if "encoder_attn_layer_norm" in key:
snake_case__ : Tuple = key.replace('encoder_attn_layer_norm' , 'cross_attention_layer_norm' )
if "final_layer_norm" in key:
snake_case__ : Optional[int] = key.replace('final_layer_norm' , 'ff_layer_norm' )
snake_case__ : Dict = state_dict[old_key]
return new_dict
def lowercase_ (A : List[Any] , A : Tuple , A : List[Any] , A : List[str] , A : str = WEIGHTS_NAME ):
snake_case__ : Dict = []
snake_case__ : str = 0
os.makedirs(A , exist_ok=A )
for expert in range(A ):
snake_case__ : Tuple = switch_checkpoint_path + F'''-rank-{expert}.pt'''
if os.path.isfile(A ):
snake_case__ : Optional[Any] = torch.load(A )['model']
remove_ignore_keys_(A )
snake_case__ : Optional[Any] = rename_fairseq_keys(A , A )
snake_case__ : Dict = os.path.join(
A , weights_name.replace('.bin' , F'''-{len(A )+1:05d}-of-???.bin''' ) )
torch.save(A , A )
sharded_state_dicts.append(expert_state.keys() )
total_size += sum([value.numel() for key, value in expert_state.items()] ) * dtype_byte_size(
expert_state[list(A )[0]].dtype )
# Add the last block
snake_case__ : Tuple = os.path.join(A , weights_name.replace('.bin' , F'''-{len(A )+1:05d}-of-???.bin''' ) )
snake_case__ : Union[str, Any] = torch.load(switch_checkpoint_path + '-shared.pt' )['model']
remove_ignore_keys_(A )
snake_case__ : str = rename_fairseq_keys(A , A )
snake_case__ : Any = shared_weights['decoder.embed_tokens.weight']
sharded_state_dicts.append(shared_weights.keys() )
# If we only have the shared weights (dummy model/experts saved on the same file)
if len(A ) == 1:
snake_case__ : Any = os.path.join(A , A )
torch.save(A , A )
return {weights_name: sharded_state_dicts[0]}, None
else:
torch.save(A , A )
# Otherwise, let's build the index
snake_case__ : Tuple = {}
for idx, shard in enumerate(A ):
snake_case__ : Optional[int] = weights_name.replace('.bin' , F'''-{idx+1:05d}-of-{len(A ):05d}.bin''' )
snake_case__ : List[Any] = os.path.join(A , weights_name.replace('.bin' , F'''-{idx+1:05d}-of-???.bin''' ) )
os.rename(A , os.path.join(A , A ) )
for key in shard:
snake_case__ : Any = shard_file
# Add the metadata
snake_case__ : int = {'total_size': total_size}
snake_case__ : Dict = {'metadata': metadata, 'weight_map': weight_map}
with open(os.path.join(A , A ) , 'w' , encoding='utf-8' ) as f:
snake_case__ : Any = json.dumps(A , indent=2 , sort_keys=A ) + '\n'
f.write(A )
return metadata, index
if __name__ == "__main__":
a_ :int = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--nllb_moe_checkpoint_path",
default="/home/arthur_huggingface_co/fairseq/weights/checkpoints/model_moe_54b/checkpoint_2_300000",
type=str,
required=False,
help="Path to a directory containing a folder per layer. Follows the original Google format.",
)
parser.add_argument("--dtype", default="float32", type=str, required=False, help="dtype of the saved model")
parser.add_argument(
"--pytorch_dump_folder_path",
default="/home/arthur_huggingface_co/fairseq/weights/checkpoints/hf-converted-moe-54b",
type=str,
required=False,
help="Path to the output pytorch model.",
)
a_ :Optional[Any] = parser.parse_args()
a_ , a_ :Optional[Any] = shard_on_the_fly(
args.nllb_moe_checkpoint_path,
args.pytorch_dump_folder_path,
128,
args.dtype,
)
a_ :List[str] = NllbMoeConfig.from_pretrained(
"facebook/nllb-200-3.3B", encoder_sparse_step=4, decoder_sparse_step=4, num_experts=128
)
config.save_pretrained(args.pytorch_dump_folder_path)
a_ :int = NllbMoeModel.from_pretrained(args.pytorch_dump_folder_path)
print("Done")
model.save_pretrained(args.pytorch_dump_folder_path)
| 277 | 0 |
'''simple docstring'''
import argparse
import json
import os
import fairseq
import torch
from fairseq.data import Dictionary
from transformers import (
UniSpeechConfig,
UniSpeechForCTC,
UniSpeechForPreTraining,
WavaVecaFeatureExtractor,
WavaVecaPhonemeCTCTokenizer,
WavaVecaProcessor,
logging,
)
logging.set_verbosity_info()
_lowerCAmelCase = logging.get_logger(__name__)
_lowerCAmelCase = {
"post_extract_proj": "feature_projection.projection",
"encoder.pos_conv.0": "encoder.pos_conv_embed.conv",
"self_attn.k_proj": "encoder.layers.*.attention.k_proj",
"self_attn.v_proj": "encoder.layers.*.attention.v_proj",
"self_attn.q_proj": "encoder.layers.*.attention.q_proj",
"self_attn.out_proj": "encoder.layers.*.attention.out_proj",
"self_attn_layer_norm": "encoder.layers.*.layer_norm",
"fc1": "encoder.layers.*.feed_forward.intermediate_dense",
"fc2": "encoder.layers.*.feed_forward.output_dense",
"final_layer_norm": "encoder.layers.*.final_layer_norm",
"encoder.layer_norm": "encoder.layer_norm",
"w2v_model.layer_norm": "feature_projection.layer_norm",
"quantizer.weight_proj": "quantizer.weight_proj",
"quantizer.vars": "quantizer.codevectors",
"project_q": "project_q",
"final_proj": "project_hid",
"w2v_encoder.proj": "ctc_proj",
"mask_emb": "masked_spec_embed",
}
_lowerCAmelCase = [
"ctc_proj",
"quantizer.weight_proj",
"quantizer.codevectors",
"project_q",
"project_hid",
]
def __lowerCAmelCase ( snake_case__ , snake_case__ , snake_case__ , snake_case__ , snake_case__ , snake_case__ ):
for attribute in key.split("." ):
if is_finetuned:
if attribute in ["quantizer", "project_q", "project_hid"]:
# those layers are only relevant for pretraining and should be dropped
return
if attribute == "ctc_proj":
# we should rename `ctc_proj` to `lm_head` for fine-tuned phoneme models
__UpperCamelCase : Optional[Any] = 'lm_head'
__UpperCamelCase : int = getattr(snake_case__ , snake_case__ )
if weight_type is not None:
__UpperCamelCase : Tuple = getattr(snake_case__ , snake_case__ ).shape
else:
__UpperCamelCase : List[Any] = hf_pointer.shape
assert hf_shape == value.shape, (
F"Shape of hf {key + '.' + weight_type if weight_type is not None else ''} is {hf_shape}, but should be"
F" {value.shape} for {full_name}"
)
if weight_type == "weight":
__UpperCamelCase : Any = value
elif weight_type == "weight_g":
__UpperCamelCase : Union[str, Any] = value
elif weight_type == "weight_v":
__UpperCamelCase : Optional[Any] = value
elif weight_type == "bias":
__UpperCamelCase : Dict = value
else:
__UpperCamelCase : Optional[Any] = value
logger.info(F"{key + '.' + weight_type if weight_type is not None else ''} was initialized from {full_name}." )
def __lowerCAmelCase ( snake_case__ , snake_case__ , snake_case__ ):
__UpperCamelCase : Optional[int] = []
__UpperCamelCase : Optional[Any] = fairseq_model.state_dict()
__UpperCamelCase : Tuple = hf_model.unispeech.feature_extractor
for name, value in fairseq_dict.items():
__UpperCamelCase : Dict = False
if "conv_layers" in name:
load_conv_layer(
snake_case__ , snake_case__ , snake_case__ , snake_case__ , hf_model.config.feat_extract_norm == "group" , )
__UpperCamelCase : Tuple = True
else:
for key, mapped_key in MAPPING.items():
__UpperCamelCase : List[Any] = 'unispeech.' + mapped_key if mapped_key not in TOP_LEVEL_KEYS else mapped_key
if key in name or key.split("w2v_model." )[-1] == name.split("." )[0]:
__UpperCamelCase : Optional[int] = True
if "*" in mapped_key:
__UpperCamelCase : int = name.split(snake_case__ )[0].split("." )[-2]
__UpperCamelCase : List[str] = mapped_key.replace("*" , snake_case__ )
if "weight_g" in name:
__UpperCamelCase : Any = 'weight_g'
elif "weight_v" in name:
__UpperCamelCase : List[Any] = 'weight_v'
elif "bias" in name:
__UpperCamelCase : int = 'bias'
elif "weight" in name:
# TODO: don't match quantizer.weight_proj
__UpperCamelCase : Any = 'weight'
else:
__UpperCamelCase : Union[str, Any] = None
set_recursively(snake_case__ , snake_case__ , snake_case__ , snake_case__ , snake_case__ , snake_case__ )
continue
if not is_used:
unused_weights.append(snake_case__ )
logger.warning(F"Unused weights: {unused_weights}" )
def __lowerCAmelCase ( snake_case__ , snake_case__ , snake_case__ , snake_case__ , snake_case__ ):
__UpperCamelCase : Optional[int] = full_name.split("conv_layers." )[-1]
__UpperCamelCase : str = name.split("." )
__UpperCamelCase : Union[str, Any] = int(items[0] )
__UpperCamelCase : Optional[Any] = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
F"{full_name} has size {value.shape}, but"
F" {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found."
)
__UpperCamelCase : int = value
logger.info(F"Feat extract conv layer {layer_id} was initialized from {full_name}." )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
F"{full_name} has size {value.shape}, but"
F" {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found."
)
__UpperCamelCase : List[str] = value
logger.info(F"Feat extract conv layer {layer_id} was initialized from {full_name}." )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
F"{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was"
" found."
)
__UpperCamelCase : List[Any] = value
logger.info(F"Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}." )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
F"{full_name} has size {value.shape}, but"
F" {feature_extractor[layer_id].layer_norm.weight.data.shape} was found."
)
__UpperCamelCase : str = value
logger.info(F"Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}." )
else:
unused_weights.append(snake_case__ )
@torch.no_grad()
def __lowerCAmelCase ( snake_case__ , snake_case__ , snake_case__=None , snake_case__=None , snake_case__=True ):
if config_path is not None:
__UpperCamelCase : Tuple = UniSpeechConfig.from_pretrained(snake_case__ )
else:
__UpperCamelCase : Tuple = UniSpeechConfig()
if is_finetuned:
if dict_path:
__UpperCamelCase : Tuple = Dictionary.load_from_json(snake_case__ )
# important change bos & pad token id since CTC symbol is <pad> and
# not <s> as in fairseq
__UpperCamelCase : Any = target_dict.pad_index
__UpperCamelCase : List[str] = target_dict.bos_index
__UpperCamelCase : List[Any] = target_dict.eos_index
__UpperCamelCase : Optional[Any] = len(target_dict.symbols )
__UpperCamelCase : List[Any] = os.path.join(snake_case__ , "vocab.json" )
if not os.path.isdir(snake_case__ ):
logger.error("--pytorch_dump_folder_path ({}) should be a directory".format(snake_case__ ) )
return
os.makedirs(snake_case__ , exist_ok=snake_case__ )
__UpperCamelCase : List[Any] = target_dict.indices
# fairseq has the <pad> and <s> switched
__UpperCamelCase : Optional[Any] = 42
__UpperCamelCase : Tuple = 43
with open(snake_case__ , "w" , encoding="utf-8" ) as vocab_handle:
json.dump(snake_case__ , snake_case__ )
__UpperCamelCase : Tuple = WavaVecaPhonemeCTCTokenizer(
snake_case__ , unk_token=target_dict.unk_word , pad_token=target_dict.pad_word , bos_token=target_dict.bos_word , eos_token=target_dict.eos_word , word_delimiter_token="|" , do_lower_case=snake_case__ , )
__UpperCamelCase : int = True if config.feat_extract_norm == 'layer' else False
__UpperCamelCase : Union[str, Any] = WavaVecaFeatureExtractor(
feature_size=1 , sampling_rate=16_000 , padding_value=0 , do_normalize=snake_case__ , return_attention_mask=snake_case__ , )
__UpperCamelCase : Union[str, Any] = WavaVecaProcessor(feature_extractor=snake_case__ , tokenizer=snake_case__ )
processor.save_pretrained(snake_case__ )
__UpperCamelCase : int = UniSpeechForCTC(snake_case__ )
else:
__UpperCamelCase : Optional[int] = UniSpeechForPreTraining(snake_case__ )
if is_finetuned:
__UpperCamelCase : str = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] , arg_overrides={"data": "/".join(dict_path.split("/" )[:-1] ), "w2v_path": checkpoint_path} )
else:
__UpperCamelCase : Tuple = fairseq.checkpoint_utils.load_model_ensemble_and_task([checkpoint_path] )
__UpperCamelCase : Tuple = model[0].eval()
recursively_load_weights(snake_case__ , snake_case__ , snake_case__ )
hf_unispeech.save_pretrained(snake_case__ )
if __name__ == "__main__":
_lowerCAmelCase = argparse.ArgumentParser()
parser.add_argument('''--pytorch_dump_folder_path''', default=None, type=str, help='''Path to the output PyTorch model.''')
parser.add_argument('''--checkpoint_path''', default=None, type=str, help='''Path to fairseq checkpoint''')
parser.add_argument('''--dict_path''', default=None, type=str, help='''Path to dict of fine-tuned model''')
parser.add_argument('''--config_path''', default=None, type=str, help='''Path to hf config.json of model to convert''')
parser.add_argument(
'''--not_finetuned''', action='''store_true''', help='''Whether the model to convert is a fine-tuned model or not'''
)
_lowerCAmelCase = parser.parse_args()
convert_unispeech_checkpoint(
args.checkpoint_path, args.pytorch_dump_folder_path, args.config_path, args.dict_path, not args.not_finetuned
)
| 298 |
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_sentencepiece_available,
is_tokenizers_available,
is_torch_available,
)
a_ :Optional[Any] = {"configuration_reformer": ["REFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP", "ReformerConfig"]}
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :str = ["ReformerTokenizer"]
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :int = ["ReformerTokenizerFast"]
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :List[str] = [
"REFORMER_PRETRAINED_MODEL_ARCHIVE_LIST",
"ReformerAttention",
"ReformerForMaskedLM",
"ReformerForQuestionAnswering",
"ReformerForSequenceClassification",
"ReformerLayer",
"ReformerModel",
"ReformerModelWithLMHead",
"ReformerPreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_reformer import REFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, ReformerConfig
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_reformer import ReformerTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_reformer_fast import ReformerTokenizerFast
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_reformer import (
REFORMER_PRETRAINED_MODEL_ARCHIVE_LIST,
ReformerAttention,
ReformerForMaskedLM,
ReformerForQuestionAnswering,
ReformerForSequenceClassification,
ReformerLayer,
ReformerModel,
ReformerModelWithLMHead,
ReformerPreTrainedModel,
)
else:
import sys
a_ :Optional[Any] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)
| 277 | 0 |
import inspect
from typing import Callable, List, Optional, Union
import torch
from transformers import (
CLIPImageProcessor,
CLIPTextModel,
CLIPTokenizer,
WhisperForConditionalGeneration,
WhisperProcessor,
)
from diffusers import (
AutoencoderKL,
DDIMScheduler,
DiffusionPipeline,
LMSDiscreteScheduler,
PNDMScheduler,
UNetaDConditionModel,
)
from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion import StableDiffusionPipelineOutput
from diffusers.pipelines.stable_diffusion.safety_checker import StableDiffusionSafetyChecker
from diffusers.utils import logging
_snake_case = logging.get_logger(__name__) # pylint: disable=invalid-name
class lowercase ( lowerCAmelCase_ ):
def __init__( self , _a , _a , _a , _a , _a , _a , _a , _a , _a , ) -> Optional[int]:
super().__init__()
if safety_checker is None:
logger.warning(
F'''You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure'''
""" that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"""
""" results in services or applications open to the public. Both the diffusers team and Hugging Face"""
""" strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"""
""" it only for use-cases that involve analyzing network behavior or auditing its results. For more"""
""" information, please have a look at https://github.com/huggingface/diffusers/pull/254 .""" )
self.register_modules(
speech_model=_snake_case , speech_processor=_snake_case , vae=_snake_case , text_encoder=_snake_case , tokenizer=_snake_case , unet=_snake_case , scheduler=_snake_case , feature_extractor=_snake_case , )
def a__ ( self , _a = "auto" ) -> Union[str, Any]:
if slice_size == "auto":
_A : Any = self.unet.config.attention_head_dim // 2
self.unet.set_attention_slice(_snake_case )
def a__ ( self ) -> Any:
self.enable_attention_slicing(_snake_case )
@torch.no_grad()
def __call__( self , _a , _a=1_6000 , _a = 512 , _a = 512 , _a = 50 , _a = 7.5 , _a = None , _a = 1 , _a = 0.0 , _a = None , _a = None , _a = "pil" , _a = True , _a = None , _a = 1 , **_a , ) -> int:
_A : Optional[int] = self.speech_processor.feature_extractor(
_snake_case , return_tensors="""pt""" , sampling_rate=_snake_case ).input_features.to(self.device )
_A : Dict = self.speech_model.generate(_snake_case , max_length=48_0000 )
_A : Optional[int] = self.speech_processor.tokenizer.batch_decode(_snake_case , skip_special_tokens=_snake_case , normalize=_snake_case )[
0
]
if isinstance(_snake_case , _snake_case ):
_A : Union[str, Any] = 1
elif isinstance(_snake_case , _snake_case ):
_A : int = len(_snake_case )
else:
raise ValueError(F'''`prompt` has to be of type `str` or `list` but is {type(_snake_case )}''' )
if height % 8 != 0 or width % 8 != 0:
raise ValueError(F'''`height` and `width` have to be divisible by 8 but are {height} and {width}.''' )
if (callback_steps is None) or (
callback_steps is not None and (not isinstance(_snake_case , _snake_case ) or callback_steps <= 0)
):
raise ValueError(
F'''`callback_steps` has to be a positive integer but is {callback_steps} of type'''
F''' {type(_snake_case )}.''' )
# get prompt text embeddings
_A : Dict = self.tokenizer(
_snake_case , padding="""max_length""" , max_length=self.tokenizer.model_max_length , return_tensors="""pt""" , )
_A : Dict = text_inputs.input_ids
if text_input_ids.shape[-1] > self.tokenizer.model_max_length:
_A : List[Any] = self.tokenizer.batch_decode(text_input_ids[:, self.tokenizer.model_max_length :] )
logger.warning(
"""The following part of your input was truncated because CLIP can only handle sequences up to"""
F''' {self.tokenizer.model_max_length} tokens: {removed_text}''' )
_A : Union[str, Any] = text_input_ids[:, : self.tokenizer.model_max_length]
_A : Optional[Any] = self.text_encoder(text_input_ids.to(self.device ) )[0]
# duplicate text embeddings for each generation per prompt, using mps friendly method
_A : Optional[int] = text_embeddings.shape
_A : Optional[int] = text_embeddings.repeat(1 , _snake_case , 1 )
_A : str = text_embeddings.view(bs_embed * num_images_per_prompt , _snake_case , -1 )
# here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
# of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
# corresponds to doing no classifier free guidance.
_A : Optional[Any] = guidance_scale > 1.0
# get unconditional embeddings for classifier free guidance
if do_classifier_free_guidance:
_A : List[str]
if negative_prompt is None:
_A : Optional[Any] = [''] * batch_size
elif type(_snake_case ) is not type(_snake_case ):
raise TypeError(
F'''`negative_prompt` should be the same type to `prompt`, but got {type(_snake_case )} !='''
F''' {type(_snake_case )}.''' )
elif isinstance(_snake_case , _snake_case ):
_A : int = [negative_prompt]
elif batch_size != len(_snake_case ):
raise ValueError(
F'''`negative_prompt`: {negative_prompt} has batch size {len(_snake_case )}, but `prompt`:'''
F''' {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches'''
""" the batch size of `prompt`.""" )
else:
_A : Tuple = negative_prompt
_A : Dict = text_input_ids.shape[-1]
_A : int = self.tokenizer(
_snake_case , padding="""max_length""" , max_length=_snake_case , truncation=_snake_case , return_tensors="""pt""" , )
_A : Optional[int] = self.text_encoder(uncond_input.input_ids.to(self.device ) )[0]
# duplicate unconditional embeddings for each generation per prompt, using mps friendly method
_A : List[Any] = uncond_embeddings.shape[1]
_A : str = uncond_embeddings.repeat(1 , _snake_case , 1 )
_A : Union[str, Any] = uncond_embeddings.view(batch_size * num_images_per_prompt , _snake_case , -1 )
# For classifier free guidance, we need to do two forward passes.
# Here we concatenate the unconditional and text embeddings into a single batch
# to avoid doing two forward passes
_A : Tuple = torch.cat([uncond_embeddings, text_embeddings] )
# get the initial random noise unless the user supplied it
# Unlike in other pipelines, latents need to be generated in the target device
# for 1-to-1 results reproducibility with the CompVis implementation.
# However this currently doesn't work in `mps`.
_A : str = (batch_size * num_images_per_prompt, self.unet.config.in_channels, height // 8, width // 8)
_A : str = text_embeddings.dtype
if latents is None:
if self.device.type == "mps":
# randn does not exist on mps
_A : Dict = torch.randn(_snake_case , generator=_snake_case , device="""cpu""" , dtype=_snake_case ).to(
self.device )
else:
_A : str = torch.randn(_snake_case , generator=_snake_case , device=self.device , dtype=_snake_case )
else:
if latents.shape != latents_shape:
raise ValueError(F'''Unexpected latents shape, got {latents.shape}, expected {latents_shape}''' )
_A : int = latents.to(self.device )
# set timesteps
self.scheduler.set_timesteps(_snake_case )
# Some schedulers like PNDM have timesteps as arrays
# It's more optimized to move all timesteps to correct device beforehand
_A : Any = self.scheduler.timesteps.to(self.device )
# scale the initial noise by the standard deviation required by the scheduler
_A : Any = latents * self.scheduler.init_noise_sigma
# prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
# eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
# eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
# and should be between [0, 1]
_A : List[str] = 'eta' in set(inspect.signature(self.scheduler.step ).parameters.keys() )
_A : Any = {}
if accepts_eta:
_A : Optional[int] = eta
for i, t in enumerate(self.progress_bar(_snake_case ) ):
# expand the latents if we are doing classifier free guidance
_A : List[Any] = torch.cat([latents] * 2 ) if do_classifier_free_guidance else latents
_A : Tuple = self.scheduler.scale_model_input(_snake_case , _snake_case )
# predict the noise residual
_A : str = self.unet(_snake_case , _snake_case , encoder_hidden_states=_snake_case ).sample
# perform guidance
if do_classifier_free_guidance:
_A : List[str] = noise_pred.chunk(2 )
_A : Optional[int] = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
# compute the previous noisy sample x_t -> x_t-1
_A : Union[str, Any] = self.scheduler.step(_snake_case , _snake_case , _snake_case , **_snake_case ).prev_sample
# call the callback, if provided
if callback is not None and i % callback_steps == 0:
callback(_snake_case , _snake_case , _snake_case )
_A : Optional[int] = 1 / 0.18215 * latents
_A : Tuple = self.vae.decode(_snake_case ).sample
_A : Tuple = (image / 2 + 0.5).clamp(0 , 1 )
# we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
_A : int = image.cpu().permute(0 , 2 , 3 , 1 ).float().numpy()
if output_type == "pil":
_A : str = self.numpy_to_pil(_snake_case )
if not return_dict:
return image
return StableDiffusionPipelineOutput(images=_snake_case , nsfw_content_detected=_snake_case )
| 26 |
import itertools
import random
import unittest
import numpy as np
from transformers import BatchFeature, SpeechTaFeatureExtractor
from transformers.testing_utils import require_torch
from transformers.utils.import_utils import is_torch_available
from ...test_sequence_feature_extraction_common import SequenceFeatureExtractionTestMixin
if is_torch_available():
import torch
a_ :Any = random.Random()
def lowercase_ (A : int , A : Union[str, Any]=1.0 , A : List[str]=None , A : Any=None ):
if rng is None:
snake_case__ : List[str] = global_rng
snake_case__ : int = []
for batch_idx in range(shape[0] ):
values.append([] )
for _ in range(shape[1] ):
values[-1].append(rng.random() * scale )
return values
@require_torch
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
def __init__( self : Optional[Any], _snake_case : List[str], _snake_case : Tuple=7, _snake_case : Union[str, Any]=4_0_0, _snake_case : Any=2_0_0_0, _snake_case : Dict=1, _snake_case : Optional[Any]=0.0, _snake_case : List[Any]=1_6_0_0_0, _snake_case : List[Any]=True, _snake_case : List[Any]=8_0, _snake_case : Dict=1_6, _snake_case : str=6_4, _snake_case : Tuple="hann_window", _snake_case : Union[str, Any]=8_0, _snake_case : Optional[Any]=7_6_0_0, _snake_case : str=1e-10, _snake_case : Any=True, ) ->Union[str, Any]:
snake_case__ : Optional[int] = parent
snake_case__ : Optional[Any] = batch_size
snake_case__ : List[Any] = min_seq_length
snake_case__ : List[Any] = max_seq_length
snake_case__ : Any = (self.max_seq_length - self.min_seq_length) // (self.batch_size - 1)
snake_case__ : Tuple = feature_size
snake_case__ : List[Any] = padding_value
snake_case__ : Any = sampling_rate
snake_case__ : Dict = do_normalize
snake_case__ : Union[str, Any] = num_mel_bins
snake_case__ : Any = hop_length
snake_case__ : Any = win_length
snake_case__ : Any = win_function
snake_case__ : Optional[int] = fmin
snake_case__ : int = fmax
snake_case__ : Union[str, Any] = mel_floor
snake_case__ : Union[str, Any] = return_attention_mask
def lowercase_ ( self : Optional[int] ) ->List[str]:
return {
"feature_size": self.feature_size,
"padding_value": self.padding_value,
"sampling_rate": self.sampling_rate,
"do_normalize": self.do_normalize,
"num_mel_bins": self.num_mel_bins,
"hop_length": self.hop_length,
"win_length": self.win_length,
"win_function": self.win_function,
"fmin": self.fmin,
"fmax": self.fmax,
"mel_floor": self.mel_floor,
"return_attention_mask": self.return_attention_mask,
}
def lowercase_ ( self : Any, _snake_case : Optional[Any]=False, _snake_case : List[str]=False ) ->Union[str, Any]:
def _flatten(_snake_case : List[str] ):
return list(itertools.chain(*_snake_case ) )
if equal_length:
snake_case__ : Any = floats_list((self.batch_size, self.max_seq_length) )
else:
# make sure that inputs increase in size
snake_case__ : int = [
_flatten(floats_list((x, self.feature_size) ) )
for x in range(self.min_seq_length, self.max_seq_length, self.seq_length_diff )
]
if numpify:
snake_case__ : Any = [np.asarray(_snake_case ) for x in speech_inputs]
return speech_inputs
def lowercase_ ( self : Union[str, Any], _snake_case : str=False, _snake_case : Dict=False ) ->List[str]:
if equal_length:
snake_case__ : Optional[Any] = [floats_list((self.max_seq_length, self.num_mel_bins) ) for _ in range(self.batch_size )]
else:
# make sure that inputs increase in size
snake_case__ : List[str] = [
floats_list((x, self.num_mel_bins) )
for x in range(self.min_seq_length, self.max_seq_length, self.seq_length_diff )
]
if numpify:
snake_case__ : int = [np.asarray(_snake_case ) for x in speech_inputs]
return speech_inputs
@require_torch
class snake_case__ ( lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = SpeechTaFeatureExtractor
def lowercase_ ( self : int ) ->Union[str, Any]:
snake_case__ : List[str] = SpeechTaFeatureExtractionTester(self )
def lowercase_ ( self : Any, _snake_case : Dict ) ->Any:
self.assertTrue(np.all(np.mean(_snake_case, axis=0 ) < 1e-3 ) )
self.assertTrue(np.all(np.abs(np.var(_snake_case, axis=0 ) - 1 ) < 1e-3 ) )
def lowercase_ ( self : List[Any] ) ->Union[str, Any]:
# Tests that all call wrap to encode_plus and batch_encode_plus
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
# create three inputs of length 800, 1000, and 1200
snake_case__ : int = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Tuple = [np.asarray(_snake_case ) for speech_input in speech_inputs]
# Test not batched input
snake_case__ : str = feat_extract(speech_inputs[0], return_tensors='np' ).input_values
snake_case__ : List[str] = feat_extract(np_speech_inputs[0], return_tensors='np' ).input_values
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test batched
snake_case__ : Any = feat_extract(_snake_case, return_tensors='np' ).input_values
snake_case__ : Union[str, Any] = feat_extract(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
def lowercase_ ( self : int ) ->Optional[int]:
snake_case__ : List[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Tuple = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : int = ['longest', 'max_length', 'do_not_pad']
snake_case__ : List[str] = [None, 1_6_0_0, None]
for max_length, padding in zip(_snake_case, _snake_case ):
snake_case__ : Optional[int] = feat_extract(_snake_case, padding=_snake_case, max_length=_snake_case, return_tensors='np' )
snake_case__ : Optional[int] = processed.input_values
self._check_zero_mean_unit_variance(input_values[0][:8_0_0] )
self.assertTrue(input_values[0][8_0_0:].sum() < 1e-6 )
self._check_zero_mean_unit_variance(input_values[1][:1_0_0_0] )
self.assertTrue(input_values[0][1_0_0_0:].sum() < 1e-6 )
self._check_zero_mean_unit_variance(input_values[2][:1_2_0_0] )
def lowercase_ ( self : Union[str, Any] ) ->Optional[Any]:
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Tuple = range(8_0_0, 1_4_0_0, 2_0_0 )
snake_case__ : Optional[Any] = [floats_list((1, x) )[0] for x in lengths]
snake_case__ : Union[str, Any] = ['longest', 'max_length', 'do_not_pad']
snake_case__ : str = [None, 1_6_0_0, None]
for max_length, padding in zip(_snake_case, _snake_case ):
snake_case__ : List[str] = feat_extract(_snake_case, max_length=_snake_case, padding=_snake_case )
snake_case__ : Tuple = processed.input_values
self._check_zero_mean_unit_variance(input_values[0][:8_0_0] )
self._check_zero_mean_unit_variance(input_values[1][:1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2][:1_2_0_0] )
def lowercase_ ( self : List[Any] ) ->Optional[Any]:
snake_case__ : Any = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : str = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Optional[Any] = feat_extract(
_snake_case, truncation=_snake_case, max_length=1_0_0_0, padding='max_length', return_tensors='np' )
snake_case__ : int = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1] )
self._check_zero_mean_unit_variance(input_values[2] )
def lowercase_ ( self : int ) ->Union[str, Any]:
snake_case__ : str = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Dict = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : str = feat_extract(
_snake_case, truncation=_snake_case, max_length=1_0_0_0, padding='longest', return_tensors='np' )
snake_case__ : Dict = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1, :1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2] )
# make sure that if max_length < longest -> then pad to max_length
self.assertTrue(input_values.shape == (3, 1_0_0_0) )
snake_case__ : Tuple = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : List[str] = feat_extract(
_snake_case, truncation=_snake_case, max_length=2_0_0_0, padding='longest', return_tensors='np' )
snake_case__ : Optional[Any] = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1, :1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2] )
# make sure that if max_length > longest -> then pad to longest
self.assertTrue(input_values.shape == (3, 1_2_0_0) )
def lowercase_ ( self : List[str] ) ->Dict:
snake_case__ : Dict = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : List[Any] = np.random.rand(1_0_0 ).astype(np.floataa )
snake_case__ : int = np_speech_inputs.tolist()
for inputs in [py_speech_inputs, np_speech_inputs]:
snake_case__ : int = feature_extractor.pad([{'input_values': inputs}], return_tensors='np' )
self.assertTrue(np_processed.input_values.dtype == np.floataa )
snake_case__ : Optional[int] = feature_extractor.pad([{'input_values': inputs}], return_tensors='pt' )
self.assertTrue(pt_processed.input_values.dtype == torch.floataa )
def lowercase_ ( self : Optional[int] ) ->Optional[Any]:
# Tests that all call wrap to encode_plus and batch_encode_plus
snake_case__ : str = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
# create three inputs of length 800, 1000, and 1200
snake_case__ : List[Any] = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Dict = [np.asarray(_snake_case ) for speech_input in speech_inputs]
# Test feature size
snake_case__ : Optional[int] = feature_extractor(audio_target=_snake_case, padding=_snake_case, return_tensors='np' ).input_values
self.assertTrue(input_values.ndim == 3 )
self.assertTrue(input_values.shape[-1] == feature_extractor.num_mel_bins )
# Test not batched input
snake_case__ : Dict = feature_extractor(speech_inputs[0], return_tensors='np' ).input_values
snake_case__ : Any = feature_extractor(np_speech_inputs[0], return_tensors='np' ).input_values
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test batched
snake_case__ : Dict = feature_extractor(_snake_case, return_tensors='np' ).input_values
snake_case__ : Dict = feature_extractor(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test 2-D numpy arrays are batched.
snake_case__ : Optional[Any] = [floats_list((1, x) )[0] for x in (8_0_0, 8_0_0, 8_0_0)]
snake_case__ : int = np.asarray(_snake_case )
snake_case__ : Union[str, Any] = feature_extractor(_snake_case, return_tensors='np' ).input_values
snake_case__ : Union[str, Any] = feature_extractor(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
def lowercase_ ( self : Union[str, Any] ) ->str:
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : List[str] = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : Tuple = BatchFeature({input_name: speech_inputs} )
self.assertTrue(all(len(_snake_case ) == len(_snake_case ) for x, y in zip(_snake_case, processed_features[input_name] ) ) )
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target(equal_length=_snake_case )
snake_case__ : Union[str, Any] = BatchFeature({input_name: speech_inputs}, tensor_type='np' )
snake_case__ : Dict = processed_features[input_name]
if len(batch_features_input.shape ) < 3:
snake_case__ : List[str] = batch_features_input[:, :, None]
self.assertTrue(
batch_features_input.shape
== (self.feat_extract_tester.batch_size, len(speech_inputs[0] ), self.feat_extract_tester.num_mel_bins) )
@require_torch
def lowercase_ ( self : List[str] ) ->Any:
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target(equal_length=_snake_case )
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Tuple = feat_extract.model_input_names[0]
snake_case__ : List[Any] = BatchFeature({input_name: speech_inputs}, tensor_type='pt' )
snake_case__ : Tuple = processed_features[input_name]
if len(batch_features_input.shape ) < 3:
snake_case__ : Any = batch_features_input[:, :, None]
self.assertTrue(
batch_features_input.shape
== (self.feat_extract_tester.batch_size, len(speech_inputs[0] ), self.feat_extract_tester.num_mel_bins) )
@require_torch
def lowercase_ ( self : Optional[int] ) ->Tuple:
snake_case__ : Dict = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Union[str, Any] = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : List[str] = BatchFeature({input_name: speech_inputs} )
snake_case__ : int = feat_extract.num_mel_bins # hack!
snake_case__ : Tuple = feat_extract.pad(_snake_case, padding='longest', return_tensors='np' )[input_name]
snake_case__ : Union[str, Any] = feat_extract.pad(_snake_case, padding='longest', return_tensors='pt' )[input_name]
self.assertTrue(abs(input_np.astype(np.floataa ).sum() - input_pt.numpy().astype(np.floataa ).sum() ) < 1e-2 )
def lowercase_ ( self : int ) ->Any:
snake_case__ : Any = self.feat_extract_dict
snake_case__ : List[Any] = True
snake_case__ : Union[str, Any] = self.feature_extraction_class(**_snake_case )
snake_case__ : Any = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : List[Any] = [len(_snake_case ) for x in speech_inputs]
snake_case__ : Union[str, Any] = feat_extract.model_input_names[0]
snake_case__ : Optional[int] = BatchFeature({input_name: speech_inputs} )
snake_case__ : List[str] = feat_extract.num_mel_bins # hack!
snake_case__ : str = feat_extract.pad(_snake_case, padding='longest', return_tensors='np' )
self.assertIn('attention_mask', _snake_case )
self.assertListEqual(list(processed.attention_mask.shape ), list(processed[input_name].shape[:2] ) )
self.assertListEqual(processed.attention_mask.sum(-1 ).tolist(), _snake_case )
def lowercase_ ( self : Optional[int] ) ->str:
snake_case__ : int = self.feat_extract_dict
snake_case__ : List[str] = True
snake_case__ : Tuple = self.feature_extraction_class(**_snake_case )
snake_case__ : List[str] = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : str = [len(_snake_case ) for x in speech_inputs]
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : Optional[int] = BatchFeature({input_name: speech_inputs} )
snake_case__ : Optional[Any] = min(_snake_case )
snake_case__ : Union[str, Any] = feat_extract.num_mel_bins # hack!
snake_case__ : Tuple = feat_extract.pad(
_snake_case, padding='max_length', max_length=_snake_case, truncation=_snake_case, return_tensors='np' )
self.assertIn('attention_mask', _snake_case )
self.assertListEqual(
list(processed_pad.attention_mask.shape ), [processed_pad[input_name].shape[0], max_length] )
self.assertListEqual(
processed_pad.attention_mask[:, :max_length].sum(-1 ).tolist(), [max_length for x in speech_inputs] )
def lowercase_ ( self : List[Any], _snake_case : Optional[int] ) ->Optional[Any]:
from datasets import load_dataset
snake_case__ : str = load_dataset('hf-internal-testing/librispeech_asr_dummy', 'clean', split='validation' )
# automatic decoding with librispeech
snake_case__ : Dict = ds.sort('id' ).select(range(_snake_case ) )[:num_samples]['audio']
return [x["array"] for x in speech_samples]
def lowercase_ ( self : str ) ->str:
# fmt: off
snake_case__ : List[Any] = torch.tensor(
[2.3804e-03, 2.0752e-03, 1.9836e-03, 2.1057e-03, 1.6174e-03,
3.0518e-04, 9.1553e-05, 3.3569e-04, 9.7656e-04, 1.8311e-03,
2.0142e-03, 2.1057e-03, 1.7395e-03, 4.5776e-04, -3.9673e-04,
4.5776e-04, 1.0071e-03, 9.1553e-05, 4.8828e-04, 1.1597e-03,
7.3242e-04, 9.4604e-04, 1.8005e-03, 1.8311e-03, 8.8501e-04,
4.2725e-04, 4.8828e-04, 7.3242e-04, 1.0986e-03, 2.1057e-03] )
# fmt: on
snake_case__ : Union[str, Any] = self._load_datasamples(1 )
snake_case__ : Optional[int] = SpeechTaFeatureExtractor()
snake_case__ : List[Any] = feature_extractor(_snake_case, return_tensors='pt' ).input_values
self.assertEquals(input_values.shape, (1, 9_3_6_8_0) )
self.assertTrue(torch.allclose(input_values[0, :3_0], _snake_case, atol=1e-6 ) )
def lowercase_ ( self : Any ) ->str:
# fmt: off
snake_case__ : Optional[Any] = torch.tensor(
[-2.6_8_7_0, -3.0_1_0_4, -3.1_3_5_6, -3.5_3_5_2, -3.0_0_4_4, -3.0_3_5_3, -3.4_7_1_9, -3.6_7_7_7,
-3.1_5_2_0, -2.9_4_3_5, -2.6_5_5_3, -2.8_7_9_5, -2.9_9_4_4, -2.5_9_2_1, -3.0_2_7_9, -3.0_3_8_6,
-3.0_8_6_4, -3.1_2_9_1, -3.2_3_5_3, -2.7_4_4_4, -2.6_8_3_1, -2.7_2_8_7, -3.1_7_6_1, -3.1_5_7_1,
-3.2_7_2_6, -3.0_5_8_2, -3.1_0_0_7, -3.4_5_3_3, -3.4_6_9_5, -3.0_9_9_8] )
# fmt: on
snake_case__ : List[str] = self._load_datasamples(1 )
snake_case__ : str = SpeechTaFeatureExtractor()
snake_case__ : Optional[Any] = feature_extractor(audio_target=_snake_case, return_tensors='pt' ).input_values
self.assertEquals(input_values.shape, (1, 3_6_6, 8_0) )
self.assertTrue(torch.allclose(input_values[0, 0, :3_0], _snake_case, atol=1e-4 ) )
| 277 | 0 |
import dataclasses
import json
import warnings
from dataclasses import dataclass, field
from time import time
from typing import List
from ..utils import logging
UpperCAmelCase__ = logging.get_logger(__name__)
def A ( _UpperCAmelCase : Tuple=None , _UpperCAmelCase : Tuple=None ) -> Dict:
'''simple docstring'''
return field(default_factory=lambda: default , metadata=_UpperCAmelCase )
@dataclass
class __lowerCAmelCase :
UpperCamelCase = list_field(
default=[] , metadata={
'''help''': (
'''Model checkpoints to be provided to the AutoModel classes. Leave blank to benchmark the base version'''
''' of all available models'''
)
} , )
UpperCamelCase = list_field(
default=[8] , metadata={'''help''': '''List of batch sizes for which memory and time performance will be evaluated'''} )
UpperCamelCase = list_field(
default=[8, 3_2, 1_2_8, 5_1_2] , metadata={'''help''': '''List of sequence lengths for which memory and time performance will be evaluated'''} , )
UpperCamelCase = field(
default=lowerCAmelCase_ , metadata={'''help''': '''Whether to benchmark inference of model. Inference can be disabled via --no-inference.'''} , )
UpperCamelCase = field(
default=lowerCAmelCase_ , metadata={'''help''': '''Whether to run on available cuda devices. Cuda can be disabled via --no-cuda.'''} , )
UpperCamelCase = field(
default=lowerCAmelCase_ , metadata={'''help''': '''Whether to run on available tpu devices. TPU can be disabled via --no-tpu.'''} )
UpperCamelCase = field(default=lowerCAmelCase_ , metadata={'''help''': '''Use FP16 to accelerate inference.'''} )
UpperCamelCase = field(default=lowerCAmelCase_ , metadata={'''help''': '''Benchmark training of model'''} )
UpperCamelCase = field(default=lowerCAmelCase_ , metadata={'''help''': '''Verbose memory tracing'''} )
UpperCamelCase = field(
default=lowerCAmelCase_ , metadata={'''help''': '''Whether to perform speed measurements. Speed measurements can be disabled via --no-speed.'''} , )
UpperCamelCase = field(
default=lowerCAmelCase_ , metadata={
'''help''': '''Whether to perform memory measurements. Memory measurements can be disabled via --no-memory'''
} , )
UpperCamelCase = field(default=lowerCAmelCase_ , metadata={'''help''': '''Trace memory line by line'''} )
UpperCamelCase = field(default=lowerCAmelCase_ , metadata={'''help''': '''Save result to a CSV file'''} )
UpperCamelCase = field(default=lowerCAmelCase_ , metadata={'''help''': '''Save all print statements in a log file'''} )
UpperCamelCase = field(default=lowerCAmelCase_ , metadata={'''help''': '''Whether to print environment information'''} )
UpperCamelCase = field(
default=lowerCAmelCase_ , metadata={
'''help''': (
'''Whether to use multiprocessing for memory and speed measurement. It is highly recommended to use'''
''' multiprocessing for accurate CPU and GPU memory measurements. This option should only be disabled'''
''' for debugging / testing and on TPU.'''
)
} , )
UpperCamelCase = field(
default=f'''inference_time_{round(time() )}.csv''' , metadata={'''help''': '''CSV filename used if saving time results to csv.'''} , )
UpperCamelCase = field(
default=f'''inference_memory_{round(time() )}.csv''' , metadata={'''help''': '''CSV filename used if saving memory results to csv.'''} , )
UpperCamelCase = field(
default=f'''train_time_{round(time() )}.csv''' , metadata={'''help''': '''CSV filename used if saving time results to csv for training.'''} , )
UpperCamelCase = field(
default=f'''train_memory_{round(time() )}.csv''' , metadata={'''help''': '''CSV filename used if saving memory results to csv for training.'''} , )
UpperCamelCase = field(
default=f'''env_info_{round(time() )}.csv''' , metadata={'''help''': '''CSV filename used if saving environment information.'''} , )
UpperCamelCase = field(
default=f'''log_{round(time() )}.csv''' , metadata={'''help''': '''Log filename used if print statements are saved in log.'''} , )
UpperCamelCase = field(default=3 , metadata={'''help''': '''Times an experiment will be run.'''} )
UpperCamelCase = field(
default=lowerCAmelCase_ , metadata={
'''help''': (
'''Instead of loading the model as defined in `config.architectures` if exists, just load the pretrain'''
''' model weights.'''
)
} , )
def _lowerCamelCase ( self : Optional[Any]) -> List[str]:
"""simple docstring"""
warnings.warn(
F"The class {self.__class__} is deprecated. Hugging Face Benchmarking utils"
' are deprecated in general and it is advised to use external Benchmarking libraries '
' to benchmark Transformer models.' , _snake_case , )
def _lowerCamelCase ( self : Tuple) -> int:
"""simple docstring"""
return json.dumps(dataclasses.asdict(self) , indent=2)
@property
def _lowerCamelCase ( self : Optional[Any]) -> List[str]:
"""simple docstring"""
if len(self.models) <= 0:
raise ValueError(
'Please make sure you provide at least one model name / model identifier, *e.g.* `--models'
' bert-base-cased` or `args.models = [\'bert-base-cased\'].')
return self.models
@property
def _lowerCamelCase ( self : Union[str, Any]) -> Tuple:
"""simple docstring"""
if not self.multi_process:
return False
elif self.is_tpu:
logger.info('Multiprocessing is currently not possible on TPU.')
return False
else:
return True
| 339 |
# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from ..models.auto import AutoModelForSeqaSeqLM, AutoTokenizer
from .base import PipelineTool
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """philschmid/bart-large-cnn-samsum"""
_SCREAMING_SNAKE_CASE = (
"""This is a tool that summarizes an English text. It takes an input `text` containing the text to summarize, """
"""and returns a summary of the text."""
)
_SCREAMING_SNAKE_CASE = """summarizer"""
_SCREAMING_SNAKE_CASE = AutoTokenizer
_SCREAMING_SNAKE_CASE = AutoModelForSeqaSeqLM
_SCREAMING_SNAKE_CASE = ["""text"""]
_SCREAMING_SNAKE_CASE = ["""text"""]
def lowercase_ ( self : Optional[Any], _snake_case : str ) ->Any:
return self.pre_processor(_snake_case, return_tensors='pt', truncation=_snake_case )
def lowercase_ ( self : int, _snake_case : List[Any] ) ->Any:
return self.model.generate(**_snake_case )[0]
def lowercase_ ( self : int, _snake_case : int ) ->str:
return self.pre_processor.decode(_snake_case, skip_special_tokens=_snake_case, clean_up_tokenization_spaces=_snake_case )
| 277 | 0 |
# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from ..models.auto import AutoModelForSeqaSeqLM, AutoTokenizer
from .base import PipelineTool
class UpperCamelCase_ ( lowerCAmelCase_ ):
'''simple docstring'''
lowerCAmelCase = '''philschmid/bart-large-cnn-samsum'''
lowerCAmelCase = (
'''This is a tool that summarizes an English text. It takes an input `text` containing the text to summarize, '''
'''and returns a summary of the text.'''
)
lowerCAmelCase = '''summarizer'''
lowerCAmelCase = AutoTokenizer
lowerCAmelCase = AutoModelForSeqaSeqLM
lowerCAmelCase = ['''text''']
lowerCAmelCase = ['''text''']
def _UpperCamelCase ( self , a ) -> Any:
return self.pre_processor(_snake_case , return_tensors='pt' , truncation=_snake_case )
def _UpperCamelCase ( self , a ) -> Any:
return self.model.generate(**_snake_case )[0]
def _UpperCamelCase ( self , a ) -> str:
return self.pre_processor.decode(_snake_case , skip_special_tokens=_snake_case , clean_up_tokenization_spaces=_snake_case )
| 178 |
import argparse
import torch
from transformers import LxmertConfig, LxmertForPreTraining, load_tf_weights_in_lxmert
from transformers.utils import logging
logging.set_verbosity_info()
def lowercase_ (A : str , A : List[Any] , A : Any ):
# Initialise PyTorch model
snake_case__ : List[Any] = LxmertConfig.from_json_file(A )
print(F'''Building PyTorch model from configuration: {config}''' )
snake_case__ : List[str] = LxmertForPreTraining(A )
# Load weights from tf checkpoint
load_tf_weights_in_lxmert(A , A , A )
# Save pytorch-model
print(F'''Save PyTorch model to {pytorch_dump_path}''' )
torch.save(model.state_dict() , A )
if __name__ == "__main__":
a_ :Union[str, Any] = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--tf_checkpoint_path", default=None, type=str, required=True, help="Path to the TensorFlow checkpoint path."
)
parser.add_argument(
"--config_file",
default=None,
type=str,
required=True,
help="The config json file corresponding to the pre-trained model. \nThis specifies the model architecture.",
)
parser.add_argument(
"--pytorch_dump_path", default=None, type=str, required=True, help="Path to the output PyTorch model."
)
a_ :Optional[int] = parser.parse_args()
convert_tf_checkpoint_to_pytorch(args.tf_checkpoint_path, args.config_file, args.pytorch_dump_path)
| 277 | 0 |
"""simple docstring"""
from typing import List
from ...configuration_utils import PretrainedConfig
from ...utils import logging
__UpperCamelCase = logging.get_logger(__name__)
__UpperCamelCase = {
"snap-research/efficientformer-l1-300": (
"https://huggingface.co/snap-research/efficientformer-l1-300/resolve/main/config.json"
),
}
class lowerCAmelCase ( lowerCAmelCase_ ):
'''simple docstring'''
SCREAMING_SNAKE_CASE_ : Tuple = """efficientformer"""
def __init__( self , lowerCAmelCase__ = [3, 2, 6, 4] , lowerCAmelCase__ = [48, 96, 224, 448] , lowerCAmelCase__ = [True, True, True, True] , lowerCAmelCase__ = 448 , lowerCAmelCase__ = 32 , lowerCAmelCase__ = 4 , lowerCAmelCase__ = 7 , lowerCAmelCase__ = 5 , lowerCAmelCase__ = 8 , lowerCAmelCase__ = 4 , lowerCAmelCase__ = 0.0 , lowerCAmelCase__ = 16 , lowerCAmelCase__ = 3 , lowerCAmelCase__ = 3 , lowerCAmelCase__ = 3 , lowerCAmelCase__ = 2 , lowerCAmelCase__ = 1 , lowerCAmelCase__ = 0.0 , lowerCAmelCase__ = 1 , lowerCAmelCase__ = True , lowerCAmelCase__ = True , lowerCAmelCase__ = 1e-5 , lowerCAmelCase__ = "gelu" , lowerCAmelCase__ = 0.02 , lowerCAmelCase__ = 1e-12 , lowerCAmelCase__ = 224 , lowerCAmelCase__ = 1e-05 , **lowerCAmelCase__ , ) -> None:
super().__init__(**_snake_case )
SCREAMING_SNAKE_CASE = hidden_act
SCREAMING_SNAKE_CASE = hidden_dropout_prob
SCREAMING_SNAKE_CASE = hidden_sizes
SCREAMING_SNAKE_CASE = num_hidden_layers
SCREAMING_SNAKE_CASE = num_attention_heads
SCREAMING_SNAKE_CASE = initializer_range
SCREAMING_SNAKE_CASE = layer_norm_eps
SCREAMING_SNAKE_CASE = patch_size
SCREAMING_SNAKE_CASE = num_channels
SCREAMING_SNAKE_CASE = depths
SCREAMING_SNAKE_CASE = mlp_expansion_ratio
SCREAMING_SNAKE_CASE = downsamples
SCREAMING_SNAKE_CASE = dim
SCREAMING_SNAKE_CASE = key_dim
SCREAMING_SNAKE_CASE = attention_ratio
SCREAMING_SNAKE_CASE = resolution
SCREAMING_SNAKE_CASE = pool_size
SCREAMING_SNAKE_CASE = downsample_patch_size
SCREAMING_SNAKE_CASE = downsample_stride
SCREAMING_SNAKE_CASE = downsample_pad
SCREAMING_SNAKE_CASE = drop_path_rate
SCREAMING_SNAKE_CASE = num_metaad_blocks
SCREAMING_SNAKE_CASE = distillation
SCREAMING_SNAKE_CASE = use_layer_scale
SCREAMING_SNAKE_CASE = layer_scale_init_value
SCREAMING_SNAKE_CASE = image_size
SCREAMING_SNAKE_CASE = batch_norm_eps
| 113 |
import argparse
import fairseq
import torch
from torch import nn
from transformers import (
MBartaaTokenizer,
MBartConfig,
MBartForCausalLM,
SpeechEncoderDecoderConfig,
SpeechEncoderDecoderModel,
WavaVecaConfig,
WavaVecaFeatureExtractor,
WavaVecaModel,
logging,
)
logging.set_verbosity_info()
a_ :Tuple = logging.get_logger(__name__)
a_ :List[Any] = {
"post_extract_proj": "feature_projection.projection",
"encoder.pos_conv.0": "encoder.pos_conv_embed.conv",
"self_attn.k_proj": "encoder.layers.*.attention.k_proj",
"self_attn.v_proj": "encoder.layers.*.attention.v_proj",
"self_attn.q_proj": "encoder.layers.*.attention.q_proj",
"self_attn.out_proj": "encoder.layers.*.attention.out_proj",
"self_attn_layer_norm": "encoder.layers.*.layer_norm",
"fc1": "encoder.layers.*.feed_forward.intermediate_dense",
"fc2": "encoder.layers.*.feed_forward.output_dense",
"final_layer_norm": "encoder.layers.*.final_layer_norm",
"encoder.layer_norm": "encoder.layer_norm",
"w2v_model.layer_norm": "feature_projection.layer_norm",
"quantizer.weight_proj": "quantizer.weight_proj",
"quantizer.vars": "quantizer.codevectors",
"project_q": "project_q",
"final_proj": "project_hid",
"w2v_encoder.proj": "lm_head",
"mask_emb": "masked_spec_embed",
}
a_ :Optional[int] = [
"lm_head",
"quantizer.weight_proj",
"quantizer.codevectors",
"project_q",
"project_hid",
]
def lowercase_ (A : Union[str, Any] , A : str , A : Dict , A : Optional[Any] , A : Optional[Any] ):
for attribute in key.split('.' ):
snake_case__ : Any = getattr(A , A )
if weight_type is not None:
snake_case__ : Optional[Any] = getattr(A , A ).shape
else:
snake_case__ : Optional[int] = hf_pointer.shape
assert hf_shape == value.shape, (
F'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
F''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
snake_case__ : Tuple = value
elif weight_type == "weight_g":
snake_case__ : Tuple = value
elif weight_type == "weight_v":
snake_case__ : List[Any] = value
elif weight_type == "bias":
snake_case__ : List[Any] = value
else:
snake_case__ : Optional[Any] = value
logger.info(F'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def lowercase_ (A : str , A : Any ):
snake_case__ : Union[str, Any] = []
snake_case__ : Union[str, Any] = fairseq_model.state_dict()
snake_case__ : Union[str, Any] = hf_model.feature_extractor
snake_case__ : Any = hf_model.adapter
for name, value in fairseq_dict.items():
snake_case__ : Any = False
if "conv_layers" in name:
load_conv_layer(
A , A , A , A , hf_model.config.feat_extract_norm == 'group' , )
snake_case__ : List[Any] = True
elif any(x in name for x in ['adaptor', 'w2v_encoder.proj.', 'w2v_proj_ln.'] ):
load_adapter(A , A , A , A )
snake_case__ : Optional[Any] = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split('w2v_model.' )[-1] == name.split('.' )[0]:
snake_case__ : Tuple = True
if "*" in mapped_key:
snake_case__ : List[Any] = name.split(A )[0].split('.' )[-2]
snake_case__ : Optional[int] = mapped_key.replace('*' , A )
if "weight_g" in name:
snake_case__ : Optional[int] = 'weight_g'
elif "weight_v" in name:
snake_case__ : Optional[Any] = 'weight_v'
elif "bias" in name:
snake_case__ : Union[str, Any] = 'bias'
elif "weight" in name:
snake_case__ : Optional[int] = 'weight'
else:
snake_case__ : Tuple = None
set_recursively(A , A , A , A , A )
continue
if not is_used:
unused_weights.append(A )
logger.warning(F'''Unused weights: {unused_weights}''' )
def lowercase_ (A : Union[str, Any] , A : Any , A : str , A : str , A : int ):
snake_case__ : str = full_name.split('conv_layers.' )[-1]
snake_case__ : Optional[int] = name.split('.' )
snake_case__ : Tuple = int(items[0] )
snake_case__ : Any = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
snake_case__ : Union[str, Any] = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
snake_case__ : Union[str, Any] = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
F'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
snake_case__ : Optional[int] = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
snake_case__ : Optional[Any] = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(A )
def lowercase_ (A : Optional[Any] , A : Any , A : Tuple , A : Any ):
snake_case__ : List[str] = full_name.split('adaptor.' )[-1]
snake_case__ : Tuple = name.split('.' )
if items[1].isdigit():
snake_case__ : Optional[int] = int(items[1] )
else:
snake_case__ : Any = None
if "adaptor" not in full_name:
if "proj_ln" in full_name:
# has to be layer norm
if "bias" in name:
assert (
value.shape == adapter.proj_layer_norm.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj_layer_norm.bias.data.shape} was found.'''
snake_case__ : List[Any] = value
logger.info(F'''Adapter proj layer norm bias was initialized from {full_name}.''' )
if "weight" in name:
assert (
value.shape == adapter.proj_layer_norm.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj_layer_norm.weight.data.shape} was found.'''
snake_case__ : int = value
else:
# has to be projection layer
if "bias" in name:
assert (
value.shape == adapter.proj.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj.bias.data.shape} was found.'''
snake_case__ : str = value
logger.info(F'''Adapter proj layer bias was initialized from {full_name}.''' )
if "weight" in name:
assert (
value.shape == adapter.proj.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj.weight.data.shape} was found.'''
snake_case__ : Dict = value
logger.info(F'''Adapter proj layer weight was initialized from {full_name}.''' )
elif isinstance(A , A ):
if "bias" in name:
assert (
value.shape == adapter.layers[layer_id].conv.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.layers[layer_id].conv.bias.data.shape} was found.'''
snake_case__ : List[str] = value
logger.info(F'''Adapter layer {layer_id} bias was initialized from {full_name}.''' )
elif "weight" in name:
assert (
value.shape == adapter.layers[layer_id].conv.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.layers[layer_id].conv.weight.data.shape} was found.'''
snake_case__ : List[str] = value
logger.info(F'''Adapter layer {layer_id} bias was initialized from {full_name}.''' )
else:
unused_weights.append(A )
def lowercase_ (A : int ):
snake_case__ , snake_case__ : Union[str, Any] = emb.weight.shape
snake_case__ : int = nn.Linear(A , A , bias=A )
snake_case__ : Optional[Any] = emb.weight.data
return lin_layer
@torch.no_grad()
def lowercase_ (A : Tuple , A : Tuple , A : Any , A : Optional[Any] , A : int , A : Optional[Any] , A : Union[str, Any] , A : Union[str, Any] , A : Optional[Any] , A : List[Any] , A : Union[str, Any] , ):
snake_case__ : Optional[Any] = WavaVecaConfig.from_pretrained(
A , add_adapter=A , adapter_stride=A , adapter_kernel_size=A , use_auth_token=A , output_hidden_size=A , )
snake_case__ : Dict = MBartConfig.from_pretrained(A )
# load model
snake_case__ , snake_case__ , snake_case__ : Any = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] , arg_overrides={
'config_yaml': config_yaml_path,
'data': '/'.join(dict_path.split('/' )[:-1] ),
'w2v_path': checkpoint_path,
'load_pretrained_decoder_from': None,
} , )
snake_case__ : List[Any] = model[0].eval()
# load feature extractor
snake_case__ : str = WavaVecaFeatureExtractor.from_pretrained(A , use_auth_token=A )
# set weights for wav2vec2 encoder
snake_case__ : List[str] = WavaVecaModel(A )
recursively_load_weights_wavaveca(model.encoder , A )
# load decoder weights
snake_case__ : Any = MBartForCausalLM(A )
snake_case__ , snake_case__ : int = hf_decoder.model.decoder.load_state_dict(model.decoder.state_dict() , strict=A )
logger.warning(F'''The following keys are missing when loading the decoder weights: {missing_keys}''' )
logger.warning(F'''The following keys are unexpected when loading the decoder weights: {unexpected_keys}''' )
snake_case__ : Union[str, Any] = SpeechEncoderDecoderModel(encoder=A , decoder=A )
snake_case__ : str = False
snake_case__ : int = MBartaaTokenizer(A )
tokenizer.save_pretrained(A )
snake_case__ : Any = hf_wavavec.config.to_dict()
snake_case__ : Tuple = tokenizer.pad_token_id
snake_case__ : Union[str, Any] = tokenizer.bos_token_id
snake_case__ : Dict = tokenizer.eos_token_id
snake_case__ : Optional[int] = 'mbart50'
snake_case__ : Union[str, Any] = 'wav2vec2'
snake_case__ : List[str] = tokenizer.eos_token_id
snake_case__ : Union[str, Any] = 2_5_0_0_0_4
snake_case__ : int = tokenizer.eos_token_id
snake_case__ : Union[str, Any] = SpeechEncoderDecoderConfig.from_dict(A )
hf_wavavec.save_pretrained(A )
feature_extractor.save_pretrained(A )
if __name__ == "__main__":
a_ :str = argparse.ArgumentParser()
parser.add_argument("--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model.")
parser.add_argument("--checkpoint_path", default=None, type=str, help="Path to fairseq checkpoint")
parser.add_argument("--dict_path", default=None, type=str, help="Path to dict of fine-tuned model")
parser.add_argument("--config_yaml_path", default=None, type=str, help="Path to yaml file of fine-tuned model")
parser.add_argument(
"--encoder_config_path",
default="facebook/wav2vec2-xls-r-1b",
type=str,
help="Path to hf encoder wav2vec2 checkpoint config",
)
parser.add_argument(
"--decoder_config_path",
default="facebook/mbart-large-50-one-to-many-mmt",
type=str,
help="Path to hf decoder checkpoint config",
)
parser.add_argument("--add_adapter", default=True, type=bool, help="whethere to add model adapter layers")
parser.add_argument("--adapter_stride", default=2, type=int, help="stride of adapter layers")
parser.add_argument("--adapter_kernel_size", default=3, type=int, help="kernel size of adapter layers")
parser.add_argument("--encoder_output_dim", default=1_024, type=int, help="encoder output dim")
parser.add_argument("--start_token_id", default=250_004, type=int, help="`decoder_start_token_id` of model config")
a_ :Union[str, Any] = parser.parse_args()
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.dict_path,
args.config_yaml_path,
encoder_config_path=args.encoder_config_path,
decoder_config_path=args.decoder_config_path,
add_adapter=args.add_adapter,
adapter_kernel_size=args.adapter_kernel_size,
adapter_stride=args.adapter_stride,
decoder_start_token_id=args.start_token_id,
encoder_output_dim=args.encoder_output_dim,
)
| 277 | 0 |
'''simple docstring'''
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_flax_available, is_torch_available
_A : int = {
"configuration_longt5": ["LONGT5_PRETRAINED_CONFIG_ARCHIVE_MAP", "LongT5Config", "LongT5OnnxConfig"],
}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
_A : List[str] = [
"LONGT5_PRETRAINED_MODEL_ARCHIVE_LIST",
"LongT5EncoderModel",
"LongT5ForConditionalGeneration",
"LongT5Model",
"LongT5PreTrainedModel",
]
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
_A : int = [
"FlaxLongT5ForConditionalGeneration",
"FlaxLongT5Model",
"FlaxLongT5PreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_longta import LONGT5_PRETRAINED_CONFIG_ARCHIVE_MAP, LongTaConfig, LongTaOnnxConfig
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_longta import (
LONGT5_PRETRAINED_MODEL_ARCHIVE_LIST,
LongTaEncoderModel,
LongTaForConditionalGeneration,
LongTaModel,
LongTaPreTrainedModel,
)
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_flax_longta import (
FlaxLongTaForConditionalGeneration,
FlaxLongTaModel,
FlaxLongTaPreTrainedModel,
)
else:
import sys
_A : Optional[int] = _LazyModule(__name__, globals()['''__file__'''], _import_structure, module_spec=__spec__)
| 229 |
from collections import OrderedDict
from typing import TYPE_CHECKING, Any, Mapping, Optional, Union
from ...configuration_utils import PretrainedConfig
from ...onnx import OnnxConfig
from ...utils import logging
if TYPE_CHECKING:
from ... import FeatureExtractionMixin, PreTrainedTokenizerBase, TensorType
a_ :Tuple = logging.get_logger(__name__)
a_ :Union[str, Any] = {
"microsoft/deberta-v2-xlarge": "https://huggingface.co/microsoft/deberta-v2-xlarge/resolve/main/config.json",
"microsoft/deberta-v2-xxlarge": "https://huggingface.co/microsoft/deberta-v2-xxlarge/resolve/main/config.json",
"microsoft/deberta-v2-xlarge-mnli": (
"https://huggingface.co/microsoft/deberta-v2-xlarge-mnli/resolve/main/config.json"
),
"microsoft/deberta-v2-xxlarge-mnli": (
"https://huggingface.co/microsoft/deberta-v2-xxlarge-mnli/resolve/main/config.json"
),
}
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """deberta-v2"""
def __init__( self : Union[str, Any], _snake_case : Dict=1_2_8_1_0_0, _snake_case : Any=1_5_3_6, _snake_case : Tuple=2_4, _snake_case : int=2_4, _snake_case : Optional[int]=6_1_4_4, _snake_case : Optional[int]="gelu", _snake_case : Optional[int]=0.1, _snake_case : List[str]=0.1, _snake_case : str=5_1_2, _snake_case : Optional[int]=0, _snake_case : Optional[int]=0.0_2, _snake_case : Dict=1e-7, _snake_case : int=False, _snake_case : Any=-1, _snake_case : List[str]=0, _snake_case : Tuple=True, _snake_case : Any=None, _snake_case : Union[str, Any]=0, _snake_case : Tuple="gelu", **_snake_case : Union[str, Any], ) ->Optional[int]:
super().__init__(**_snake_case )
snake_case__ : Dict = hidden_size
snake_case__ : Optional[int] = num_hidden_layers
snake_case__ : Any = num_attention_heads
snake_case__ : List[Any] = intermediate_size
snake_case__ : List[Any] = hidden_act
snake_case__ : Union[str, Any] = hidden_dropout_prob
snake_case__ : Dict = attention_probs_dropout_prob
snake_case__ : List[str] = max_position_embeddings
snake_case__ : List[str] = type_vocab_size
snake_case__ : Optional[Any] = initializer_range
snake_case__ : Optional[int] = relative_attention
snake_case__ : Tuple = max_relative_positions
snake_case__ : Union[str, Any] = pad_token_id
snake_case__ : Optional[int] = position_biased_input
# Backwards compatibility
if type(_snake_case ) == str:
snake_case__ : int = [x.strip() for x in pos_att_type.lower().split('|' )]
snake_case__ : List[str] = pos_att_type
snake_case__ : Union[str, Any] = vocab_size
snake_case__ : Optional[int] = layer_norm_eps
snake_case__ : Optional[int] = kwargs.get('pooler_hidden_size', _snake_case )
snake_case__ : int = pooler_dropout
snake_case__ : str = pooler_hidden_act
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
@property
def lowercase_ ( self : Optional[int] ) ->Mapping[str, Mapping[int, str]]:
if self.task == "multiple-choice":
snake_case__ : List[Any] = {0: 'batch', 1: 'choice', 2: 'sequence'}
else:
snake_case__ : int = {0: 'batch', 1: 'sequence'}
if self._config.type_vocab_size > 0:
return OrderedDict(
[('input_ids', dynamic_axis), ('attention_mask', dynamic_axis), ('token_type_ids', dynamic_axis)] )
else:
return OrderedDict([('input_ids', dynamic_axis), ('attention_mask', dynamic_axis)] )
@property
def lowercase_ ( self : Dict ) ->int:
return 1_2
def lowercase_ ( self : Tuple, _snake_case : Union["PreTrainedTokenizerBase", "FeatureExtractionMixin"], _snake_case : int = -1, _snake_case : int = -1, _snake_case : int = -1, _snake_case : bool = False, _snake_case : Optional["TensorType"] = None, _snake_case : int = 3, _snake_case : int = 4_0, _snake_case : int = 4_0, _snake_case : "PreTrainedTokenizerBase" = None, ) ->Mapping[str, Any]:
snake_case__ : Union[str, Any] = super().generate_dummy_inputs(preprocessor=_snake_case, framework=_snake_case )
if self._config.type_vocab_size == 0 and "token_type_ids" in dummy_inputs:
del dummy_inputs["token_type_ids"]
return dummy_inputs
| 277 | 0 |
"""simple docstring"""
import requests
UpperCAmelCase = "YOUR API KEY"
def lowercase ( a__ : str , a__ : str = giphy_api_key ) -> Optional[Any]:
_UpperCamelCase = '+'.join(query.split() )
_UpperCamelCase = F'''https://api.giphy.com/v1/gifs/search?q={formatted_query}&api_key={api_key}'''
_UpperCamelCase = requests.get(a__ ).json()['data']
return [gif["url"] for gif in gifs]
if __name__ == "__main__":
print("""\n""".join(get_gifs("""space ship""")))
| 256 |
import argparse
import json
import pickle
from pathlib import Path
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import MaskFormerConfig, MaskFormerForInstanceSegmentation, MaskFormerImageProcessor, SwinConfig
from transformers.utils import logging
logging.set_verbosity_info()
a_ :str = logging.get_logger(__name__)
def lowercase_ (A : str ):
snake_case__ : Tuple = SwinConfig.from_pretrained(
'microsoft/swin-tiny-patch4-window7-224' , out_features=['stage1', 'stage2', 'stage3', 'stage4'] )
snake_case__ : List[Any] = MaskFormerConfig(backbone_config=A )
snake_case__ : Union[str, Any] = 'huggingface/label-files'
if "ade20k-full" in model_name:
# this should be ok
snake_case__ : Dict = 8_4_7
snake_case__ : List[str] = 'maskformer-ade20k-full-id2label.json'
elif "ade" in model_name:
# this should be ok
snake_case__ : Union[str, Any] = 1_5_0
snake_case__ : Any = 'ade20k-id2label.json'
elif "coco-stuff" in model_name:
# this should be ok
snake_case__ : List[str] = 1_7_1
snake_case__ : Union[str, Any] = 'maskformer-coco-stuff-id2label.json'
elif "coco" in model_name:
# TODO
snake_case__ : Dict = 1_3_3
snake_case__ : str = 'coco-panoptic-id2label.json'
elif "cityscapes" in model_name:
# this should be ok
snake_case__ : List[str] = 1_9
snake_case__ : Union[str, Any] = 'cityscapes-id2label.json'
elif "vistas" in model_name:
# this should be ok
snake_case__ : Tuple = 6_5
snake_case__ : List[str] = 'mapillary-vistas-id2label.json'
snake_case__ : Dict = json.load(open(hf_hub_download(A , A , repo_type='dataset' ) , 'r' ) )
snake_case__ : List[str] = {int(A ): v for k, v in idalabel.items()}
return config
def lowercase_ (A : Any ):
snake_case__ : Optional[int] = []
# stem
# fmt: off
rename_keys.append(('backbone.patch_embed.proj.weight', 'model.pixel_level_module.encoder.model.embeddings.patch_embeddings.projection.weight') )
rename_keys.append(('backbone.patch_embed.proj.bias', 'model.pixel_level_module.encoder.model.embeddings.patch_embeddings.projection.bias') )
rename_keys.append(('backbone.patch_embed.norm.weight', 'model.pixel_level_module.encoder.model.embeddings.norm.weight') )
rename_keys.append(('backbone.patch_embed.norm.bias', 'model.pixel_level_module.encoder.model.embeddings.norm.bias') )
# stages
for i in range(len(config.backbone_config.depths ) ):
for j in range(config.backbone_config.depths[i] ):
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm1.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_before.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm1.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_before.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.relative_position_bias_table''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.self.relative_position_bias_table''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.relative_position_index''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.self.relative_position_index''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.proj.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.output.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.proj.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.output.dense.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm2.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_after.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm2.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_after.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc1.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.intermediate.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc1.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.intermediate.dense.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc2.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.output.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc2.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.output.dense.bias''') )
if i < 3:
rename_keys.append((F'''backbone.layers.{i}.downsample.reduction.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.reduction.weight''') )
rename_keys.append((F'''backbone.layers.{i}.downsample.norm.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.norm.weight''') )
rename_keys.append((F'''backbone.layers.{i}.downsample.norm.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.norm.bias''') )
rename_keys.append((F'''backbone.norm{i}.weight''', F'''model.pixel_level_module.encoder.hidden_states_norms.{i}.weight''') )
rename_keys.append((F'''backbone.norm{i}.bias''', F'''model.pixel_level_module.encoder.hidden_states_norms.{i}.bias''') )
# FPN
rename_keys.append(('sem_seg_head.layer_4.weight', 'model.pixel_level_module.decoder.fpn.stem.0.weight') )
rename_keys.append(('sem_seg_head.layer_4.norm.weight', 'model.pixel_level_module.decoder.fpn.stem.1.weight') )
rename_keys.append(('sem_seg_head.layer_4.norm.bias', 'model.pixel_level_module.decoder.fpn.stem.1.bias') )
for source_index, target_index in zip(range(3 , 0 , -1 ) , range(0 , 3 ) ):
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.0.weight''') )
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.norm.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.1.weight''') )
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.norm.bias''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.1.bias''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.0.weight''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.norm.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.1.weight''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.norm.bias''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.1.bias''') )
rename_keys.append(('sem_seg_head.mask_features.weight', 'model.pixel_level_module.decoder.mask_projection.weight') )
rename_keys.append(('sem_seg_head.mask_features.bias', 'model.pixel_level_module.decoder.mask_projection.bias') )
# Transformer decoder
for idx in range(config.decoder_config.decoder_layers ):
# self-attention out projection
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.out_proj.weight''', F'''model.transformer_module.decoder.layers.{idx}.self_attn.out_proj.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.out_proj.bias''', F'''model.transformer_module.decoder.layers.{idx}.self_attn.out_proj.bias''') )
# cross-attention out projection
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.out_proj.weight''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn.out_proj.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.out_proj.bias''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn.out_proj.bias''') )
# MLP 1
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear1.weight''', F'''model.transformer_module.decoder.layers.{idx}.fc1.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear1.bias''', F'''model.transformer_module.decoder.layers.{idx}.fc1.bias''') )
# MLP 2
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear2.weight''', F'''model.transformer_module.decoder.layers.{idx}.fc2.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear2.bias''', F'''model.transformer_module.decoder.layers.{idx}.fc2.bias''') )
# layernorm 1 (self-attention layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm1.weight''', F'''model.transformer_module.decoder.layers.{idx}.self_attn_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm1.bias''', F'''model.transformer_module.decoder.layers.{idx}.self_attn_layer_norm.bias''') )
# layernorm 2 (cross-attention layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm2.weight''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm2.bias''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn_layer_norm.bias''') )
# layernorm 3 (final layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm3.weight''', F'''model.transformer_module.decoder.layers.{idx}.final_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm3.bias''', F'''model.transformer_module.decoder.layers.{idx}.final_layer_norm.bias''') )
rename_keys.append(('sem_seg_head.predictor.transformer.decoder.norm.weight', 'model.transformer_module.decoder.layernorm.weight') )
rename_keys.append(('sem_seg_head.predictor.transformer.decoder.norm.bias', 'model.transformer_module.decoder.layernorm.bias') )
# heads on top
rename_keys.append(('sem_seg_head.predictor.query_embed.weight', 'model.transformer_module.queries_embedder.weight') )
rename_keys.append(('sem_seg_head.predictor.input_proj.weight', 'model.transformer_module.input_projection.weight') )
rename_keys.append(('sem_seg_head.predictor.input_proj.bias', 'model.transformer_module.input_projection.bias') )
rename_keys.append(('sem_seg_head.predictor.class_embed.weight', 'class_predictor.weight') )
rename_keys.append(('sem_seg_head.predictor.class_embed.bias', 'class_predictor.bias') )
for i in range(3 ):
rename_keys.append((F'''sem_seg_head.predictor.mask_embed.layers.{i}.weight''', F'''mask_embedder.{i}.0.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.mask_embed.layers.{i}.bias''', F'''mask_embedder.{i}.0.bias''') )
# fmt: on
return rename_keys
def lowercase_ (A : Tuple , A : Tuple , A : Optional[Any] ):
snake_case__ : Optional[int] = dct.pop(A )
snake_case__ : Union[str, Any] = val
def lowercase_ (A : Optional[Any] , A : Tuple ):
snake_case__ : Optional[int] = [int(backbone_config.embed_dim * 2**i ) for i in range(len(backbone_config.depths ) )]
for i in range(len(backbone_config.depths ) ):
snake_case__ : Optional[int] = num_features[i]
for j in range(backbone_config.depths[i] ):
# fmt: off
# read in weights + bias of input projection layer (in original implementation, this is a single matrix + bias)
snake_case__ : int = state_dict.pop(F'''backbone.layers.{i}.blocks.{j}.attn.qkv.weight''' )
snake_case__ : Tuple = state_dict.pop(F'''backbone.layers.{i}.blocks.{j}.attn.qkv.bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : str = in_proj_weight[:dim, :]
snake_case__ : int = in_proj_bias[: dim]
snake_case__ : List[Any] = in_proj_weight[
dim : dim * 2, :
]
snake_case__ : List[str] = in_proj_bias[
dim : dim * 2
]
snake_case__ : List[Any] = in_proj_weight[
-dim :, :
]
snake_case__ : Dict = in_proj_bias[-dim :]
# fmt: on
def lowercase_ (A : List[str] , A : List[Any] ):
# fmt: off
snake_case__ : str = config.decoder_config.hidden_size
for idx in range(config.decoder_config.decoder_layers ):
# read in weights + bias of self-attention input projection layer (in the original implementation, this is a single matrix + bias)
snake_case__ : List[Any] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.in_proj_weight''' )
snake_case__ : int = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.in_proj_bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : Any = in_proj_weight[: hidden_size, :]
snake_case__ : Tuple = in_proj_bias[:config.hidden_size]
snake_case__ : List[str] = in_proj_weight[hidden_size : hidden_size * 2, :]
snake_case__ : Dict = in_proj_bias[hidden_size : hidden_size * 2]
snake_case__ : Any = in_proj_weight[-hidden_size :, :]
snake_case__ : int = in_proj_bias[-hidden_size :]
# read in weights + bias of cross-attention input projection layer (in the original implementation, this is a single matrix + bias)
snake_case__ : List[Any] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.in_proj_weight''' )
snake_case__ : List[str] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.in_proj_bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : Optional[int] = in_proj_weight[: hidden_size, :]
snake_case__ : Optional[Any] = in_proj_bias[:config.hidden_size]
snake_case__ : int = in_proj_weight[hidden_size : hidden_size * 2, :]
snake_case__ : List[str] = in_proj_bias[hidden_size : hidden_size * 2]
snake_case__ : List[str] = in_proj_weight[-hidden_size :, :]
snake_case__ : str = in_proj_bias[-hidden_size :]
# fmt: on
def lowercase_ ():
snake_case__ : Any = 'http://images.cocodataset.org/val2017/000000039769.jpg'
snake_case__ : int = Image.open(requests.get(A , stream=A ).raw )
return im
@torch.no_grad()
def lowercase_ (A : str , A : str , A : str , A : bool = False ):
snake_case__ : Optional[int] = get_maskformer_config(A )
# load original state_dict
with open(A , 'rb' ) as f:
snake_case__ : List[Any] = pickle.load(A )
snake_case__ : Optional[int] = data['model']
# for name, param in state_dict.items():
# print(name, param.shape)
# rename keys
snake_case__ : List[str] = create_rename_keys(A )
for src, dest in rename_keys:
rename_key(A , A , A )
read_in_swin_q_k_v(A , config.backbone_config )
read_in_decoder_q_k_v(A , A )
# update to torch tensors
for key, value in state_dict.items():
snake_case__ : int = torch.from_numpy(A )
# load 🤗 model
snake_case__ : str = MaskFormerForInstanceSegmentation(A )
model.eval()
for name, param in model.named_parameters():
print(A , param.shape )
snake_case__ , snake_case__ : Union[str, Any] = model.load_state_dict(A , strict=A )
assert missing_keys == [
"model.pixel_level_module.encoder.model.layernorm.weight",
"model.pixel_level_module.encoder.model.layernorm.bias",
]
assert len(A ) == 0, F'''Unexpected keys: {unexpected_keys}'''
# verify results
snake_case__ : Optional[Any] = prepare_img()
if "vistas" in model_name:
snake_case__ : int = 6_5
elif "cityscapes" in model_name:
snake_case__ : Dict = 6_5_5_3_5
else:
snake_case__ : Tuple = 2_5_5
snake_case__ : Optional[int] = True if 'ade' in model_name else False
snake_case__ : Dict = MaskFormerImageProcessor(ignore_index=A , reduce_labels=A )
snake_case__ : Any = image_processor(A , return_tensors='pt' )
snake_case__ : Any = model(**A )
print('Logits:' , outputs.class_queries_logits[0, :3, :3] )
if model_name == "maskformer-swin-tiny-ade":
snake_case__ : Tuple = torch.tensor(
[[3.6353, -4.4770, -2.6065], [0.5081, -4.2394, -3.5343], [2.1909, -5.0353, -1.9323]] )
assert torch.allclose(outputs.class_queries_logits[0, :3, :3] , A , atol=1e-4 )
print('Looks ok!' )
if pytorch_dump_folder_path is not None:
print(F'''Saving model and image processor to {pytorch_dump_folder_path}''' )
Path(A ).mkdir(exist_ok=A )
model.save_pretrained(A )
image_processor.save_pretrained(A )
if push_to_hub:
print('Pushing model and image processor to the hub...' )
model.push_to_hub(F'''nielsr/{model_name}''' )
image_processor.push_to_hub(F'''nielsr/{model_name}''' )
if __name__ == "__main__":
a_ :Optional[int] = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--model_name",
default="maskformer-swin-tiny-ade",
type=str,
help=("Name of the MaskFormer model you'd like to convert",),
)
parser.add_argument(
"--checkpoint_path",
default="/Users/nielsrogge/Documents/MaskFormer_checkpoints/MaskFormer-Swin-tiny-ADE20k/model.pkl",
type=str,
help="Path to the original state dict (.pth file).",
)
parser.add_argument(
"--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model directory."
)
parser.add_argument(
"--push_to_hub", action="store_true", help="Whether or not to push the converted model to the 🤗 hub."
)
a_ :Dict = parser.parse_args()
convert_maskformer_checkpoint(
args.model_name, args.checkpoint_path, args.pytorch_dump_folder_path, args.push_to_hub
)
| 277 | 0 |
from timeit import timeit
def _a ( SCREAMING_SNAKE_CASE : int ) -> Optional[Any]:
"""simple docstring"""
if number < 0:
raise ValueError('the value of input must not be negative' )
__lowerCAmelCase: Dict = 0
while number:
number &= number - 1
result += 1
return result
def _a ( SCREAMING_SNAKE_CASE : int ) -> Optional[Any]:
"""simple docstring"""
if number < 0:
raise ValueError('the value of input must not be negative' )
__lowerCAmelCase: List[str] = 0
while number:
if number % 2 == 1:
result += 1
number >>= 1
return result
def _a ( ) -> str:
"""simple docstring"""
def do_benchmark(SCREAMING_SNAKE_CASE : int ) -> None:
__lowerCAmelCase: Dict = 'import __main__ as z'
print(f'''Benchmark when {number = }:''' )
print(f'''{get_set_bits_count_using_modulo_operator(SCREAMING_SNAKE_CASE ) = }''' )
__lowerCAmelCase: int = timeit('z.get_set_bits_count_using_modulo_operator(25)' , setup=SCREAMING_SNAKE_CASE )
print(f'''timeit() runs in {timing} seconds''' )
print(f'''{get_set_bits_count_using_brian_kernighans_algorithm(SCREAMING_SNAKE_CASE ) = }''' )
__lowerCAmelCase: Tuple = timeit(
'z.get_set_bits_count_using_brian_kernighans_algorithm(25)' , setup=SCREAMING_SNAKE_CASE , )
print(f'''timeit() runs in {timing} seconds''' )
for number in (25, 37, 58, 0):
do_benchmark(SCREAMING_SNAKE_CASE )
print()
if __name__ == "__main__":
import doctest
doctest.testmod()
benchmark()
| 322 |
import unittest
from transformers import is_torch_available
from transformers.testing_utils import require_torch, slow, torch_device
from ...generation.test_utils import GenerationTesterMixin
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers import (
OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST,
OpenAIGPTConfig,
OpenAIGPTDoubleHeadsModel,
OpenAIGPTForSequenceClassification,
OpenAIGPTLMHeadModel,
OpenAIGPTModel,
)
class snake_case__ :
"""simple docstring"""
def __init__( self : List[str], _snake_case : Any, _snake_case : int=1_3, _snake_case : Optional[int]=7, _snake_case : int=True, _snake_case : Optional[Any]=True, _snake_case : Optional[Any]=True, _snake_case : Union[str, Any]=9_9, _snake_case : Optional[Any]=3_2, _snake_case : Tuple=5, _snake_case : str=4, _snake_case : Any=3_7, _snake_case : int="gelu", _snake_case : Optional[Any]=0.1, _snake_case : str=0.1, _snake_case : str=5_1_2, _snake_case : Dict=1_6, _snake_case : str=2, _snake_case : Union[str, Any]=0.0_2, _snake_case : Optional[int]=3, _snake_case : Union[str, Any]=4, _snake_case : Tuple=None, ) ->Optional[Any]:
snake_case__ : Optional[int] = parent
snake_case__ : List[Any] = batch_size
snake_case__ : Tuple = seq_length
snake_case__ : str = is_training
snake_case__ : Optional[int] = use_token_type_ids
snake_case__ : Any = use_labels
snake_case__ : Dict = vocab_size
snake_case__ : str = hidden_size
snake_case__ : Union[str, Any] = num_hidden_layers
snake_case__ : List[str] = num_attention_heads
snake_case__ : Union[str, Any] = intermediate_size
snake_case__ : List[Any] = hidden_act
snake_case__ : int = hidden_dropout_prob
snake_case__ : str = attention_probs_dropout_prob
snake_case__ : Any = max_position_embeddings
snake_case__ : Union[str, Any] = type_vocab_size
snake_case__ : Optional[Any] = type_sequence_label_size
snake_case__ : Optional[int] = initializer_range
snake_case__ : Optional[int] = num_labels
snake_case__ : str = num_choices
snake_case__ : int = scope
snake_case__ : List[str] = self.vocab_size - 1
def lowercase_ ( self : Union[str, Any] ) ->Tuple:
snake_case__ : List[str] = ids_tensor([self.batch_size, self.seq_length], self.vocab_size )
snake_case__ : List[str] = None
if self.use_token_type_ids:
snake_case__ : Optional[Any] = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size )
snake_case__ : Tuple = None
snake_case__ : str = None
snake_case__ : List[Any] = None
if self.use_labels:
snake_case__ : Dict = ids_tensor([self.batch_size], self.type_sequence_label_size )
snake_case__ : int = ids_tensor([self.batch_size, self.seq_length], self.num_labels )
snake_case__ : List[str] = ids_tensor([self.batch_size], self.num_choices )
snake_case__ : Union[str, Any] = OpenAIGPTConfig(
vocab_size=self.vocab_size, n_embd=self.hidden_size, n_layer=self.num_hidden_layers, n_head=self.num_attention_heads, n_positions=self.max_position_embeddings, pad_token_id=self.pad_token_id, )
snake_case__ : List[str] = ids_tensor([self.num_hidden_layers, self.num_attention_heads], 2 )
return (
config,
input_ids,
head_mask,
token_type_ids,
sequence_labels,
token_labels,
choice_labels,
)
def lowercase_ ( self : Any, _snake_case : List[str], _snake_case : Any, _snake_case : List[Any], _snake_case : Tuple, *_snake_case : Optional[Any] ) ->Tuple:
snake_case__ : Union[str, Any] = OpenAIGPTModel(config=_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Optional[Any] = model(_snake_case, token_type_ids=_snake_case, head_mask=_snake_case )
snake_case__ : Union[str, Any] = model(_snake_case, token_type_ids=_snake_case )
snake_case__ : Optional[Any] = model(_snake_case )
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size) )
def lowercase_ ( self : Optional[int], _snake_case : Optional[Any], _snake_case : Union[str, Any], _snake_case : Optional[int], _snake_case : List[Any], *_snake_case : Dict ) ->Optional[int]:
snake_case__ : Optional[Any] = OpenAIGPTLMHeadModel(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Tuple = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.loss.shape, () )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size) )
def lowercase_ ( self : int, _snake_case : Tuple, _snake_case : List[str], _snake_case : List[Any], _snake_case : List[Any], *_snake_case : List[Any] ) ->Optional[int]:
snake_case__ : List[str] = OpenAIGPTDoubleHeadsModel(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Optional[Any] = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.loss.shape, () )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size) )
def lowercase_ ( self : Optional[int], _snake_case : Tuple, _snake_case : Dict, _snake_case : List[str], _snake_case : Optional[Any], *_snake_case : Union[str, Any] ) ->str:
snake_case__ : List[str] = self.num_labels
snake_case__ : Dict = OpenAIGPTForSequenceClassification(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : List[str] = ids_tensor([self.batch_size], self.type_sequence_label_size )
snake_case__ : List[str] = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels) )
def lowercase_ ( self : Dict ) ->int:
snake_case__ : List[Any] = self.prepare_config_and_inputs()
(
(
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) ,
) : Optional[Any] = config_and_inputs
snake_case__ : str = {
'input_ids': input_ids,
'token_type_ids': token_type_ids,
'head_mask': head_mask,
}
return config, inputs_dict
@require_torch
class snake_case__ ( lowerCAmelCase_ , lowerCAmelCase_ , lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = (
(OpenAIGPTModel, OpenAIGPTLMHeadModel, OpenAIGPTDoubleHeadsModel, OpenAIGPTForSequenceClassification)
if is_torch_available()
else ()
)
_SCREAMING_SNAKE_CASE = (
(OpenAIGPTLMHeadModel,) if is_torch_available() else ()
) # TODO (PVP): Add Double HeadsModel when generate() function is changed accordingly
_SCREAMING_SNAKE_CASE = (
{
"""feature-extraction""": OpenAIGPTModel,
"""text-classification""": OpenAIGPTForSequenceClassification,
"""text-generation""": OpenAIGPTLMHeadModel,
"""zero-shot""": OpenAIGPTForSequenceClassification,
}
if is_torch_available()
else {}
)
def lowercase_ ( self : Optional[int], _snake_case : Union[str, Any], _snake_case : int, _snake_case : Tuple, _snake_case : Tuple, _snake_case : List[str] ) ->Optional[Any]:
if pipeline_test_casse_name == "ZeroShotClassificationPipelineTests":
# Get `tokenizer does not have a padding token` error for both fast/slow tokenizers.
# `OpenAIGPTConfig` was never used in pipeline tests, either because of a missing checkpoint or because a
# tiny config could not be created.
return True
return False
def lowercase_ ( self : Optional[Any], _snake_case : Union[str, Any], _snake_case : List[str], _snake_case : Any=False ) ->Tuple:
snake_case__ : Optional[int] = super()._prepare_for_class(_snake_case, _snake_case, return_labels=_snake_case )
if return_labels:
if model_class.__name__ == "OpenAIGPTDoubleHeadsModel":
snake_case__ : Union[str, Any] = torch.zeros(
(self.model_tester.batch_size, self.model_tester.num_choices, self.model_tester.seq_length), dtype=torch.long, device=_snake_case, )
snake_case__ : List[Any] = inputs_dict['labels']
snake_case__ : List[Any] = inputs_dict['labels']
snake_case__ : Any = torch.zeros(
(self.model_tester.batch_size, self.model_tester.num_choices), dtype=torch.long, device=_snake_case, )
snake_case__ : Tuple = torch.zeros(
self.model_tester.batch_size, dtype=torch.long, device=_snake_case )
return inputs_dict
def lowercase_ ( self : Union[str, Any] ) ->List[str]:
snake_case__ : List[str] = OpenAIGPTModelTester(self )
snake_case__ : Any = ConfigTester(self, config_class=_snake_case, n_embd=3_7 )
def lowercase_ ( self : Optional[int] ) ->str:
self.config_tester.run_common_tests()
def lowercase_ ( self : int ) ->Tuple:
snake_case__ : str = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_openai_gpt_model(*_snake_case )
def lowercase_ ( self : Tuple ) ->List[str]:
snake_case__ : Optional[int] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_lm_head_model(*_snake_case )
def lowercase_ ( self : Dict ) ->int:
snake_case__ : Optional[Any] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_double_lm_head_model(*_snake_case )
def lowercase_ ( self : int ) ->str:
snake_case__ : List[Any] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_openai_gpt_for_sequence_classification(*_snake_case )
@slow
def lowercase_ ( self : Optional[Any] ) ->str:
for model_name in OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
snake_case__ : Optional[int] = OpenAIGPTModel.from_pretrained(_snake_case )
self.assertIsNotNone(_snake_case )
@require_torch
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
@slow
def lowercase_ ( self : Tuple ) ->Optional[int]:
snake_case__ : Union[str, Any] = OpenAIGPTLMHeadModel.from_pretrained('openai-gpt' )
model.to(_snake_case )
snake_case__ : Tuple = torch.tensor([[4_8_1, 4_7_3_5, 5_4_4]], dtype=torch.long, device=_snake_case ) # the president is
snake_case__ : int = [
4_8_1,
4_7_3_5,
5_4_4,
2_4_6,
9_6_3,
8_7_0,
7_6_2,
2_3_9,
2_4_4,
4_0_4_7_7,
2_4_4,
2_4_9,
7_1_9,
8_8_1,
4_8_7,
5_4_4,
2_4_0,
2_4_4,
6_0_3,
4_8_1,
] # the president is a very good man. " \n " i\'m sure he is, " said the
snake_case__ : Optional[int] = model.generate(_snake_case, do_sample=_snake_case )
self.assertListEqual(output_ids[0].tolist(), _snake_case )
| 277 | 0 |
import collections
import inspect
import unittest
from transformers import SwinvaConfig
from transformers.testing_utils import require_torch, require_vision, slow, torch_device
from transformers.utils import cached_property, is_torch_available, is_vision_available
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, _config_zero_init, floats_tensor, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from torch import nn
from transformers import SwinvaForImageClassification, SwinvaForMaskedImageModeling, SwinvaModel
from transformers.models.swinva.modeling_swinva import SWINV2_PRETRAINED_MODEL_ARCHIVE_LIST
if is_vision_available():
from PIL import Image
from transformers import AutoImageProcessor
class __lowerCAmelCase :
def __init__( self , lowerCAmelCase__ , lowerCAmelCase__=1_3 , lowerCAmelCase__=3_2 , lowerCAmelCase__=2 , lowerCAmelCase__=3 , lowerCAmelCase__=1_6 , lowerCAmelCase__=[1, 2, 1] , lowerCAmelCase__=[2, 2, 4] , lowerCAmelCase__=2 , lowerCAmelCase__=2.0 , lowerCAmelCase__=True , lowerCAmelCase__=0.0 , lowerCAmelCase__=0.0 , lowerCAmelCase__=0.1 , lowerCAmelCase__="gelu" , lowerCAmelCase__=False , lowerCAmelCase__=True , lowerCAmelCase__=0.02 , lowerCAmelCase__=1E-5 , lowerCAmelCase__=True , lowerCAmelCase__=None , lowerCAmelCase__=True , lowerCAmelCase__=1_0 , lowerCAmelCase__=8 , ) -> Union[str, Any]:
'''simple docstring'''
a__ : Any =parent
a__ : Tuple =batch_size
a__ : Tuple =image_size
a__ : Any =patch_size
a__ : Optional[int] =num_channels
a__ : Tuple =embed_dim
a__ : Any =depths
a__ : Any =num_heads
a__ : List[str] =window_size
a__ : Dict =mlp_ratio
a__ : Optional[int] =qkv_bias
a__ : Optional[Any] =hidden_dropout_prob
a__ : List[str] =attention_probs_dropout_prob
a__ : Union[str, Any] =drop_path_rate
a__ : str =hidden_act
a__ : Union[str, Any] =use_absolute_embeddings
a__ : Union[str, Any] =patch_norm
a__ : Any =layer_norm_eps
a__ : Tuple =initializer_range
a__ : Dict =is_training
a__ : Any =scope
a__ : Optional[Any] =use_labels
a__ : str =type_sequence_label_size
a__ : List[Any] =encoder_stride
def _lowercase ( self ) -> str:
'''simple docstring'''
a__ : Tuple =floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size] )
a__ : List[Any] =None
if self.use_labels:
a__ : Optional[Any] =ids_tensor([self.batch_size] , self.type_sequence_label_size )
a__ : Any =self.get_config()
return config, pixel_values, labels
def _lowercase ( self ) -> Optional[int]:
'''simple docstring'''
return SwinvaConfig(
image_size=self.image_size , patch_size=self.patch_size , num_channels=self.num_channels , embed_dim=self.embed_dim , depths=self.depths , num_heads=self.num_heads , window_size=self.window_size , mlp_ratio=self.mlp_ratio , qkv_bias=self.qkv_bias , hidden_dropout_prob=self.hidden_dropout_prob , attention_probs_dropout_prob=self.attention_probs_dropout_prob , drop_path_rate=self.drop_path_rate , hidden_act=self.hidden_act , use_absolute_embeddings=self.use_absolute_embeddings , path_norm=self.patch_norm , layer_norm_eps=self.layer_norm_eps , initializer_range=self.initializer_range , encoder_stride=self.encoder_stride , )
def _lowercase ( self , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ ) -> Dict:
'''simple docstring'''
a__ : List[Any] =SwinvaModel(config=_snake_case )
model.to(_snake_case )
model.eval()
a__ : Optional[int] =model(_snake_case )
a__ : List[Any] =((config.image_size // config.patch_size) ** 2) // (4 ** (len(config.depths ) - 1))
a__ : List[Any] =int(config.embed_dim * 2 ** (len(config.depths ) - 1) )
self.parent.assertEqual(result.last_hidden_state.shape , (self.batch_size, expected_seq_len, expected_dim) )
def _lowercase ( self , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ ) -> List[Any]:
'''simple docstring'''
a__ : List[str] =SwinvaForMaskedImageModeling(config=_snake_case )
model.to(_snake_case )
model.eval()
a__ : Union[str, Any] =model(_snake_case )
self.parent.assertEqual(
result.logits.shape , (self.batch_size, self.num_channels, self.image_size, self.image_size) )
# test greyscale images
a__ : Optional[Any] =1
a__ : Optional[int] =SwinvaForMaskedImageModeling(_snake_case )
model.to(_snake_case )
model.eval()
a__ : Tuple =floats_tensor([self.batch_size, 1, self.image_size, self.image_size] )
a__ : Any =model(_snake_case )
self.parent.assertEqual(result.logits.shape , (self.batch_size, 1, self.image_size, self.image_size) )
def _lowercase ( self , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ ) -> Any:
'''simple docstring'''
a__ : Tuple =self.type_sequence_label_size
a__ : int =SwinvaForImageClassification(_snake_case )
model.to(_snake_case )
model.eval()
a__ : Tuple =model(_snake_case , labels=_snake_case )
self.parent.assertEqual(result.logits.shape , (self.batch_size, self.type_sequence_label_size) )
def _lowercase ( self ) -> Dict:
'''simple docstring'''
a__ : str =self.prepare_config_and_inputs()
a__ : List[str] =config_and_inputs
a__ : Union[str, Any] ={'pixel_values': pixel_values}
return config, inputs_dict
@require_torch
class __lowerCAmelCase ( lowerCAmelCase_ , lowerCAmelCase_ , unittest.TestCase):
_lowercase : Union[str, Any] = (
(SwinvaModel, SwinvaForImageClassification, SwinvaForMaskedImageModeling) if is_torch_available() else ()
)
_lowercase : Optional[Any] = (
{"""feature-extraction""": SwinvaModel, """image-classification""": SwinvaForImageClassification}
if is_torch_available()
else {}
)
_lowercase : int = False
_lowercase : List[str] = False
_lowercase : List[Any] = False
_lowercase : int = False
def _lowercase ( self ) -> Dict:
'''simple docstring'''
a__ : Optional[int] =SwinvaModelTester(self )
a__ : int =ConfigTester(self , config_class=_snake_case , embed_dim=3_7 )
def _lowercase ( self ) -> int:
'''simple docstring'''
self.config_tester.create_and_test_config_to_json_string()
self.config_tester.create_and_test_config_to_json_file()
self.config_tester.create_and_test_config_from_and_save_pretrained()
self.config_tester.create_and_test_config_with_num_labels()
self.config_tester.check_config_can_be_init_without_params()
self.config_tester.check_config_arguments_init()
def _lowercase ( self ) -> str:
'''simple docstring'''
a__ : int =self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*_snake_case )
@unittest.skip(reason="Got `CUDA error: misaligned address` with PyTorch 2.0.0." )
def _lowercase ( self ) -> Union[str, Any]:
'''simple docstring'''
pass
@unittest.skip(reason="Swinv2 does not use inputs_embeds" )
def _lowercase ( self ) -> Union[str, Any]:
'''simple docstring'''
pass
def _lowercase ( self ) -> Union[str, Any]:
'''simple docstring'''
a__ : Optional[int] =self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
a__ : Union[str, Any] =model_class(_snake_case )
self.assertIsInstance(model.get_input_embeddings() , (nn.Module) )
a__ : Union[str, Any] =model.get_output_embeddings()
self.assertTrue(x is None or isinstance(_snake_case , nn.Linear ) )
def _lowercase ( self ) -> Optional[int]:
'''simple docstring'''
a__ : Any =self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
a__ : Any =model_class(_snake_case )
a__ : Dict =inspect.signature(model.forward )
# signature.parameters is an OrderedDict => so arg_names order is deterministic
a__ : Optional[Any] =[*signature.parameters.keys()]
a__ : List[Any] =['pixel_values']
self.assertListEqual(arg_names[:1] , _snake_case )
def _lowercase ( self ) -> Union[str, Any]:
'''simple docstring'''
a__ : Any =self.model_tester.prepare_config_and_inputs_for_common()
a__ : int =True
for model_class in self.all_model_classes:
a__ : str =True
a__ : Union[str, Any] =False
a__ : Tuple =True
a__ : int =model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
a__ : Optional[int] =model(**self._prepare_for_class(_snake_case , _snake_case ) )
a__ : List[str] =outputs.attentions
a__ : List[Any] =len(self.model_tester.depths )
self.assertEqual(len(_snake_case ) , _snake_case )
# check that output_attentions also work using config
del inputs_dict["output_attentions"]
a__ : str =True
a__ : Tuple =config.window_size**2
a__ : Optional[int] =model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
a__ : str =model(**self._prepare_for_class(_snake_case , _snake_case ) )
a__ : Tuple =outputs.attentions
self.assertEqual(len(_snake_case ) , _snake_case )
self.assertListEqual(
list(attentions[0].shape[-3:] ) , [self.model_tester.num_heads[0], window_size_squared, window_size_squared] , )
a__ : Optional[Any] =len(_snake_case )
# Check attention is always last and order is fine
a__ : Optional[int] =True
a__ : Dict =True
a__ : List[Any] =model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
a__ : Optional[int] =model(**self._prepare_for_class(_snake_case , _snake_case ) )
if hasattr(self.model_tester , "num_hidden_states_types" ):
a__ : str =self.model_tester.num_hidden_states_types
else:
# also another +1 for reshaped_hidden_states
a__ : Dict =2
self.assertEqual(out_len + added_hidden_states , len(_snake_case ) )
a__ : Any =outputs.attentions
self.assertEqual(len(_snake_case ) , _snake_case )
self.assertListEqual(
list(self_attentions[0].shape[-3:] ) , [self.model_tester.num_heads[0], window_size_squared, window_size_squared] , )
def _lowercase ( self , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ , lowerCAmelCase__ ) -> str:
'''simple docstring'''
a__ : Dict =model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
a__ : List[Any] =model(**self._prepare_for_class(_snake_case , _snake_case ) )
a__ : Dict =outputs.hidden_states
a__ : int =getattr(
self.model_tester , "expected_num_hidden_layers" , len(self.model_tester.depths ) + 1 )
self.assertEqual(len(_snake_case ) , _snake_case )
# Swinv2 has a different seq_length
a__ : int =(
config.patch_size
if isinstance(config.patch_size , collections.abc.Iterable )
else (config.patch_size, config.patch_size)
)
a__ : Optional[Any] =(image_size[1] // patch_size[1]) * (image_size[0] // patch_size[0])
self.assertListEqual(
list(hidden_states[0].shape[-2:] ) , [num_patches, self.model_tester.embed_dim] , )
a__ : Union[str, Any] =outputs.reshaped_hidden_states
self.assertEqual(len(_snake_case ) , _snake_case )
a__ : str =reshaped_hidden_states[0].shape
a__ : Any =(
reshaped_hidden_states[0].view(_snake_case , _snake_case , height * width ).permute(0 , 2 , 1 )
)
self.assertListEqual(
list(reshaped_hidden_states.shape[-2:] ) , [num_patches, self.model_tester.embed_dim] , )
def _lowercase ( self ) -> List[Any]:
'''simple docstring'''
a__ : Any =self.model_tester.prepare_config_and_inputs_for_common()
a__ : List[Any] =(
self.model_tester.image_size
if isinstance(self.model_tester.image_size , collections.abc.Iterable )
else (self.model_tester.image_size, self.model_tester.image_size)
)
for model_class in self.all_model_classes:
a__ : Optional[int] =True
self.check_hidden_states_output(_snake_case , _snake_case , _snake_case , _snake_case )
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
a__ : Dict =True
self.check_hidden_states_output(_snake_case , _snake_case , _snake_case , _snake_case )
def _lowercase ( self ) -> str:
'''simple docstring'''
a__ : Optional[Any] =self.model_tester.prepare_config_and_inputs_for_common()
a__ : List[str] =3
a__ : Union[str, Any] =(
self.model_tester.image_size
if isinstance(self.model_tester.image_size , collections.abc.Iterable )
else (self.model_tester.image_size, self.model_tester.image_size)
)
a__ : str =(
config.patch_size
if isinstance(config.patch_size , collections.abc.Iterable )
else (config.patch_size, config.patch_size)
)
a__ : Tuple =image_size[0] + patch_size[0] - (image_size[0] % patch_size[0])
a__ : Optional[Any] =image_size[1] + patch_size[1] - (image_size[1] % patch_size[1])
for model_class in self.all_model_classes:
a__ : int =True
self.check_hidden_states_output(_snake_case , _snake_case , _snake_case , (padded_height, padded_width) )
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
a__ : List[str] =True
self.check_hidden_states_output(_snake_case , _snake_case , _snake_case , (padded_height, padded_width) )
def _lowercase ( self ) -> Optional[int]:
'''simple docstring'''
a__ : Any =self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_masked_image_modeling(*_snake_case )
def _lowercase ( self ) -> str:
'''simple docstring'''
a__ : Any =self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_image_classification(*_snake_case )
@slow
def _lowercase ( self ) -> Union[str, Any]:
'''simple docstring'''
for model_name in SWINV2_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
a__ : Dict =SwinvaModel.from_pretrained(_snake_case )
self.assertIsNotNone(_snake_case )
def _lowercase ( self ) -> List[str]:
'''simple docstring'''
a__ : Any =self.model_tester.prepare_config_and_inputs_for_common()
a__ : List[Any] =_config_zero_init(_snake_case )
for model_class in self.all_model_classes:
a__ : List[str] =model_class(config=_snake_case )
for name, param in model.named_parameters():
if "embeddings" not in name and "logit_scale" not in name and param.requires_grad:
self.assertIn(
((param.data.mean() * 1E9).round() / 1E9).item() , [0.0, 1.0] , msg=F'''Parameter {name} of model {model_class} seems not properly initialized''' , )
@require_vision
@require_torch
class __lowerCAmelCase ( unittest.TestCase):
@cached_property
def _lowercase ( self ) -> List[str]:
'''simple docstring'''
return (
AutoImageProcessor.from_pretrained("microsoft/swinv2-tiny-patch4-window8-256" )
if is_vision_available()
else None
)
@slow
def _lowercase ( self ) -> List[Any]:
'''simple docstring'''
a__ : Any =SwinvaForImageClassification.from_pretrained("microsoft/swinv2-tiny-patch4-window8-256" ).to(
_snake_case )
a__ : int =self.default_image_processor
a__ : Union[str, Any] =Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png" )
a__ : Optional[Any] =image_processor(images=_snake_case , return_tensors="pt" ).to(_snake_case )
# forward pass
with torch.no_grad():
a__ : List[str] =model(**_snake_case )
# verify the logits
a__ : int =torch.Size((1, 1_0_0_0) )
self.assertEqual(outputs.logits.shape , _snake_case )
a__ : Optional[int] =torch.tensor([-0.39_47, -0.43_06, 0.00_26] ).to(_snake_case )
self.assertTrue(torch.allclose(outputs.logits[0, :3] , _snake_case , atol=1E-4 ) )
| 95 |
import os
import unittest
from transformers.models.transfo_xl.tokenization_transfo_xl import VOCAB_FILES_NAMES, TransfoXLTokenizer
from ...test_tokenization_common import TokenizerTesterMixin
class snake_case__ ( lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = TransfoXLTokenizer
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
def lowercase_ ( self : Optional[int] ) ->Any:
super().setUp()
snake_case__ : Tuple = [
'<unk>',
'[CLS]',
'[SEP]',
'want',
'unwanted',
'wa',
'un',
'running',
',',
'low',
'l',
]
snake_case__ : Any = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['vocab_file'] )
with open(self.vocab_file, 'w', encoding='utf-8' ) as vocab_writer:
vocab_writer.write(''.join([x + '\n' for x in vocab_tokens] ) )
def lowercase_ ( self : Union[str, Any], **_snake_case : List[Any] ) ->Dict:
snake_case__ : str = True
return TransfoXLTokenizer.from_pretrained(self.tmpdirname, **_snake_case )
def lowercase_ ( self : Optional[Any], _snake_case : str ) ->Dict:
snake_case__ : List[Any] = '<unk> UNwanted , running'
snake_case__ : List[Any] = '<unk> unwanted, running'
return input_text, output_text
def lowercase_ ( self : List[Any] ) ->Tuple:
snake_case__ : Dict = TransfoXLTokenizer(vocab_file=self.vocab_file, lower_case=_snake_case )
snake_case__ : str = tokenizer.tokenize('<unk> UNwanted , running' )
self.assertListEqual(_snake_case, ['<unk>', 'unwanted', ',', 'running'] )
self.assertListEqual(tokenizer.convert_tokens_to_ids(_snake_case ), [0, 4, 8, 7] )
def lowercase_ ( self : List[str] ) ->List[Any]:
snake_case__ : str = TransfoXLTokenizer(lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHeLLo ! how \n Are yoU ? ' ), ['hello', '!', 'how', 'are', 'you', '?'] )
def lowercase_ ( self : Optional[int] ) ->Optional[Any]:
snake_case__ : Optional[int] = TransfoXLTokenizer(lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHeLLo ! how \n Are yoU ? ' ), ['HeLLo', '!', 'how', 'Are', 'yoU', '?'] )
def lowercase_ ( self : Optional[int] ) ->Union[str, Any]:
snake_case__ : List[Any] = TransfoXLTokenizer(lower_case=_snake_case )
snake_case__ : Dict = 'Hello (bracket) and side-scrolled [and] Henry\'s $5,000 with 3.34 m. What\'s up!?'
snake_case__ : List[Any] = [
'Hello',
'(',
'bracket',
')',
'and',
'side',
'@-@',
'scrolled',
'[',
'and',
']',
'Henry',
'\'s',
'$',
'5',
'@,@',
'000',
'with',
'3',
'@.@',
'34',
'm',
'.',
'What',
'\'s',
'up',
'!',
'?',
]
self.assertListEqual(tokenizer.tokenize(_snake_case ), _snake_case )
self.assertEqual(tokenizer.convert_tokens_to_string(_snake_case ), _snake_case )
def lowercase_ ( self : Dict ) ->Any:
snake_case__ : Dict = self.get_tokenizer()
snake_case__ : Optional[Any] = len(_snake_case )
tokenizer.add_tokens(['new1', 'new2'] )
tokenizer.move_added_token('new1', 1 )
# Check that moved token is not copied (duplicate)
self.assertEqual(len(_snake_case ), original_len + 2 )
# Check that token is moved to specified id
self.assertEqual(tokenizer.encode('new1' ), [1] )
self.assertEqual(tokenizer.decode([1] ), 'new1' )
| 277 | 0 |
from math import factorial, pi
def a ( snake_case__: float , snake_case__: int = 30 ):
'''simple docstring'''
if not isinstance(snake_case__ , (int, float) ):
raise ValueError('''maclaurin_sin() requires either an int or float for theta''' )
if not isinstance(snake_case__ , snake_case__ ) or accuracy <= 0:
raise ValueError('''maclaurin_sin() requires a positive int for accuracy''' )
lowercase_ = float(snake_case__ )
lowercase_ = theta // (2 * pi)
theta -= 2 * div * pi
return sum(
(-1) ** r * theta ** (2 * r + 1) / factorial(2 * r + 1 ) for r in range(snake_case__ ) )
def a ( snake_case__: float , snake_case__: int = 30 ):
'''simple docstring'''
if not isinstance(snake_case__ , (int, float) ):
raise ValueError('''maclaurin_cos() requires either an int or float for theta''' )
if not isinstance(snake_case__ , snake_case__ ) or accuracy <= 0:
raise ValueError('''maclaurin_cos() requires a positive int for accuracy''' )
lowercase_ = float(snake_case__ )
lowercase_ = theta // (2 * pi)
theta -= 2 * div * pi
return sum((-1) ** r * theta ** (2 * r) / factorial(2 * r ) for r in range(snake_case__ ) )
if __name__ == "__main__":
import doctest
doctest.testmod()
print(maclaurin_sin(1_0))
print(maclaurin_sin(-1_0))
print(maclaurin_sin(1_0, 1_5))
print(maclaurin_sin(-1_0, 1_5))
print(maclaurin_cos(5))
print(maclaurin_cos(-5))
print(maclaurin_cos(1_0, 1_5))
print(maclaurin_cos(-1_0, 1_5))
| 30 |
from ...configuration_utils import PretrainedConfig
from ...utils import logging
a_ :Optional[int] = logging.get_logger(__name__)
a_ :Dict = {"openai-gpt": "https://huggingface.co/openai-gpt/resolve/main/config.json"}
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """openai-gpt"""
_SCREAMING_SNAKE_CASE = {
"""max_position_embeddings""": """n_positions""",
"""hidden_size""": """n_embd""",
"""num_attention_heads""": """n_head""",
"""num_hidden_layers""": """n_layer""",
}
def __init__( self : Optional[int], _snake_case : Dict=4_0_4_7_8, _snake_case : str=5_1_2, _snake_case : int=7_6_8, _snake_case : Tuple=1_2, _snake_case : Any=1_2, _snake_case : str="gelu", _snake_case : List[str]=0.1, _snake_case : Any=0.1, _snake_case : Dict=0.1, _snake_case : int=1e-5, _snake_case : Optional[Any]=0.0_2, _snake_case : List[Any]="cls_index", _snake_case : Any=True, _snake_case : Any=None, _snake_case : int=True, _snake_case : Optional[Any]=0.1, **_snake_case : List[Any], ) ->Optional[int]:
snake_case__ : int = vocab_size
snake_case__ : Dict = n_positions
snake_case__ : str = n_embd
snake_case__ : str = n_layer
snake_case__ : List[Any] = n_head
snake_case__ : List[Any] = afn
snake_case__ : Optional[Any] = resid_pdrop
snake_case__ : List[str] = embd_pdrop
snake_case__ : List[Any] = attn_pdrop
snake_case__ : Optional[int] = layer_norm_epsilon
snake_case__ : str = initializer_range
snake_case__ : List[str] = summary_type
snake_case__ : Optional[int] = summary_use_proj
snake_case__ : List[str] = summary_activation
snake_case__ : Optional[Any] = summary_first_dropout
snake_case__ : int = summary_proj_to_labels
super().__init__(**_snake_case )
| 277 | 0 |
"""simple docstring"""
from __future__ import annotations
class __lowerCAmelCase :
'''simple docstring'''
def __init__( self , _a=None ):
__a = data
__a = None
def __repr__( self ):
__a = []
__a = self
while temp:
string_rep.append(f'''{temp.data}''' )
__a = temp.next
return "->".join(_snake_case )
def lowercase ( lowerCAmelCase__ : list ) -> Optional[Any]:
if not elements_list:
raise Exception('''The Elements List is empty''' )
__a = Node(elements_list[0] )
for i in range(1 , len(lowerCAmelCase__ ) ):
__a = Node(elements_list[i] )
__a = current.next
return head
def lowercase ( lowerCAmelCase__ : Node ) -> Any:
if head_node is not None and isinstance(lowerCAmelCase__ , lowerCAmelCase__ ):
print_reverse(head_node.next )
print(head_node.data )
def lowercase ( ) -> Tuple:
from doctest import testmod
testmod()
__a = make_linked_list([14, 52, 14, 12, 43] )
print('''Linked List:''' )
print(lowerCAmelCase__ )
print('''Elements in Reverse:''' )
print_reverse(lowerCAmelCase__ )
if __name__ == "__main__":
main()
| 45 |
import argparse
import logging
import os
from datetime import datetime
import numpy as np
import torch
from torch import nn
from torch.utils.data import DataLoader, RandomSampler, TensorDataset
from tqdm import tqdm
from transformers import GPTaLMHeadModel
a_ :Optional[Any] = logging.getLogger(__name__)
def lowercase_ (A : List[Any] , A : List[Any] ):
# save results
if os.path.exists(A ):
if os.path.exists(os.path.join(A , 'config.json' ) ) and os.path.isfile(
os.path.join(A , 'config.json' ) ):
os.remove(os.path.join(A , 'config.json' ) )
if os.path.exists(os.path.join(A , 'pytorch_model.bin' ) ) and os.path.isfile(
os.path.join(A , 'pytorch_model.bin' ) ):
os.remove(os.path.join(A , 'pytorch_model.bin' ) )
else:
os.makedirs(A )
model.save_pretrained(A )
def lowercase_ (A : Any , A : Optional[Any]=False ):
snake_case__ : str = 2
if unlogit:
snake_case__ : Dict = torch.pow(A , A )
snake_case__ : Any = p * torch.log(A )
snake_case__ : Tuple = 0
return -plogp.sum(dim=-1 )
def lowercase_ (A : List[str] ):
logger.info('lv, h >\t' + '\t'.join(F'''{x + 1}''' for x in range(len(A ) ) ) )
for row in range(len(A ) ):
if tensor.dtype != torch.long:
logger.info(F'''layer {row + 1}:\t''' + '\t'.join(F'''{x:.5f}''' for x in tensor[row].cpu().data ) )
else:
logger.info(F'''layer {row + 1}:\t''' + '\t'.join(F'''{x:d}''' for x in tensor[row].cpu().data ) )
def lowercase_ (A : Tuple , A : Optional[Any] , A : str , A : int=True , A : Optional[int]=True , A : Any=None , A : int=False ):
snake_case__ , snake_case__ : Optional[Any] = model.config.num_hidden_layers, model.config.num_attention_heads
snake_case__ : int = torch.zeros(A , A ).to(args.device )
snake_case__ : Any = torch.zeros(A , A ).to(args.device )
if head_mask is None:
snake_case__ : Dict = torch.ones(A , A ).to(args.device )
head_mask.requires_grad_(requires_grad=A )
# If actually pruned attention multi-head, set head mask to None to avoid shape mismatch
if actually_pruned:
snake_case__ : Optional[int] = None
snake_case__ : List[Any] = 0.0
snake_case__ : str = 0.0
for step, inputs in enumerate(tqdm(A , desc='Iteration' , disable=args.local_rank not in [-1, 0] ) ):
snake_case__ : Union[str, Any] = tuple(t.to(args.device ) for t in inputs )
((snake_case__) , ) : Optional[Any] = inputs
# Do a forward pass (not with torch.no_grad() since we need gradients for importance score - see below)
snake_case__ : Union[str, Any] = model(A , labels=A , head_mask=A )
# (loss), lm_logits, presents, (all hidden_states), (attentions)
snake_case__ , snake_case__ , snake_case__ : Dict = (
outputs[0],
outputs[1],
outputs[-1],
) # Loss and logits are the first, attention the last
loss.backward() # Backpropagate to populate the gradients in the head mask
total_loss += loss.detach().cpu().numpy()
if compute_entropy:
for layer, attn in enumerate(A ):
snake_case__ : Optional[Any] = entropy(attn.detach() , A )
attn_entropy[layer] += masked_entropy.sum(-1 ).sum(0 ).sum(0 ).detach()
if compute_importance:
head_importance += head_mask.grad.abs().detach()
tot_tokens += torch.ones_like(A ).float().detach().sum().data
# Normalize
attn_entropy /= tot_tokens
head_importance /= tot_tokens
# Layerwise importance normalization
if not args.dont_normalize_importance_by_layer:
snake_case__ : Union[str, Any] = 2
snake_case__ : List[Any] = torch.pow(torch.pow(A , A ).sum(-1 ) , 1 / exponent )
head_importance /= norm_by_layer.unsqueeze(-1 ) + 1e-20
if not args.dont_normalize_global_importance:
snake_case__ : Tuple = (head_importance - head_importance.min()) / (head_importance.max() - head_importance.min())
# Print matrices
if compute_entropy:
logger.info('Attention entropies' )
print_ad_tensor(A )
if compute_importance:
logger.info('Head importance scores' )
print_ad_tensor(A )
logger.info('Head ranked by importance scores' )
snake_case__ : Tuple = torch.zeros(head_importance.numel() , dtype=torch.long , device=args.device )
snake_case__ : Union[str, Any] = torch.arange(
head_importance.numel() , device=args.device )
snake_case__ : str = head_ranks.view_as(A )
print_ad_tensor(A )
return attn_entropy, head_importance, total_loss
def lowercase_ (A : Optional[int] , A : Dict , A : Optional[int] ):
snake_case__ , snake_case__ , snake_case__ : Any = compute_heads_importance(A , A , A , compute_entropy=A )
snake_case__ : Tuple = 1 / loss # instead of downsteam score use the LM loss
logger.info('Pruning: original score: %f, threshold: %f' , A , original_score * args.masking_threshold )
snake_case__ : Optional[Any] = torch.ones_like(A )
snake_case__ : Union[str, Any] = max(1 , int(new_head_mask.numel() * args.masking_amount ) )
snake_case__ : Dict = original_score
while current_score >= original_score * args.masking_threshold:
snake_case__ : int = new_head_mask.clone().detach() # save current head mask
# heads from least important to most - keep only not-masked heads
snake_case__ : List[Any] = float('Inf' )
snake_case__ : Union[str, Any] = head_importance.view(-1 ).sort()[1]
if len(A ) <= num_to_mask:
print('BREAK BY num_to_mask' )
break
# mask heads
snake_case__ : int = current_heads_to_mask[:num_to_mask]
logger.info('Heads to mask: %s' , str(current_heads_to_mask.tolist() ) )
snake_case__ : int = new_head_mask.view(-1 )
snake_case__ : int = 0.0
snake_case__ : Union[str, Any] = new_head_mask.view_as(A )
snake_case__ : List[str] = new_head_mask.clone().detach()
print_ad_tensor(A )
# Compute metric and head importance again
snake_case__ , snake_case__ , snake_case__ : Any = compute_heads_importance(
A , A , A , compute_entropy=A , head_mask=A )
snake_case__ : Dict = 1 / loss
logger.info(
'Masking: current score: %f, remaining heads %d (%.1f percents)' , A , new_head_mask.sum() , new_head_mask.sum() / new_head_mask.numel() * 1_0_0 , )
logger.info('Final head mask' )
print_ad_tensor(A )
np.save(os.path.join(args.output_dir , 'head_mask.npy' ) , head_mask.detach().cpu().numpy() )
return head_mask
def lowercase_ (A : List[str] , A : Tuple , A : Optional[Any] , A : int ):
snake_case__ : Any = datetime.now()
snake_case__ , snake_case__ , snake_case__ : str = compute_heads_importance(
A , A , A , compute_entropy=A , compute_importance=A , head_mask=A )
snake_case__ : Tuple = 1 / loss
snake_case__ : Dict = datetime.now() - before_time
snake_case__ : Union[str, Any] = sum(p.numel() for p in model.parameters() )
snake_case__ : Optional[Any] = {
layer: (1 - head_mask[layer].long()).nonzero().squeeze().tolist() for layer in range(len(A ) )
}
for k, v in heads_to_prune.items():
if isinstance(A , A ):
snake_case__ : Any = [
v,
]
assert sum(len(A ) for h in heads_to_prune.values() ) == (1 - head_mask.long()).sum().item()
model.prune_heads(A )
snake_case__ : Dict = sum(p.numel() for p in model.parameters() )
snake_case__ : Tuple = datetime.now()
snake_case__ , snake_case__ , snake_case__ : Dict = compute_heads_importance(
A , A , A , compute_entropy=A , compute_importance=A , head_mask=A , actually_pruned=A , )
snake_case__ : Any = 1 / loss
snake_case__ : int = datetime.now() - before_time
logger.info(
'Pruning: original num of params: %.2e, after pruning %.2e (%.1f percents)' , A , A , pruned_num_params / original_num_params * 1_0_0 , )
logger.info('Pruning: score with masking: %f score with pruning: %f' , A , A )
logger.info('Pruning: speed ratio (original timing / new timing): %f percents' , original_time / new_time * 1_0_0 )
save_model(A , args.output_dir )
def lowercase_ ():
snake_case__ : str = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
'--data_dir' , default=A , type=A , required=A , help='The input data dir. Should contain the .tsv files (or other data files) for the task.' , )
parser.add_argument(
'--model_name_or_path' , default=A , type=A , required=A , help='Path to pretrained model or model identifier from huggingface.co/models' , )
parser.add_argument(
'--output_dir' , default=A , type=A , required=A , help='The output directory where the model predictions and checkpoints will be written.' , )
# Other parameters
parser.add_argument(
'--config_name' , default='' , type=A , help='Pretrained config name or path if not the same as model_name_or_path' , )
parser.add_argument(
'--tokenizer_name' , default='' , type=A , help='Pretrained tokenizer name or path if not the same as model_name_or_path' , )
parser.add_argument(
'--cache_dir' , default=A , type=A , help='Where do you want to store the pre-trained models downloaded from s3' , )
parser.add_argument(
'--data_subset' , type=A , default=-1 , help='If > 0: limit the data to a subset of data_subset instances.' )
parser.add_argument(
'--overwrite_output_dir' , action='store_true' , help='Whether to overwrite data in output directory' )
parser.add_argument(
'--overwrite_cache' , action='store_true' , help='Overwrite the cached training and evaluation sets' )
parser.add_argument(
'--dont_normalize_importance_by_layer' , action='store_true' , help='Don\'t normalize importance score by layers' )
parser.add_argument(
'--dont_normalize_global_importance' , action='store_true' , help='Don\'t normalize all importance scores between 0 and 1' , )
parser.add_argument(
'--try_masking' , action='store_true' , help='Whether to try to mask head until a threshold of accuracy.' )
parser.add_argument(
'--masking_threshold' , default=0.9 , type=A , help='masking threshold in term of metrics (stop masking when metric < threshold * original metric value).' , )
parser.add_argument(
'--masking_amount' , default=0.1 , type=A , help='Amount to heads to masking at each masking step.' )
parser.add_argument('--metric_name' , default='acc' , type=A , help='Metric to use for head masking.' )
parser.add_argument(
'--max_seq_length' , default=1_2_8 , type=A , help=(
'The maximum total input sequence length after WordPiece tokenization. \n'
'Sequences longer than this will be truncated, sequences shorter padded.'
) , )
parser.add_argument('--batch_size' , default=1 , type=A , help='Batch size.' )
parser.add_argument('--seed' , type=A , default=4_2 )
parser.add_argument('--local_rank' , type=A , default=-1 , help='local_rank for distributed training on gpus' )
parser.add_argument('--no_cuda' , action='store_true' , help='Whether not to use CUDA when available' )
parser.add_argument('--server_ip' , type=A , default='' , help='Can be used for distant debugging.' )
parser.add_argument('--server_port' , type=A , default='' , help='Can be used for distant debugging.' )
snake_case__ : Optional[int] = parser.parse_args()
if args.server_ip and args.server_port:
# Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script
import ptvsd
print('Waiting for debugger attach' )
ptvsd.enable_attach(address=(args.server_ip, args.server_port) , redirect_output=A )
ptvsd.wait_for_attach()
# Setup devices and distributed training
if args.local_rank == -1 or args.no_cuda:
snake_case__ : List[Any] = torch.device('cuda' if torch.cuda.is_available() and not args.no_cuda else 'cpu' )
snake_case__ : Optional[Any] = 0 if args.no_cuda else torch.cuda.device_count()
else:
torch.cuda.set_device(args.local_rank )
snake_case__ : int = torch.device('cuda' , args.local_rank )
snake_case__ : List[str] = 1
torch.distributed.init_process_group(backend='nccl' ) # Initializes the distributed backend
# Setup logging
logging.basicConfig(level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN )
logger.info('device: {} n_gpu: {}, distributed: {}'.format(args.device , args.n_gpu , bool(args.local_rank != -1 ) ) )
snake_case__ : Any = GPTaLMHeadModel.from_pretrained(args.model_name_or_path )
# Distributed and parallel training
model.to(args.device )
if args.local_rank != -1:
snake_case__ : List[str] = nn.parallel.DistributedDataParallel(
A , device_ids=[args.local_rank] , output_device=args.local_rank , find_unused_parameters=A )
elif args.n_gpu > 1:
snake_case__ : Optional[int] = nn.DataParallel(A )
# Print/save training arguments
os.makedirs(args.output_dir , exist_ok=A )
torch.save(A , os.path.join(args.output_dir , 'run_args.bin' ) )
logger.info('Training/evaluation parameters %s' , A )
# Prepare dataset
snake_case__ : Optional[Any] = np.concatenate(
[
np.loadtxt(args.data_dir , dtype=np.intaa ),
] )
snake_case__ : List[str] = (torch.from_numpy(A ),)
snake_case__ : int = TensorDataset(*A )
snake_case__ : Union[str, Any] = RandomSampler(A )
snake_case__ : Any = DataLoader(A , sampler=A , batch_size=args.batch_size )
# Compute head entropy and importance score
compute_heads_importance(A , A , A )
# Try head masking (set heads to zero until the score goes under a threshole)
# and head pruning (remove masked heads and see the effect on the network)
if args.try_masking and args.masking_threshold > 0.0 and args.masking_threshold < 1.0:
snake_case__ : Dict = mask_heads(A , A , A )
prune_heads(A , A , A , A )
if __name__ == "__main__":
main()
| 277 | 0 |
'''simple docstring'''
import socket
def __lowerCAmelCase ( ):
__UpperCamelCase : Optional[Any] = socket.socket(socket.AF_INET , socket.SOCK_STREAM )
__UpperCamelCase : Dict = socket.gethostname()
__UpperCamelCase : List[str] = 12_312
sock.connect((host, port) )
sock.send(b"Hello server!" )
with open("Received_file" , "wb" ) as out_file:
print("File opened" )
print("Receiving data..." )
while True:
__UpperCamelCase : Tuple = sock.recv(1_024 )
if not data:
break
out_file.write(snake_case__ )
print("Successfully received the file" )
sock.close()
print("Connection closed" )
if __name__ == "__main__":
main()
| 298 |
import argparse
import json
from collections import OrderedDict
from pathlib import Path
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import (
SegformerConfig,
SegformerForImageClassification,
SegformerForSemanticSegmentation,
SegformerImageProcessor,
)
from transformers.utils import logging
logging.set_verbosity_info()
a_ :Dict = logging.get_logger(__name__)
def lowercase_ (A : Optional[Any] , A : Any=False ):
snake_case__ : List[Any] = OrderedDict()
for key, value in state_dict.items():
if encoder_only and not key.startswith('head' ):
snake_case__ : str = 'segformer.encoder.' + key
if key.startswith('backbone' ):
snake_case__ : str = key.replace('backbone' , 'segformer.encoder' )
if "patch_embed" in key:
# replace for example patch_embed1 by patch_embeddings.0
snake_case__ : Optional[int] = key[key.find('patch_embed' ) + len('patch_embed' )]
snake_case__ : int = key.replace(F'''patch_embed{idx}''' , F'''patch_embeddings.{int(A )-1}''' )
if "norm" in key:
snake_case__ : Optional[int] = key.replace('norm' , 'layer_norm' )
if "segformer.encoder.layer_norm" in key:
# replace for example layer_norm1 by layer_norm.0
snake_case__ : Tuple = key[key.find('segformer.encoder.layer_norm' ) + len('segformer.encoder.layer_norm' )]
snake_case__ : Union[str, Any] = key.replace(F'''layer_norm{idx}''' , F'''layer_norm.{int(A )-1}''' )
if "layer_norm1" in key:
snake_case__ : List[Any] = key.replace('layer_norm1' , 'layer_norm_1' )
if "layer_norm2" in key:
snake_case__ : List[Any] = key.replace('layer_norm2' , 'layer_norm_2' )
if "block" in key:
# replace for example block1 by block.0
snake_case__ : List[Any] = key[key.find('block' ) + len('block' )]
snake_case__ : List[Any] = key.replace(F'''block{idx}''' , F'''block.{int(A )-1}''' )
if "attn.q" in key:
snake_case__ : int = key.replace('attn.q' , 'attention.self.query' )
if "attn.proj" in key:
snake_case__ : str = key.replace('attn.proj' , 'attention.output.dense' )
if "attn" in key:
snake_case__ : Optional[int] = key.replace('attn' , 'attention.self' )
if "fc1" in key:
snake_case__ : str = key.replace('fc1' , 'dense1' )
if "fc2" in key:
snake_case__ : Dict = key.replace('fc2' , 'dense2' )
if "linear_pred" in key:
snake_case__ : Union[str, Any] = key.replace('linear_pred' , 'classifier' )
if "linear_fuse" in key:
snake_case__ : List[str] = key.replace('linear_fuse.conv' , 'linear_fuse' )
snake_case__ : List[Any] = key.replace('linear_fuse.bn' , 'batch_norm' )
if "linear_c" in key:
# replace for example linear_c4 by linear_c.3
snake_case__ : Optional[int] = key[key.find('linear_c' ) + len('linear_c' )]
snake_case__ : Tuple = key.replace(F'''linear_c{idx}''' , F'''linear_c.{int(A )-1}''' )
if key.startswith('head' ):
snake_case__ : Tuple = key.replace('head' , 'classifier' )
snake_case__ : Optional[int] = value
return new_state_dict
def lowercase_ (A : Tuple , A : Optional[int] ):
# for each of the encoder blocks:
for i in range(config.num_encoder_blocks ):
for j in range(config.depths[i] ):
# read in weights + bias of keys and values (which is a single matrix in the original implementation)
snake_case__ : List[str] = state_dict.pop(F'''segformer.encoder.block.{i}.{j}.attention.self.kv.weight''' )
snake_case__ : Optional[Any] = state_dict.pop(F'''segformer.encoder.block.{i}.{j}.attention.self.kv.bias''' )
# next, add keys and values (in that order) to the state dict
snake_case__ : str = kv_weight[
: config.hidden_sizes[i], :
]
snake_case__ : Dict = kv_bias[: config.hidden_sizes[i]]
snake_case__ : List[str] = kv_weight[
config.hidden_sizes[i] :, :
]
snake_case__ : List[Any] = kv_bias[
config.hidden_sizes[i] :
]
def lowercase_ ():
snake_case__ : Union[str, Any] = 'http://images.cocodataset.org/val2017/000000039769.jpg'
snake_case__ : Dict = Image.open(requests.get(A , stream=A ).raw )
return image
@torch.no_grad()
def lowercase_ (A : Any , A : Union[str, Any] , A : Optional[Any] ):
snake_case__ : List[str] = SegformerConfig()
snake_case__ : Dict = False
# set attributes based on model_name
snake_case__ : Optional[int] = 'huggingface/label-files'
if "segformer" in model_name:
snake_case__ : str = model_name[len('segformer.' ) : len('segformer.' ) + 2]
if "ade" in model_name:
snake_case__ : Optional[int] = 1_5_0
snake_case__ : int = 'ade20k-id2label.json'
snake_case__ : List[Any] = (1, 1_5_0, 1_2_8, 1_2_8)
elif "city" in model_name:
snake_case__ : str = 1_9
snake_case__ : List[str] = 'cityscapes-id2label.json'
snake_case__ : Optional[Any] = (1, 1_9, 1_2_8, 1_2_8)
else:
raise ValueError(F'''Model {model_name} not supported''' )
elif "mit" in model_name:
snake_case__ : str = True
snake_case__ : Union[str, Any] = model_name[4:6]
snake_case__ : Optional[Any] = 1_0_0_0
snake_case__ : Optional[int] = 'imagenet-1k-id2label.json'
snake_case__ : List[Any] = (1, 1_0_0_0)
else:
raise ValueError(F'''Model {model_name} not supported''' )
# set config attributes
snake_case__ : str = json.load(open(hf_hub_download(A , A , repo_type='dataset' ) , 'r' ) )
snake_case__ : List[Any] = {int(A ): v for k, v in idalabel.items()}
snake_case__ : Union[str, Any] = idalabel
snake_case__ : Tuple = {v: k for k, v in idalabel.items()}
if size == "b0":
pass
elif size == "b1":
snake_case__ : List[Any] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : Tuple = 2_5_6
elif size == "b2":
snake_case__ : List[str] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : int = 7_6_8
snake_case__ : List[Any] = [3, 4, 6, 3]
elif size == "b3":
snake_case__ : Optional[Any] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : int = 7_6_8
snake_case__ : Optional[Any] = [3, 4, 1_8, 3]
elif size == "b4":
snake_case__ : str = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : Optional[Any] = 7_6_8
snake_case__ : Union[str, Any] = [3, 8, 2_7, 3]
elif size == "b5":
snake_case__ : List[str] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : Optional[Any] = 7_6_8
snake_case__ : Any = [3, 6, 4_0, 3]
else:
raise ValueError(F'''Size {size} not supported''' )
# load image processor (only resize + normalize)
snake_case__ : Dict = SegformerImageProcessor(
image_scale=(5_1_2, 5_1_2) , keep_ratio=A , align=A , do_random_crop=A )
# prepare image
snake_case__ : List[str] = prepare_img()
snake_case__ : Dict = image_processor(images=A , return_tensors='pt' ).pixel_values
logger.info(F'''Converting model {model_name}...''' )
# load original state dict
if encoder_only:
snake_case__ : Tuple = torch.load(A , map_location=torch.device('cpu' ) )
else:
snake_case__ : int = torch.load(A , map_location=torch.device('cpu' ) )['state_dict']
# rename keys
snake_case__ : List[Any] = rename_keys(A , encoder_only=A )
if not encoder_only:
del state_dict["decode_head.conv_seg.weight"]
del state_dict["decode_head.conv_seg.bias"]
# key and value matrices need special treatment
read_in_k_v(A , A )
# create HuggingFace model and load state dict
if encoder_only:
snake_case__ : str = False
snake_case__ : List[Any] = SegformerForImageClassification(A )
else:
snake_case__ : Dict = SegformerForSemanticSegmentation(A )
model.load_state_dict(A )
model.eval()
# forward pass
snake_case__ : int = model(A )
snake_case__ : Any = outputs.logits
# set expected_slice based on model name
# ADE20k checkpoints
if model_name == "segformer.b0.512x512.ade.160k":
snake_case__ : Dict = torch.tensor(
[
[[-4.6310, -5.5232, -6.2356], [-5.1921, -6.1444, -6.5996], [-5.4424, -6.2790, -6.7574]],
[[-12.1391, -13.3122, -13.9554], [-12.8732, -13.9352, -14.3563], [-12.9438, -13.8226, -14.2513]],
[[-12.5134, -13.4686, -14.4915], [-12.8669, -14.4343, -14.7758], [-13.2523, -14.5819, -15.0694]],
] )
elif model_name == "segformer.b1.512x512.ade.160k":
snake_case__ : Optional[int] = torch.tensor(
[
[[-7.5820, -8.7231, -8.3215], [-8.0600, -10.3529, -10.0304], [-7.5208, -9.4103, -9.6239]],
[[-12.6918, -13.8994, -13.7137], [-13.3196, -15.7523, -15.4789], [-12.9343, -14.8757, -14.9689]],
[[-11.1911, -11.9421, -11.3243], [-11.3342, -13.6839, -13.3581], [-10.3909, -12.1832, -12.4858]],
] )
elif model_name == "segformer.b2.512x512.ade.160k":
snake_case__ : List[Any] = torch.tensor(
[
[[-11.8173, -14.3850, -16.3128], [-14.5648, -16.5804, -18.6568], [-14.7223, -15.7387, -18.4218]],
[[-15.7290, -17.9171, -19.4423], [-18.3105, -19.9448, -21.4661], [-17.9296, -18.6497, -20.7910]],
[[-15.0783, -17.0336, -18.2789], [-16.8771, -18.6870, -20.1612], [-16.2454, -17.1426, -19.5055]],
] )
elif model_name == "segformer.b3.512x512.ade.160k":
snake_case__ : Union[str, Any] = torch.tensor(
[
[[-9.0878, -10.2081, -10.1891], [-9.3144, -10.7941, -10.9843], [-9.2294, -10.3855, -10.5704]],
[[-12.2316, -13.9068, -13.6102], [-12.9161, -14.3702, -14.3235], [-12.5233, -13.7174, -13.7932]],
[[-14.6275, -15.2490, -14.9727], [-14.3400, -15.9687, -16.2827], [-14.1484, -15.4033, -15.8937]],
] )
elif model_name == "segformer.b4.512x512.ade.160k":
snake_case__ : Dict = torch.tensor(
[
[[-12.3144, -13.2447, -14.0802], [-13.3614, -14.5816, -15.6117], [-13.3340, -14.4433, -16.2219]],
[[-19.2781, -20.4128, -20.7506], [-20.6153, -21.6566, -22.0998], [-19.9800, -21.0430, -22.1494]],
[[-18.8739, -19.7804, -21.1834], [-20.1233, -21.6765, -23.2944], [-20.0315, -21.2641, -23.6944]],
] )
elif model_name == "segformer.b5.640x640.ade.160k":
snake_case__ : List[Any] = torch.tensor(
[
[[-9.5524, -12.0835, -11.7348], [-10.5229, -13.6446, -14.5662], [-9.5842, -12.8851, -13.9414]],
[[-15.3432, -17.5323, -17.0818], [-16.3330, -18.9255, -19.2101], [-15.1340, -17.7848, -18.3971]],
[[-12.6072, -14.9486, -14.6631], [-13.7629, -17.0907, -17.7745], [-12.7899, -16.1695, -17.1671]],
] )
# Cityscapes checkpoints
elif model_name == "segformer.b0.1024x1024.city.160k":
snake_case__ : str = torch.tensor(
[
[[-11.9295, -13.4057, -14.8106], [-13.3431, -14.8179, -15.3781], [-14.2836, -15.5942, -16.1588]],
[[-11.4906, -12.8067, -13.6564], [-13.1189, -14.0500, -14.1543], [-13.8748, -14.5136, -14.8789]],
[[0.5374, 0.1067, -0.4742], [0.1141, -0.2255, -0.7099], [-0.3000, -0.5924, -1.3105]],
] )
elif model_name == "segformer.b0.512x1024.city.160k":
snake_case__ : Tuple = torch.tensor(
[
[[-7.8217, -9.8767, -10.1717], [-9.4438, -10.9058, -11.4047], [-9.7939, -12.3495, -12.1079]],
[[-7.1514, -9.5336, -10.0860], [-9.7776, -11.6822, -11.8439], [-10.1411, -12.7655, -12.8972]],
[[0.3021, 0.0805, -0.2310], [-0.0328, -0.1605, -0.2714], [-0.1408, -0.5477, -0.6976]],
] )
elif model_name == "segformer.b0.640x1280.city.160k":
snake_case__ : Any = torch.tensor(
[
[
[-1.1_372e01, -1.2_787e01, -1.3_477e01],
[-1.2_536e01, -1.4_194e01, -1.4_409e01],
[-1.3_217e01, -1.4_888e01, -1.5_327e01],
],
[
[-1.4_791e01, -1.7_122e01, -1.8_277e01],
[-1.7_163e01, -1.9_192e01, -1.9_533e01],
[-1.7_897e01, -1.9_991e01, -2.0_315e01],
],
[
[7.6_723e-01, 4.1_921e-01, -7.7_878e-02],
[4.7_772e-01, 9.5_557e-03, -2.8_082e-01],
[3.6_032e-01, -2.4_826e-01, -5.1_168e-01],
],
] )
elif model_name == "segformer.b0.768x768.city.160k":
snake_case__ : Optional[int] = torch.tensor(
[
[[-9.4959, -11.3087, -11.7479], [-11.0025, -12.6540, -12.3319], [-11.4064, -13.0487, -12.9905]],
[[-9.8905, -11.3084, -12.0854], [-11.1726, -12.7698, -12.9583], [-11.5985, -13.3278, -14.1774]],
[[0.2213, 0.0192, -0.2466], [-0.1731, -0.4213, -0.4874], [-0.3126, -0.6541, -1.1389]],
] )
elif model_name == "segformer.b1.1024x1024.city.160k":
snake_case__ : Union[str, Any] = torch.tensor(
[
[[-13.5748, -13.9111, -12.6500], [-14.3500, -15.3683, -14.2328], [-14.7532, -16.0424, -15.6087]],
[[-17.1651, -15.8725, -12.9653], [-17.2580, -17.3718, -14.8223], [-16.6058, -16.8783, -16.7452]],
[[-3.6456, -3.0209, -1.4203], [-3.0797, -3.1959, -2.0000], [-1.8757, -1.9217, -1.6997]],
] )
elif model_name == "segformer.b2.1024x1024.city.160k":
snake_case__ : List[str] = torch.tensor(
[
[[-16.0976, -16.4856, -17.3962], [-16.6234, -19.0342, -19.7685], [-16.0900, -18.0661, -19.1180]],
[[-18.4750, -18.8488, -19.5074], [-19.4030, -22.1570, -22.5977], [-19.1191, -20.8486, -22.3783]],
[[-4.5178, -5.5037, -6.5109], [-5.0884, -7.2174, -8.0334], [-4.4156, -5.8117, -7.2970]],
] )
elif model_name == "segformer.b3.1024x1024.city.160k":
snake_case__ : List[Any] = torch.tensor(
[
[[-14.2081, -14.4732, -14.1977], [-14.5867, -16.4423, -16.6356], [-13.4441, -14.9685, -16.8696]],
[[-14.4576, -14.7073, -15.0451], [-15.0816, -17.6237, -17.9873], [-14.4213, -16.0199, -18.5992]],
[[-4.7349, -4.9588, -5.0966], [-4.3210, -6.9325, -7.2591], [-3.4312, -4.7484, -7.1917]],
] )
elif model_name == "segformer.b4.1024x1024.city.160k":
snake_case__ : str = torch.tensor(
[
[[-11.7737, -11.9526, -11.3273], [-13.6692, -14.4574, -13.8878], [-13.8937, -14.6924, -15.9345]],
[[-14.6706, -14.5330, -14.1306], [-16.1502, -16.8180, -16.4269], [-16.8338, -17.8939, -20.1746]],
[[1.0491, 0.8289, 1.0310], [1.1044, 0.5219, 0.8055], [1.0899, 0.6926, 0.5590]],
] )
elif model_name == "segformer.b5.1024x1024.city.160k":
snake_case__ : List[str] = torch.tensor(
[
[[-12.5641, -13.4777, -13.0684], [-13.9587, -15.8983, -16.6557], [-13.3109, -15.7350, -16.3141]],
[[-14.7074, -15.4352, -14.5944], [-16.6353, -18.1663, -18.6120], [-15.1702, -18.0329, -18.1547]],
[[-1.7990, -2.0951, -1.7784], [-2.6397, -3.8245, -3.9686], [-1.5264, -2.8126, -2.9316]],
] )
else:
snake_case__ : Tuple = logits.argmax(-1 ).item()
print('Predicted class:' , model.config.idalabel[predicted_class_idx] )
# verify logits
if not encoder_only:
assert logits.shape == expected_shape
assert torch.allclose(logits[0, :3, :3, :3] , A , atol=1e-2 )
# finally, save model and image processor
logger.info(F'''Saving PyTorch model and image processor to {pytorch_dump_folder_path}...''' )
Path(A ).mkdir(exist_ok=A )
model.save_pretrained(A )
image_processor.save_pretrained(A )
if __name__ == "__main__":
a_ :Optional[int] = argparse.ArgumentParser()
parser.add_argument(
"--model_name",
default="segformer.b0.512x512.ade.160k",
type=str,
help="Name of the model you'd like to convert.",
)
parser.add_argument(
"--checkpoint_path", default=None, type=str, help="Path to the original PyTorch checkpoint (.pth file)."
)
parser.add_argument(
"--pytorch_dump_folder_path", default=None, type=str, help="Path to the folder to output PyTorch model."
)
a_ :Union[str, Any] = parser.parse_args()
convert_segformer_checkpoint(args.model_name, args.checkpoint_path, args.pytorch_dump_folder_path)
| 277 | 0 |
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_sentencepiece_available,
is_tokenizers_available,
is_torch_available,
)
_snake_case = {"configuration_reformer": ["REFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP", "ReformerConfig"]}
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
_snake_case = ["ReformerTokenizer"]
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
_snake_case = ["ReformerTokenizerFast"]
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
_snake_case = [
"REFORMER_PRETRAINED_MODEL_ARCHIVE_LIST",
"ReformerAttention",
"ReformerForMaskedLM",
"ReformerForQuestionAnswering",
"ReformerForSequenceClassification",
"ReformerLayer",
"ReformerModel",
"ReformerModelWithLMHead",
"ReformerPreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_reformer import REFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, ReformerConfig
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_reformer import ReformerTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_reformer_fast import ReformerTokenizerFast
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_reformer import (
REFORMER_PRETRAINED_MODEL_ARCHIVE_LIST,
ReformerAttention,
ReformerForMaskedLM,
ReformerForQuestionAnswering,
ReformerForSequenceClassification,
ReformerLayer,
ReformerModel,
ReformerModelWithLMHead,
ReformerPreTrainedModel,
)
else:
import sys
_snake_case = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)
| 26 |
import argparse
import json
import os
import fairseq
import torch
from fairseq.data import Dictionary
from transformers import (
WavaVecaConfig,
WavaVecaCTCTokenizer,
WavaVecaFeatureExtractor,
WavaVecaForCTC,
WavaVecaForPreTraining,
WavaVecaProcessor,
logging,
)
from transformers.models.wavaveca.modeling_wavaveca import WavaVecaForSequenceClassification
logging.set_verbosity_info()
a_ :List[Any] = logging.get_logger(__name__)
a_ :List[Any] = {
"post_extract_proj": "feature_projection.projection",
"encoder.pos_conv.0": "encoder.pos_conv_embed.conv",
"self_attn.k_proj": "encoder.layers.*.attention.k_proj",
"self_attn.v_proj": "encoder.layers.*.attention.v_proj",
"self_attn.q_proj": "encoder.layers.*.attention.q_proj",
"self_attn.out_proj": "encoder.layers.*.attention.out_proj",
"self_attn_layer_norm": "encoder.layers.*.layer_norm",
"fc1": "encoder.layers.*.feed_forward.intermediate_dense",
"fc2": "encoder.layers.*.feed_forward.output_dense",
"final_layer_norm": "encoder.layers.*.final_layer_norm",
"encoder.layer_norm": "encoder.layer_norm",
"adapter_layer": "encoder.layers.*.adapter_layer",
"w2v_model.layer_norm": "feature_projection.layer_norm",
"quantizer.weight_proj": "quantizer.weight_proj",
"quantizer.vars": "quantizer.codevectors",
"project_q": "project_q",
"final_proj": "project_hid",
"w2v_encoder.proj": "lm_head",
"mask_emb": "masked_spec_embed",
"pooling_layer.linear": "projector",
"pooling_layer.projection": "classifier",
}
a_ :List[Any] = [
"lm_head",
"quantizer.weight_proj",
"quantizer.codevectors",
"project_q",
"project_hid",
"projector",
"classifier",
]
def lowercase_ (A : Dict ):
snake_case__ : Optional[Any] = {}
with open(A , 'r' ) as file:
for line_number, line in enumerate(A ):
snake_case__ : Dict = line.strip()
if line:
snake_case__ : int = line.split()
snake_case__ : List[str] = line_number
snake_case__ : Dict = words[0]
snake_case__ : Optional[Any] = value
return result
def lowercase_ (A : int , A : int , A : Optional[int] , A : Optional[Any] , A : Tuple ):
for attribute in key.split('.' ):
snake_case__ : Optional[int] = getattr(A , A )
snake_case__ : Union[str, Any] = None
for param_key in PARAM_MAPPING.keys():
if full_name.endswith(A ):
snake_case__ : List[str] = PARAM_MAPPING[full_name.split('.' )[-1]]
snake_case__ : Dict = 'param'
if weight_type is not None and weight_type != "param":
snake_case__ : Union[str, Any] = getattr(A , A ).shape
elif weight_type is not None and weight_type == "param":
snake_case__ : Optional[int] = hf_pointer
for attribute in hf_param_name.split('.' ):
snake_case__ : Optional[Any] = getattr(A , A )
snake_case__ : Dict = shape_pointer.shape
# let's reduce dimension
snake_case__ : List[Any] = value[0]
else:
snake_case__ : Union[str, Any] = hf_pointer.shape
if hf_shape != value.shape:
raise ValueError(
F'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
F''' {value.shape} for {full_name}''' )
if weight_type == "weight":
snake_case__ : Any = value
elif weight_type == "weight_g":
snake_case__ : List[Any] = value
elif weight_type == "weight_v":
snake_case__ : Any = value
elif weight_type == "bias":
snake_case__ : List[Any] = value
elif weight_type == "param":
for attribute in hf_param_name.split('.' ):
snake_case__ : int = getattr(A , A )
snake_case__ : Optional[int] = value
else:
snake_case__ : Optional[Any] = value
logger.info(F'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def lowercase_ (A : Tuple , A : List[Any] , A : int , A : str , A : Tuple ):
snake_case__ : Optional[int] = None
for param_key in PARAM_MAPPING.keys():
if full_name.endswith(A ):
snake_case__ : List[str] = PARAM_MAPPING[full_name.split('.' )[-1]]
snake_case__ : str = 'param'
if weight_type is not None and weight_type != "param":
snake_case__ : int = '.'.join([key, weight_type] )
elif weight_type is not None and weight_type == "param":
snake_case__ : Any = '.'.join([key, hf_param_name] )
else:
snake_case__ : Dict = key
snake_case__ : List[str] = value if 'lm_head' in full_key else value[0]
a_ :List[str] = {
"W_a": "linear_1.weight",
"W_b": "linear_2.weight",
"b_a": "linear_1.bias",
"b_b": "linear_2.bias",
"ln_W": "norm.weight",
"ln_b": "norm.bias",
}
def lowercase_ (A : str , A : Optional[Any] , A : Optional[Any]=None , A : List[str]=None ):
snake_case__ : Optional[int] = False
for key, mapped_key in MAPPING.items():
snake_case__ : Tuple = 'wav2vec2.' + mapped_key if mapped_key not in TOP_LEVEL_KEYS else mapped_key
if key in name or key.split('w2v_model.' )[-1] == name.split('.' )[0]:
snake_case__ : Optional[int] = True
if "*" in mapped_key:
snake_case__ : List[Any] = name.split(A )[0].split('.' )[-2]
snake_case__ : Union[str, Any] = mapped_key.replace('*' , A )
if "weight_g" in name:
snake_case__ : Tuple = 'weight_g'
elif "weight_v" in name:
snake_case__ : List[str] = 'weight_v'
elif "bias" in name:
snake_case__ : Dict = 'bias'
elif "weight" in name:
# TODO: don't match quantizer.weight_proj
snake_case__ : Optional[int] = 'weight'
else:
snake_case__ : str = None
if hf_dict is not None:
rename_dict(A , A , A , A , A )
else:
set_recursively(A , A , A , A , A )
return is_used
return is_used
def lowercase_ (A : Optional[Any] , A : Dict , A : Optional[int] ):
snake_case__ : Dict = []
snake_case__ : Tuple = fairseq_model.state_dict()
snake_case__ : str = hf_model.wavaveca.feature_extractor
for name, value in fairseq_dict.items():
snake_case__ : str = False
if "conv_layers" in name:
load_conv_layer(
A , A , A , A , hf_model.config.feat_extract_norm == 'group' , )
snake_case__ : Any = True
else:
snake_case__ : Dict = load_wavaveca_layer(A , A , A )
if not is_used:
unused_weights.append(A )
logger.warning(F'''Unused weights: {unused_weights}''' )
def lowercase_ (A : Dict , A : Optional[Any] , A : Tuple , A : str , A : List[str] ):
snake_case__ : List[Any] = full_name.split('conv_layers.' )[-1]
snake_case__ : List[str] = name.split('.' )
snake_case__ : List[Any] = int(items[0] )
snake_case__ : str = int(items[1] )
if type_id == 0:
if "bias" in name:
if value.shape != feature_extractor.conv_layers[layer_id].conv.bias.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.''' )
snake_case__ : Any = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
if value.shape != feature_extractor.conv_layers[layer_id].conv.weight.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.''' )
snake_case__ : str = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
if value.shape != feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape} was found.''' )
snake_case__ : str = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
if value.shape != feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape} was found.''' )
snake_case__ : int = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(A )
@torch.no_grad()
def lowercase_ (A : Union[str, Any] , A : str , A : Tuple=None , A : List[str]=None , A : Any=True , A : Optional[int]=False ):
if config_path is not None:
snake_case__ : List[Any] = WavaVecaConfig.from_pretrained(A )
else:
snake_case__ : List[Any] = WavaVecaConfig()
if is_seq_class:
snake_case__ : Dict = read_txt_into_dict(A )
snake_case__ : Any = idalabel
snake_case__ : Union[str, Any] = WavaVecaForSequenceClassification(A )
snake_case__ : Any = WavaVecaFeatureExtractor(
feature_size=1 , sampling_rate=1_6_0_0_0 , padding_value=0 , do_normalize=A , return_attention_mask=A , )
feature_extractor.save_pretrained(A )
elif is_finetuned:
if dict_path:
snake_case__ : str = Dictionary.load(A )
# important change bos & pad token id since CTC symbol is <pad> and
# not <s> as in fairseq
snake_case__ : List[str] = target_dict.pad_index
snake_case__ : Optional[int] = target_dict.bos_index
snake_case__ : Optional[int] = target_dict.eos_index
snake_case__ : List[Any] = len(target_dict.symbols )
snake_case__ : str = os.path.join(A , 'vocab.json' )
if not os.path.isdir(A ):
logger.error('--pytorch_dump_folder_path ({}) should be a directory'.format(A ) )
return
os.makedirs(A , exist_ok=A )
snake_case__ : Optional[Any] = target_dict.indices
# fairseq has the <pad> and <s> switched
snake_case__ : Optional[Any] = 0
snake_case__ : Union[str, Any] = 1
with open(A , 'w' , encoding='utf-8' ) as vocab_handle:
json.dump(A , A )
snake_case__ : List[Any] = WavaVecaCTCTokenizer(
A , unk_token=target_dict.unk_word , pad_token=target_dict.pad_word , bos_token=target_dict.bos_word , eos_token=target_dict.eos_word , word_delimiter_token='|' , do_lower_case=A , )
snake_case__ : str = True if config.feat_extract_norm == 'layer' else False
snake_case__ : Optional[Any] = WavaVecaFeatureExtractor(
feature_size=1 , sampling_rate=1_6_0_0_0 , padding_value=0 , do_normalize=A , return_attention_mask=A , )
snake_case__ : Union[str, Any] = WavaVecaProcessor(feature_extractor=A , tokenizer=A )
processor.save_pretrained(A )
snake_case__ : str = WavaVecaForCTC(A )
else:
snake_case__ : int = WavaVecaForPreTraining(A )
if is_finetuned or is_seq_class:
snake_case__ , snake_case__ , snake_case__ : str = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] , arg_overrides={'data': '/'.join(dict_path.split('/' )[:-1] )} )
else:
snake_case__ : Tuple = argparse.Namespace(task='audio_pretraining' )
snake_case__ : str = fairseq.tasks.setup_task(A )
snake_case__ , snake_case__ , snake_case__ : Any = fairseq.checkpoint_utils.load_model_ensemble_and_task([checkpoint_path] , task=A )
snake_case__ : List[Any] = model[0].eval()
recursively_load_weights(A , A , not is_finetuned )
hf_wavavec.save_pretrained(A )
if __name__ == "__main__":
a_ :List[Any] = argparse.ArgumentParser()
parser.add_argument("--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model.")
parser.add_argument("--checkpoint_path", default=None, type=str, help="Path to fairseq checkpoint")
parser.add_argument("--dict_path", default=None, type=str, help="Path to dict of fine-tuned model")
parser.add_argument("--config_path", default=None, type=str, help="Path to hf config.json of model to convert")
parser.add_argument(
"--not_finetuned", action="store_true", help="Whether the model to convert is a fine-tuned model or not"
)
parser.add_argument(
"--is_seq_class",
action="store_true",
help="Whether the model to convert is a fine-tuned sequence classification model or not",
)
a_ :str = parser.parse_args()
a_ :Tuple = not args.not_finetuned and not args.is_seq_class
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.config_path,
args.dict_path,
is_finetuned,
args.is_seq_class,
)
| 277 | 0 |
import os
import re
import warnings
from shutil import copyfile
from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple
import sentencepiece as spm
from ...tokenization_utils import PreTrainedTokenizer
if TYPE_CHECKING:
from ...tokenization_utils_base import TextInput
from ...utils import logging
UpperCAmelCase__ = logging.get_logger(__name__)
UpperCAmelCase__ = {"vocab_file": "spiece.model"}
UpperCAmelCase__ = {
"vocab_file": {
"t5-small": "https://huggingface.co/t5-small/resolve/main/spiece.model",
"t5-base": "https://huggingface.co/t5-base/resolve/main/spiece.model",
"t5-large": "https://huggingface.co/t5-large/resolve/main/spiece.model",
"t5-3b": "https://huggingface.co/t5-3b/resolve/main/spiece.model",
"t5-11b": "https://huggingface.co/t5-11b/resolve/main/spiece.model",
}
}
# TODO(PVP) - this should be removed in Transformers v5
UpperCAmelCase__ = {
"t5-small": 512,
"t5-base": 512,
"t5-large": 512,
"t5-3b": 512,
"t5-11b": 512,
}
UpperCAmelCase__ = "▁"
class __lowerCAmelCase ( lowerCAmelCase_ ):
UpperCamelCase = VOCAB_FILES_NAMES
UpperCamelCase = PRETRAINED_VOCAB_FILES_MAP
UpperCamelCase = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
UpperCamelCase = ['''input_ids''', '''attention_mask''']
def __init__( self : Dict , A : List[Any] , A : str="</s>" , A : List[Any]="<unk>" , A : Union[str, Any]="<pad>" , A : List[Any]=1_00 , A : str=None , A : Optional[Dict[str, Any]] = None , A : Optional[Any]=True , **A : List[str] , ) -> None:
"""simple docstring"""
if extra_ids > 0 and additional_special_tokens is None:
_UpperCAmelCase = [F"<extra_id_{i}>" for i in range(_snake_case)]
elif extra_ids > 0 and additional_special_tokens is not None:
# Check that we have the right number of extra_id special tokens
_UpperCAmelCase = len(set(filter(lambda A: bool('extra_id' in str(_snake_case)) , _snake_case)))
if extra_tokens != extra_ids:
raise ValueError(
F"Both extra_ids ({extra_ids}) and additional_special_tokens ({additional_special_tokens}) are"
' provided to T5Tokenizer. In this case the additional_special_tokens must include the extra_ids'
' tokens')
if legacy:
logger.warning_once(
F"You are using the legacy behaviour of the {self.__class__}. This means that tokens that come after special tokens will not be properly handled. We recommend you to"
' read the related pull request available at https://github.com/huggingface/transformers/pull/24565')
_UpperCAmelCase = legacy
_UpperCAmelCase = {} if sp_model_kwargs is None else sp_model_kwargs
super().__init__(
eos_token=_snake_case , unk_token=_snake_case , pad_token=_snake_case , extra_ids=_snake_case , additional_special_tokens=_snake_case , sp_model_kwargs=self.sp_model_kwargs , legacy=_snake_case , **_snake_case , )
_UpperCAmelCase = vocab_file
_UpperCAmelCase = extra_ids
_UpperCAmelCase = spm.SentencePieceProcessor(**self.sp_model_kwargs)
self.sp_model.Load(_snake_case)
@staticmethod
def _lowerCamelCase ( A : Optional[int] , A : Optional[int] , A : List[Any]) -> Tuple:
"""simple docstring"""
if pretrained_model_name_or_path in TaTokenizer.max_model_input_sizes:
_UpperCAmelCase = TaTokenizer.max_model_input_sizes[pretrained_model_name_or_path]
if init_max_model_length is not None and init_max_model_length != max_model_length:
return init_max_model_length
elif init_max_model_length is None:
warnings.warn(
'This tokenizer was incorrectly instantiated with a model max length of'
F" {deprecated_max_model_length} which will be corrected in Transformers v5.\nFor now, this"
' behavior is kept to avoid breaking backwards compatibility when padding/encoding with'
' `truncation is True`.\n- Be aware that you SHOULD NOT rely on'
F" {pretrained_model_name_or_path} automatically truncating your input to"
F" {deprecated_max_model_length} when padding/encoding.\n- If you want to encode/pad to sequences"
F" longer than {deprecated_max_model_length} you can either instantiate this tokenizer with"
' `model_max_length` or pass `max_length` when encoding/padding.\n- To avoid this warning, please'
' instantiate this tokenizer with `model_max_length` set to your preferred value.' , _snake_case , )
return max_model_length
@property
def _lowerCamelCase ( self : Optional[Any]) -> Tuple:
"""simple docstring"""
return self.sp_model.get_piece_size() + self._extra_ids
def _lowerCamelCase ( self : Optional[Any]) -> List[Any]:
"""simple docstring"""
_UpperCAmelCase = {self.convert_ids_to_tokens(_snake_case): i for i in range(self.vocab_size)}
vocab.update(self.added_tokens_encoder)
return vocab
def _lowerCamelCase ( self : List[Any] , A : List[int] , A : Optional[List[int]] = None , A : bool = False) -> List[int]:
"""simple docstring"""
if already_has_special_tokens:
return super().get_special_tokens_mask(
token_ids_a=_snake_case , token_ids_a=_snake_case , already_has_special_tokens=_snake_case)
# normal case: some special tokens
if token_ids_a is None:
return ([0] * len(_snake_case)) + [1]
return ([0] * len(_snake_case)) + [1] + ([0] * len(_snake_case)) + [1]
def _lowerCamelCase ( self : Tuple) -> Union[str, Any]:
"""simple docstring"""
return list(
set(filter(lambda A: bool(re.search(R'<extra_id_\d+>' , _snake_case)) is not None , self.additional_special_tokens)))
def _lowerCamelCase ( self : Any) -> str:
"""simple docstring"""
return [self._convert_token_to_id(_snake_case) for token in self.get_sentinel_tokens()]
def _lowerCamelCase ( self : Dict , A : List[int]) -> List[int]:
"""simple docstring"""
if len(_snake_case) > 0 and token_ids[-1] == self.eos_token_id:
warnings.warn(
F"This sequence already has {self.eos_token}. In future versions this behavior may lead to duplicated"
' eos tokens being added.')
return token_ids
else:
return token_ids + [self.eos_token_id]
def _lowerCamelCase ( self : Optional[Any] , A : List[int] , A : Optional[List[int]] = None) -> List[int]:
"""simple docstring"""
_UpperCAmelCase = [self.eos_token_id]
if token_ids_a is None:
return len(token_ids_a + eos) * [0]
return len(token_ids_a + eos + token_ids_a + eos) * [0]
def _lowerCamelCase ( self : int , A : List[int] , A : Optional[List[int]] = None) -> List[int]:
"""simple docstring"""
_UpperCAmelCase = self._add_eos_if_not_present(_snake_case)
if token_ids_a is None:
return token_ids_a
else:
_UpperCAmelCase = self._add_eos_if_not_present(_snake_case)
return token_ids_a + token_ids_a
def __getstate__( self : Optional[Any]) -> Union[str, Any]:
"""simple docstring"""
_UpperCAmelCase = self.__dict__.copy()
_UpperCAmelCase = None
return state
def __setstate__( self : Optional[int] , A : Optional[int]) -> Dict:
"""simple docstring"""
_UpperCAmelCase = d
# for backward compatibility
if not hasattr(self , 'sp_model_kwargs'):
_UpperCAmelCase = {}
_UpperCAmelCase = spm.SentencePieceProcessor(**self.sp_model_kwargs)
self.sp_model.Load(self.vocab_file)
def _lowerCamelCase ( self : Tuple , A : "TextInput" , **A : Union[str, Any]) -> List[str]:
"""simple docstring"""
if not self.legacy:
_UpperCAmelCase = SPIECE_UNDERLINE + text.replace(_snake_case , ' ')
return super().tokenize(_snake_case , **_snake_case)
def _lowerCamelCase ( self : int , A : Optional[Any] , **A : Optional[Any]) -> int:
"""simple docstring"""
if not self.legacy:
_UpperCAmelCase = text.startswith(_snake_case)
if is_first:
_UpperCAmelCase = text[1:]
_UpperCAmelCase = self.sp_model.encode(_snake_case , out_type=_snake_case)
if not self.legacy and not is_first and not text.startswith(' ') and tokens[0].startswith(_snake_case):
_UpperCAmelCase = ([tokens[0][1:]] if len(tokens[0]) > 1 else []) + tokens[1:]
return tokens
def _lowerCamelCase ( self : Union[str, Any] , A : List[Any]) -> int:
"""simple docstring"""
if token.startswith('<extra_id_'):
_UpperCAmelCase = re.match(R'<extra_id_(\d+)>' , _snake_case)
_UpperCAmelCase = int(match.group(1))
return self.vocab_size - num - 1
return self.sp_model.piece_to_id(_snake_case)
def _lowerCamelCase ( self : int , A : int) -> int:
"""simple docstring"""
if index < self.sp_model.get_piece_size():
_UpperCAmelCase = self.sp_model.IdToPiece(_snake_case)
else:
_UpperCAmelCase = F"<extra_id_{self.vocab_size - 1 - index}>"
return token
def _lowerCamelCase ( self : Optional[Any] , A : Union[str, Any]) -> Optional[int]:
"""simple docstring"""
_UpperCAmelCase = []
_UpperCAmelCase = ''
_UpperCAmelCase = False
for token in tokens:
# make sure that special tokens are not decoded using sentencepiece model
if token in self.all_special_tokens:
if not prev_is_special:
out_string += " "
out_string += self.sp_model.decode(_snake_case) + token
_UpperCAmelCase = True
_UpperCAmelCase = []
else:
current_sub_tokens.append(_snake_case)
_UpperCAmelCase = False
out_string += self.sp_model.decode(_snake_case)
return out_string.strip()
def _lowerCamelCase ( self : int , A : str , A : Optional[str] = None) -> Tuple[str]:
"""simple docstring"""
if not os.path.isdir(_snake_case):
logger.error(F"Vocabulary path ({save_directory}) should be a directory")
return
_UpperCAmelCase = os.path.join(
_snake_case , (filename_prefix + '-' if filename_prefix else '') + VOCAB_FILES_NAMES['vocab_file'])
if os.path.abspath(self.vocab_file) != os.path.abspath(_snake_case) and os.path.isfile(self.vocab_file):
copyfile(self.vocab_file , _snake_case)
elif not os.path.isfile(self.vocab_file):
with open(_snake_case , 'wb') as fi:
_UpperCAmelCase = self.sp_model.serialized_model_proto()
fi.write(_snake_case)
return (out_vocab_file,)
| 339 |
from typing import Dict, List
from nltk.translate import gleu_score
import datasets
from datasets import MetricInfo
a_ :Any = "\\n@misc{wu2016googles,\n title={Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation},\n author={Yonghui Wu and Mike Schuster and Zhifeng Chen and Quoc V. Le and Mohammad Norouzi and Wolfgang Macherey\n and Maxim Krikun and Yuan Cao and Qin Gao and Klaus Macherey and Jeff Klingner and Apurva Shah and Melvin\n Johnson and Xiaobing Liu and Łukasz Kaiser and Stephan Gouws and Yoshikiyo Kato and Taku Kudo and Hideto\n Kazawa and Keith Stevens and George Kurian and Nishant Patil and Wei Wang and Cliff Young and\n Jason Smith and Jason Riesa and Alex Rudnick and Oriol Vinyals and Greg Corrado and Macduff Hughes\n and Jeffrey Dean},\n year={2016},\n eprint={1609.08144},\n archivePrefix={arXiv},\n primaryClass={cs.CL}\n}\n"
a_ :List[str] = "\\nThe BLEU score has some undesirable properties when used for single\nsentences, as it was designed to be a corpus measure. We therefore\nuse a slightly different score for our RL experiments which we call\nthe 'GLEU score'. For the GLEU score, we record all sub-sequences of\n1, 2, 3 or 4 tokens in output and target sequence (n-grams). We then\ncompute a recall, which is the ratio of the number of matching n-grams\nto the number of total n-grams in the target (ground truth) sequence,\nand a precision, which is the ratio of the number of matching n-grams\nto the number of total n-grams in the generated output sequence. Then\nGLEU score is simply the minimum of recall and precision. This GLEU\nscore's range is always between 0 (no matches) and 1 (all match) and\nit is symmetrical when switching output and target. According to\nour experiments, GLEU score correlates quite well with the BLEU\nmetric on a corpus level but does not have its drawbacks for our per\nsentence reward objective.\n"
a_ :List[str] = "\\nComputes corpus-level Google BLEU (GLEU) score of translated segments against one or more references.\nInstead of averaging the sentence level GLEU scores (i.e. macro-average precision), Wu et al. (2016) sum up the matching\ntokens and the max of hypothesis and reference tokens for each sentence, then compute using the aggregate values.\n\nArgs:\n predictions (list of str): list of translations to score.\n Each translation should be tokenized into a list of tokens.\n references (list of list of str): list of lists of references for each translation.\n Each reference should be tokenized into a list of tokens.\n min_len (int): The minimum order of n-gram this function should extract. Defaults to 1.\n max_len (int): The maximum order of n-gram this function should extract. Defaults to 4.\n\nReturns:\n 'google_bleu': google_bleu score\n\nExamples:\n Example 1:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses, references=list_of_references)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.44\n\n Example 2:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n >>> ref1b = ['It', 'is', 'a', 'guide', 'to', 'action', 'that',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'will', 'never',\n ... 'heed', 'the', 'cat', 'commands']\n >>> ref1c = ['It', 'is', 'the', 'practical', 'guide', 'for', 'the',\n ... 'rubber', 'duck', 'army', 'never', 'to', 'heed', 'the', 'directions',\n ... 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a, ref1b, ref1c], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses, references=list_of_references)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.61\n\n Example 3:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n >>> ref1b = ['It', 'is', 'a', 'guide', 'to', 'action', 'that',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'will', 'never',\n ... 'heed', 'the', 'cat', 'commands']\n >>> ref1c = ['It', 'is', 'the', 'practical', 'guide', 'for', 'the',\n ... 'rubber', 'duck', 'army', 'never', 'to', 'heed', 'the', 'directions',\n ... 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a, ref1b, ref1c], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses, references=list_of_references, min_len=2)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.53\n\n Example 4:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n >>> ref1b = ['It', 'is', 'a', 'guide', 'to', 'action', 'that',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'will', 'never',\n ... 'heed', 'the', 'cat', 'commands']\n >>> ref1c = ['It', 'is', 'the', 'practical', 'guide', 'for', 'the',\n ... 'rubber', 'duck', 'army', 'never', 'to', 'heed', 'the', 'directions',\n ... 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a, ref1b, ref1c], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses,references=list_of_references, min_len=2, max_len=6)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.4\n"
@datasets.utils.file_utils.add_start_docstrings(_DESCRIPTION , _KWARGS_DESCRIPTION )
class snake_case__ ( datasets.Metric ):
"""simple docstring"""
def lowercase_ ( self : str ) ->MetricInfo:
return datasets.MetricInfo(
description=_DESCRIPTION, citation=_CITATION, inputs_description=_KWARGS_DESCRIPTION, features=datasets.Features(
{
'predictions': datasets.Sequence(datasets.Value('string', id='token' ), id='sequence' ),
'references': datasets.Sequence(
datasets.Sequence(datasets.Value('string', id='token' ), id='sequence' ), id='references' ),
} ), )
def lowercase_ ( self : str, _snake_case : List[List[List[str]]], _snake_case : List[List[str]], _snake_case : int = 1, _snake_case : int = 4, ) ->Dict[str, float]:
return {
"google_bleu": gleu_score.corpus_gleu(
list_of_references=_snake_case, hypotheses=_snake_case, min_len=_snake_case, max_len=_snake_case )
}
| 277 | 0 |
from __future__ import annotations
import typing
from collections.abc import Iterable
import numpy as np
lowercase = typing.Union[Iterable[float], Iterable[int], np.ndarray] # noqa: UP007
lowercase = typing.Union[np.floataa, int, float] # noqa: UP007
def __UpperCAmelCase ( a_ , a_):
return np.sqrt(np.sum((np.asarray(a_) - np.asarray(a_)) ** 2))
def __UpperCAmelCase ( a_ , a_):
return sum((va - va) ** 2 for va, va in zip(a_ , a_)) ** (1 / 2)
if __name__ == "__main__":
def __UpperCAmelCase ( ):
from timeit import timeit
print('Without Numpy')
print(
timeit(
'euclidean_distance_no_np([1, 2, 3], [4, 5, 6])' , number=1_00_00 , globals=globals() , ))
print('With Numpy')
print(
timeit(
'euclidean_distance([1, 2, 3], [4, 5, 6])' , number=1_00_00 , globals=globals() , ))
benchmark()
| 178 |
from math import factorial
def lowercase_ (A : int , A : int , A : float ):
if successes > trials:
raise ValueError('successes must be lower or equal to trials' )
if trials < 0 or successes < 0:
raise ValueError('the function is defined for non-negative integers' )
if not isinstance(A , A ) or not isinstance(A , A ):
raise ValueError('the function is defined for non-negative integers' )
if not 0 < prob < 1:
raise ValueError('prob has to be in range of 1 - 0' )
snake_case__ : List[Any] = (prob**successes) * ((1 - prob) ** (trials - successes))
# Calculate the binomial coefficient: n! / k!(n-k)!
snake_case__ : List[str] = float(factorial(A ) )
coefficient /= factorial(A ) * factorial(trials - successes )
return probability * coefficient
if __name__ == "__main__":
from doctest import testmod
testmod()
print("Probability of 2 successes out of 4 trails")
print("with probability of 0.75 is:", end=" ")
print(binomial_distribution(2, 4, 0.75))
| 277 | 0 |
"""simple docstring"""
import os
import random
import sys
from . import cryptomath_module as cryptomath
from . import rabin_miller
__UpperCamelCase = 3
def lowercase (SCREAMING_SNAKE_CASE_ : int ) -> Dict:
print('Generating primitive root of p' )
while True:
SCREAMING_SNAKE_CASE = random.randrange(3 , SCREAMING_SNAKE_CASE_ )
if pow(SCREAMING_SNAKE_CASE_ , 2 , SCREAMING_SNAKE_CASE_ ) == 1:
continue
if pow(SCREAMING_SNAKE_CASE_ , SCREAMING_SNAKE_CASE_ , SCREAMING_SNAKE_CASE_ ) == 1:
continue
return g
def lowercase (SCREAMING_SNAKE_CASE_ : int ) -> List[Any]:
print('Generating prime p...' )
SCREAMING_SNAKE_CASE = rabin_miller.generate_large_prime(SCREAMING_SNAKE_CASE_ ) # select large prime number.
SCREAMING_SNAKE_CASE = primitive_root(SCREAMING_SNAKE_CASE_ ) # one primitive root on modulo p.
SCREAMING_SNAKE_CASE = random.randrange(3 , SCREAMING_SNAKE_CASE_ ) # private_key -> have to be greater than 2 for safety.
SCREAMING_SNAKE_CASE = cryptomath.find_mod_inverse(pow(SCREAMING_SNAKE_CASE_ , SCREAMING_SNAKE_CASE_ , SCREAMING_SNAKE_CASE_ ) , SCREAMING_SNAKE_CASE_ )
SCREAMING_SNAKE_CASE = (key_size, e_a, e_a, p)
SCREAMING_SNAKE_CASE = (key_size, d)
return public_key, private_key
def lowercase (SCREAMING_SNAKE_CASE_ : str , SCREAMING_SNAKE_CASE_ : int ) -> List[str]:
if os.path.exists(F'{name}_pubkey.txt' ) or os.path.exists(F'{name}_privkey.txt' ):
print('\nWARNING:' )
print(
F'"{name}_pubkey.txt" or "{name}_privkey.txt" already exists. \n'
'Use a different name or delete these files and re-run this program.' )
sys.exit()
SCREAMING_SNAKE_CASE = generate_key(SCREAMING_SNAKE_CASE_ )
print(F'\nWriting public key to file {name}_pubkey.txt...' )
with open(F'{name}_pubkey.txt' , 'w' ) as fo:
fo.write(F'{public_key[0]},{public_key[1]},{public_key[2]},{public_key[3]}' )
print(F'Writing private key to file {name}_privkey.txt...' )
with open(F'{name}_privkey.txt' , 'w' ) as fo:
fo.write(F'{private_key[0]},{private_key[1]}' )
def lowercase () -> List[str]:
print('Making key files...' )
make_key_files('elgamal' , 20_48 )
print('Key files generation successful' )
if __name__ == "__main__":
main()
| 113 |
from collections import UserDict
from typing import Union
import numpy as np
import requests
from ..utils import (
add_end_docstrings,
logging,
)
from .audio_classification import ffmpeg_read
from .base import PIPELINE_INIT_ARGS, Pipeline
a_ :List[Any] = logging.get_logger(__name__)
@add_end_docstrings(lowerCAmelCase_ )
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
def __init__( self : Optional[Any], **_snake_case : str ) ->Dict:
super().__init__(**_snake_case )
if self.framework != "pt":
raise ValueError(F'''The {self.__class__} is only available in PyTorch.''' )
# No specific FOR_XXX available yet
def __call__( self : Union[str, Any], _snake_case : Union[np.ndarray, bytes, str], **_snake_case : Tuple ) ->Dict:
return super().__call__(_snake_case, **_snake_case )
def lowercase_ ( self : Tuple, **_snake_case : Any ) ->Union[str, Any]:
snake_case__ : str = {}
if "candidate_labels" in kwargs:
snake_case__ : str = kwargs['candidate_labels']
if "hypothesis_template" in kwargs:
snake_case__ : str = kwargs['hypothesis_template']
return preprocess_params, {}, {}
def lowercase_ ( self : Dict, _snake_case : str, _snake_case : Optional[int]=None, _snake_case : List[str]="This is a sound of {}." ) ->int:
if isinstance(_snake_case, _snake_case ):
if audio.startswith('http://' ) or audio.startswith('https://' ):
# We need to actually check for a real protocol, otherwise it's impossible to use a local file
# like http_huggingface_co.png
snake_case__ : List[Any] = requests.get(_snake_case ).content
else:
with open(_snake_case, 'rb' ) as f:
snake_case__ : Union[str, Any] = f.read()
if isinstance(_snake_case, _snake_case ):
snake_case__ : List[Any] = ffmpeg_read(_snake_case, self.feature_extractor.sampling_rate )
if not isinstance(_snake_case, np.ndarray ):
raise ValueError('We expect a numpy ndarray as input' )
if len(audio.shape ) != 1:
raise ValueError('We expect a single channel audio input for ZeroShotAudioClassificationPipeline' )
snake_case__ : Tuple = self.feature_extractor(
[audio], sampling_rate=self.feature_extractor.sampling_rate, return_tensors='pt' )
snake_case__ : int = candidate_labels
snake_case__ : int = [hypothesis_template.format(_snake_case ) for x in candidate_labels]
snake_case__ : Optional[int] = self.tokenizer(_snake_case, return_tensors=self.framework, padding=_snake_case )
snake_case__ : List[Any] = [text_inputs]
return inputs
def lowercase_ ( self : Optional[int], _snake_case : Optional[Any] ) ->int:
snake_case__ : Optional[int] = model_inputs.pop('candidate_labels' )
snake_case__ : str = model_inputs.pop('text_inputs' )
if isinstance(text_inputs[0], _snake_case ):
snake_case__ : Optional[Any] = text_inputs[0]
else:
# Batching case.
snake_case__ : int = text_inputs[0][0]
snake_case__ : Any = self.model(**_snake_case, **_snake_case )
snake_case__ : List[Any] = {
'candidate_labels': candidate_labels,
'logits': outputs.logits_per_audio,
}
return model_outputs
def lowercase_ ( self : Union[str, Any], _snake_case : str ) ->List[str]:
snake_case__ : int = model_outputs.pop('candidate_labels' )
snake_case__ : List[Any] = model_outputs['logits'][0]
if self.framework == "pt":
snake_case__ : Tuple = logits.softmax(dim=0 )
snake_case__ : Union[str, Any] = probs.tolist()
else:
raise ValueError('`tf` framework not supported.' )
snake_case__ : Union[str, Any] = [
{'score': score, 'label': candidate_label}
for score, candidate_label in sorted(zip(_snake_case, _snake_case ), key=lambda _snake_case : -x[0] )
]
return result
| 277 | 0 |
'''simple docstring'''
from manim import *
class _lowercase ( lowerCAmelCase_ ):
'''simple docstring'''
def a ( self : int ) -> Any:
__lowerCAmelCase = Rectangle(height=0.5 , width=0.5 )
__lowerCAmelCase = Rectangle(height=0.4_6 , width=0.4_6 ).set_stroke(width=0 )
__lowerCAmelCase = [mem.copy() for i in range(6 )]
__lowerCAmelCase = [mem.copy() for i in range(6 )]
__lowerCAmelCase = VGroup(*_snake_case ).arrange(_snake_case , buff=0 )
__lowerCAmelCase = VGroup(*_snake_case ).arrange(_snake_case , buff=0 )
__lowerCAmelCase = VGroup(_snake_case , _snake_case ).arrange(_snake_case , buff=0 )
__lowerCAmelCase = Text("""CPU""" , font_size=24 )
__lowerCAmelCase = Group(_snake_case , _snake_case ).arrange(_snake_case , buff=0.5 , aligned_edge=_snake_case )
cpu.move_to([-2.5, -0.5, 0] )
self.add(_snake_case )
__lowerCAmelCase = [mem.copy() for i in range(1 )]
__lowerCAmelCase = VGroup(*_snake_case ).arrange(_snake_case , buff=0 )
__lowerCAmelCase = Text("""GPU""" , font_size=24 )
__lowerCAmelCase = Group(_snake_case , _snake_case ).arrange(_snake_case , buff=0.5 , aligned_edge=_snake_case )
gpu.align_to(_snake_case , _snake_case )
gpu.set_x(gpu.get_x() - 1 )
self.add(_snake_case )
__lowerCAmelCase = [mem.copy() for i in range(6 )]
__lowerCAmelCase = VGroup(*_snake_case ).arrange(_snake_case , buff=0 )
__lowerCAmelCase = Text("""Model""" , font_size=24 )
__lowerCAmelCase = Group(_snake_case , _snake_case ).arrange(_snake_case , buff=0.5 , aligned_edge=_snake_case )
model.move_to([3, -1.0, 0] )
self.play(
Create(_snake_case , run_time=1 ) , Create(_snake_case , run_time=1 ) , Create(_snake_case , run_time=1 ) , )
__lowerCAmelCase = MarkupText(
f"""First, an empty model skeleton is loaded\ninto <span fgcolor=\'{YELLOW}\'>memory</span> without using much RAM.""" , font_size=24 , )
__lowerCAmelCase = Square(side_length=2.2 )
key.move_to([-5, 2, 0] )
__lowerCAmelCase = MarkupText(
f"""<b>Key:</b>\n\n<span fgcolor=\'{YELLOW}\'>●</span> Empty Model""" , font_size=18 , )
key_text.move_to([-5, 2.4, 0] )
step_a.move_to([2, 2, 0] )
self.play(Write(_snake_case , run_time=2.5 ) , Write(_snake_case ) , Write(_snake_case ) )
self.add(_snake_case )
__lowerCAmelCase = []
__lowerCAmelCase = []
__lowerCAmelCase = []
for i, rect in enumerate(_snake_case ):
__lowerCAmelCase = Rectangle(height=0.4_6 , width=0.4_6 ).set_stroke(width=0.0 ).set_fill(_snake_case , opacity=0.7 )
cpu_target.move_to(_snake_case )
cpu_target.generate_target()
__lowerCAmelCase = 0.4_6 / 4
__lowerCAmelCase = 0.4_6 / 3
if i == 0:
cpu_target.target.next_to(cpu_left_col_base[0].get_corner(DOWN + LEFT ) , buff=0.0_2 , direction=_snake_case )
cpu_target.target.set_x(cpu_target.target.get_x() + 0.1 )
elif i == 3:
cpu_target.target.next_to(cpu_targs[0].target , direction=_snake_case , buff=0.0 )
else:
cpu_target.target.next_to(cpu_targs[i - 1].target , direction=_snake_case , buff=0.0 )
cpu_targs.append(_snake_case )
first_animations.append(rect.animate(run_time=0.5 ).set_stroke(_snake_case ) )
second_animations.append(MoveToTarget(_snake_case , run_time=1.5 ) )
self.play(*_snake_case )
self.play(*_snake_case )
self.wait()
| 229 |
import collections
import inspect
import unittest
from transformers import SwinvaConfig
from transformers.testing_utils import require_torch, require_vision, slow, torch_device
from transformers.utils import cached_property, is_torch_available, is_vision_available
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, _config_zero_init, floats_tensor, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from torch import nn
from transformers import SwinvaForImageClassification, SwinvaForMaskedImageModeling, SwinvaModel
from transformers.models.swinva.modeling_swinva import SWINV2_PRETRAINED_MODEL_ARCHIVE_LIST
if is_vision_available():
from PIL import Image
from transformers import AutoImageProcessor
class snake_case__ :
"""simple docstring"""
def __init__( self : Tuple, _snake_case : Any, _snake_case : int=1_3, _snake_case : Optional[int]=3_2, _snake_case : Tuple=2, _snake_case : Any=3, _snake_case : Tuple=1_6, _snake_case : Tuple=[1, 2, 1], _snake_case : Dict=[2, 2, 4], _snake_case : str=2, _snake_case : Union[str, Any]=2.0, _snake_case : Dict=True, _snake_case : Dict=0.0, _snake_case : str=0.0, _snake_case : str=0.1, _snake_case : List[str]="gelu", _snake_case : int=False, _snake_case : Optional[Any]=True, _snake_case : List[Any]=0.0_2, _snake_case : Union[str, Any]=1e-5, _snake_case : Union[str, Any]=True, _snake_case : List[Any]=None, _snake_case : Any=True, _snake_case : List[Any]=1_0, _snake_case : str=8, ) ->Union[str, Any]:
snake_case__ : Any = parent
snake_case__ : Tuple = batch_size
snake_case__ : Tuple = image_size
snake_case__ : Any = patch_size
snake_case__ : Optional[int] = num_channels
snake_case__ : Tuple = embed_dim
snake_case__ : Any = depths
snake_case__ : Any = num_heads
snake_case__ : List[str] = window_size
snake_case__ : Dict = mlp_ratio
snake_case__ : Optional[int] = qkv_bias
snake_case__ : Optional[Any] = hidden_dropout_prob
snake_case__ : List[str] = attention_probs_dropout_prob
snake_case__ : Union[str, Any] = drop_path_rate
snake_case__ : str = hidden_act
snake_case__ : Union[str, Any] = use_absolute_embeddings
snake_case__ : Union[str, Any] = patch_norm
snake_case__ : Any = layer_norm_eps
snake_case__ : Tuple = initializer_range
snake_case__ : Dict = is_training
snake_case__ : Any = scope
snake_case__ : Optional[Any] = use_labels
snake_case__ : str = type_sequence_label_size
snake_case__ : List[Any] = encoder_stride
def lowercase_ ( self : Tuple ) ->str:
snake_case__ : Tuple = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size] )
snake_case__ : List[Any] = None
if self.use_labels:
snake_case__ : Optional[Any] = ids_tensor([self.batch_size], self.type_sequence_label_size )
snake_case__ : Any = self.get_config()
return config, pixel_values, labels
def lowercase_ ( self : Optional[int] ) ->Optional[int]:
return SwinvaConfig(
image_size=self.image_size, patch_size=self.patch_size, num_channels=self.num_channels, embed_dim=self.embed_dim, depths=self.depths, num_heads=self.num_heads, window_size=self.window_size, mlp_ratio=self.mlp_ratio, qkv_bias=self.qkv_bias, hidden_dropout_prob=self.hidden_dropout_prob, attention_probs_dropout_prob=self.attention_probs_dropout_prob, drop_path_rate=self.drop_path_rate, hidden_act=self.hidden_act, use_absolute_embeddings=self.use_absolute_embeddings, path_norm=self.patch_norm, layer_norm_eps=self.layer_norm_eps, initializer_range=self.initializer_range, encoder_stride=self.encoder_stride, )
def lowercase_ ( self : Optional[int], _snake_case : str, _snake_case : List[str], _snake_case : int ) ->Dict:
snake_case__ : List[Any] = SwinvaModel(config=_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Optional[int] = model(_snake_case )
snake_case__ : List[Any] = ((config.image_size // config.patch_size) ** 2) // (4 ** (len(config.depths ) - 1))
snake_case__ : List[Any] = int(config.embed_dim * 2 ** (len(config.depths ) - 1) )
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, expected_seq_len, expected_dim) )
def lowercase_ ( self : Optional[Any], _snake_case : Any, _snake_case : List[str], _snake_case : Dict ) ->List[Any]:
snake_case__ : List[str] = SwinvaForMaskedImageModeling(config=_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Union[str, Any] = model(_snake_case )
self.parent.assertEqual(
result.logits.shape, (self.batch_size, self.num_channels, self.image_size, self.image_size) )
# test greyscale images
snake_case__ : Optional[Any] = 1
snake_case__ : Optional[int] = SwinvaForMaskedImageModeling(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Tuple = floats_tensor([self.batch_size, 1, self.image_size, self.image_size] )
snake_case__ : Any = model(_snake_case )
self.parent.assertEqual(result.logits.shape, (self.batch_size, 1, self.image_size, self.image_size) )
def lowercase_ ( self : List[str], _snake_case : int, _snake_case : List[Any], _snake_case : Optional[int] ) ->Any:
snake_case__ : Tuple = self.type_sequence_label_size
snake_case__ : int = SwinvaForImageClassification(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Tuple = model(_snake_case, labels=_snake_case )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.type_sequence_label_size) )
def lowercase_ ( self : Any ) ->Dict:
snake_case__ : str = self.prepare_config_and_inputs()
snake_case__ , snake_case__ , snake_case__ : List[str] = config_and_inputs
snake_case__ : Union[str, Any] = {'pixel_values': pixel_values}
return config, inputs_dict
@require_torch
class snake_case__ ( lowerCAmelCase_ , lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = (
(SwinvaModel, SwinvaForImageClassification, SwinvaForMaskedImageModeling) if is_torch_available() else ()
)
_SCREAMING_SNAKE_CASE = (
{"""feature-extraction""": SwinvaModel, """image-classification""": SwinvaForImageClassification}
if is_torch_available()
else {}
)
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
def lowercase_ ( self : Union[str, Any] ) ->Dict:
snake_case__ : Optional[int] = SwinvaModelTester(self )
snake_case__ : int = ConfigTester(self, config_class=_snake_case, embed_dim=3_7 )
def lowercase_ ( self : Tuple ) ->int:
self.config_tester.create_and_test_config_to_json_string()
self.config_tester.create_and_test_config_to_json_file()
self.config_tester.create_and_test_config_from_and_save_pretrained()
self.config_tester.create_and_test_config_with_num_labels()
self.config_tester.check_config_can_be_init_without_params()
self.config_tester.check_config_arguments_init()
def lowercase_ ( self : Any ) ->str:
snake_case__ : int = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*_snake_case )
@unittest.skip(reason='Got `CUDA error: misaligned address` with PyTorch 2.0.0.' )
def lowercase_ ( self : Any ) ->Union[str, Any]:
pass
@unittest.skip(reason='Swinv2 does not use inputs_embeds' )
def lowercase_ ( self : str ) ->Union[str, Any]:
pass
def lowercase_ ( self : Optional[Any] ) ->Union[str, Any]:
snake_case__ , snake_case__ : Optional[int] = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
snake_case__ : Union[str, Any] = model_class(_snake_case )
self.assertIsInstance(model.get_input_embeddings(), (nn.Module) )
snake_case__ : Union[str, Any] = model.get_output_embeddings()
self.assertTrue(x is None or isinstance(_snake_case, nn.Linear ) )
def lowercase_ ( self : List[str] ) ->Optional[int]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
snake_case__ : Any = model_class(_snake_case )
snake_case__ : Dict = inspect.signature(model.forward )
# signature.parameters is an OrderedDict => so arg_names order is deterministic
snake_case__ : Optional[Any] = [*signature.parameters.keys()]
snake_case__ : List[Any] = ['pixel_values']
self.assertListEqual(arg_names[:1], _snake_case )
def lowercase_ ( self : str ) ->Union[str, Any]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : int = True
for model_class in self.all_model_classes:
snake_case__ : str = True
snake_case__ : Union[str, Any] = False
snake_case__ : Tuple = True
snake_case__ : int = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : Optional[int] = model(**self._prepare_for_class(_snake_case, _snake_case ) )
snake_case__ : List[str] = outputs.attentions
snake_case__ : List[Any] = len(self.model_tester.depths )
self.assertEqual(len(_snake_case ), _snake_case )
# check that output_attentions also work using config
del inputs_dict["output_attentions"]
snake_case__ : str = True
snake_case__ : Tuple = config.window_size**2
snake_case__ : Optional[int] = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : str = model(**self._prepare_for_class(_snake_case, _snake_case ) )
snake_case__ : Tuple = outputs.attentions
self.assertEqual(len(_snake_case ), _snake_case )
self.assertListEqual(
list(attentions[0].shape[-3:] ), [self.model_tester.num_heads[0], window_size_squared, window_size_squared], )
snake_case__ : Optional[Any] = len(_snake_case )
# Check attention is always last and order is fine
snake_case__ : Optional[int] = True
snake_case__ : Dict = True
snake_case__ : List[Any] = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : Optional[int] = model(**self._prepare_for_class(_snake_case, _snake_case ) )
if hasattr(self.model_tester, 'num_hidden_states_types' ):
snake_case__ : str = self.model_tester.num_hidden_states_types
else:
# also another +1 for reshaped_hidden_states
snake_case__ : Dict = 2
self.assertEqual(out_len + added_hidden_states, len(_snake_case ) )
snake_case__ : Any = outputs.attentions
self.assertEqual(len(_snake_case ), _snake_case )
self.assertListEqual(
list(self_attentions[0].shape[-3:] ), [self.model_tester.num_heads[0], window_size_squared, window_size_squared], )
def lowercase_ ( self : Dict, _snake_case : Tuple, _snake_case : Any, _snake_case : int, _snake_case : Optional[int] ) ->str:
snake_case__ : Dict = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : List[Any] = model(**self._prepare_for_class(_snake_case, _snake_case ) )
snake_case__ : Dict = outputs.hidden_states
snake_case__ : int = getattr(
self.model_tester, 'expected_num_hidden_layers', len(self.model_tester.depths ) + 1 )
self.assertEqual(len(_snake_case ), _snake_case )
# Swinv2 has a different seq_length
snake_case__ : int = (
config.patch_size
if isinstance(config.patch_size, collections.abc.Iterable )
else (config.patch_size, config.patch_size)
)
snake_case__ : Optional[Any] = (image_size[1] // patch_size[1]) * (image_size[0] // patch_size[0])
self.assertListEqual(
list(hidden_states[0].shape[-2:] ), [num_patches, self.model_tester.embed_dim], )
snake_case__ : Union[str, Any] = outputs.reshaped_hidden_states
self.assertEqual(len(_snake_case ), _snake_case )
snake_case__ , snake_case__ , snake_case__ , snake_case__ : str = reshaped_hidden_states[0].shape
snake_case__ : Any = (
reshaped_hidden_states[0].view(_snake_case, _snake_case, height * width ).permute(0, 2, 1 )
)
self.assertListEqual(
list(reshaped_hidden_states.shape[-2:] ), [num_patches, self.model_tester.embed_dim], )
def lowercase_ ( self : str ) ->List[Any]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : List[Any] = (
self.model_tester.image_size
if isinstance(self.model_tester.image_size, collections.abc.Iterable )
else (self.model_tester.image_size, self.model_tester.image_size)
)
for model_class in self.all_model_classes:
snake_case__ : Optional[int] = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, _snake_case )
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
snake_case__ : Dict = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, _snake_case )
def lowercase_ ( self : List[str] ) ->str:
snake_case__ , snake_case__ : Optional[Any] = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : List[str] = 3
snake_case__ : Union[str, Any] = (
self.model_tester.image_size
if isinstance(self.model_tester.image_size, collections.abc.Iterable )
else (self.model_tester.image_size, self.model_tester.image_size)
)
snake_case__ : str = (
config.patch_size
if isinstance(config.patch_size, collections.abc.Iterable )
else (config.patch_size, config.patch_size)
)
snake_case__ : Tuple = image_size[0] + patch_size[0] - (image_size[0] % patch_size[0])
snake_case__ : Optional[Any] = image_size[1] + patch_size[1] - (image_size[1] % patch_size[1])
for model_class in self.all_model_classes:
snake_case__ : int = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, (padded_height, padded_width) )
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
snake_case__ : List[str] = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, (padded_height, padded_width) )
def lowercase_ ( self : List[str] ) ->Optional[int]:
snake_case__ : Any = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_masked_image_modeling(*_snake_case )
def lowercase_ ( self : List[Any] ) ->str:
snake_case__ : Any = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_image_classification(*_snake_case )
@slow
def lowercase_ ( self : str ) ->Union[str, Any]:
for model_name in SWINV2_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
snake_case__ : Dict = SwinvaModel.from_pretrained(_snake_case )
self.assertIsNotNone(_snake_case )
def lowercase_ ( self : Optional[int] ) ->List[str]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : List[Any] = _config_zero_init(_snake_case )
for model_class in self.all_model_classes:
snake_case__ : List[str] = model_class(config=_snake_case )
for name, param in model.named_parameters():
if "embeddings" not in name and "logit_scale" not in name and param.requires_grad:
self.assertIn(
((param.data.mean() * 1e9).round() / 1e9).item(), [0.0, 1.0], msg=F'''Parameter {name} of model {model_class} seems not properly initialized''', )
@require_vision
@require_torch
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
@cached_property
def lowercase_ ( self : Union[str, Any] ) ->List[str]:
return (
AutoImageProcessor.from_pretrained('microsoft/swinv2-tiny-patch4-window8-256' )
if is_vision_available()
else None
)
@slow
def lowercase_ ( self : int ) ->List[Any]:
snake_case__ : Any = SwinvaForImageClassification.from_pretrained('microsoft/swinv2-tiny-patch4-window8-256' ).to(
_snake_case )
snake_case__ : int = self.default_image_processor
snake_case__ : Union[str, Any] = Image.open('./tests/fixtures/tests_samples/COCO/000000039769.png' )
snake_case__ : Optional[Any] = image_processor(images=_snake_case, return_tensors='pt' ).to(_snake_case )
# forward pass
with torch.no_grad():
snake_case__ : List[str] = model(**_snake_case )
# verify the logits
snake_case__ : int = torch.Size((1, 1_0_0_0) )
self.assertEqual(outputs.logits.shape, _snake_case )
snake_case__ : Optional[int] = torch.tensor([-0.3_9_4_7, -0.4_3_0_6, 0.0_0_2_6] ).to(_snake_case )
self.assertTrue(torch.allclose(outputs.logits[0, :3], _snake_case, atol=1e-4 ) )
| 277 | 0 |
"""simple docstring"""
from __future__ import annotations
from typing import Any
def lowercase ( a__ : list[Any] ) -> int:
create_state_space_tree(a__ , [] , 0 )
def lowercase ( a__ : list[Any] , a__ : list[Any] , a__ : int ) -> List[Any]:
if index == len(a__ ):
print(a__ )
return
create_state_space_tree(a__ , a__ , index + 1 )
current_subsequence.append(sequence[index] )
create_state_space_tree(a__ , a__ , index + 1 )
current_subsequence.pop()
if __name__ == "__main__":
UpperCAmelCase = [3, 1, 2, 4]
generate_all_subsequences(seq)
seq.clear()
seq.extend(["""A""", """B""", """C"""])
generate_all_subsequences(seq)
| 256 |
import unittest
import numpy as np
from transformers.testing_utils import require_torch, require_vision
from transformers.utils import is_torch_available, is_vision_available
from ...test_image_processing_common import ImageProcessingSavingTestMixin, prepare_image_inputs
if is_torch_available():
import torch
if is_vision_available():
from PIL import Image
from transformers import PoolFormerImageProcessor
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
def __init__( self : Optional[int], _snake_case : List[Any], _snake_case : str=7, _snake_case : Tuple=3, _snake_case : List[str]=3_0, _snake_case : Tuple=4_0_0, _snake_case : Any=True, _snake_case : List[Any]=None, _snake_case : int=0.9, _snake_case : Optional[Any]=None, _snake_case : str=True, _snake_case : Union[str, Any]=[0.5, 0.5, 0.5], _snake_case : Union[str, Any]=[0.5, 0.5, 0.5], ) ->List[Any]:
snake_case__ : int = size if size is not None else {'shortest_edge': 3_0}
snake_case__ : Tuple = crop_size if crop_size is not None else {'height': 3_0, 'width': 3_0}
snake_case__ : Union[str, Any] = parent
snake_case__ : Dict = batch_size
snake_case__ : int = num_channels
snake_case__ : Tuple = min_resolution
snake_case__ : Any = max_resolution
snake_case__ : List[Any] = do_resize_and_center_crop
snake_case__ : str = size
snake_case__ : str = crop_pct
snake_case__ : List[str] = crop_size
snake_case__ : Optional[int] = do_normalize
snake_case__ : Tuple = image_mean
snake_case__ : Tuple = image_std
def lowercase_ ( self : Optional[int] ) ->int:
return {
"size": self.size,
"do_resize_and_center_crop": self.do_resize_and_center_crop,
"crop_pct": self.crop_pct,
"crop_size": self.crop_size,
"do_normalize": self.do_normalize,
"image_mean": self.image_mean,
"image_std": self.image_std,
}
@require_torch
@require_vision
class snake_case__ ( lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = PoolFormerImageProcessor if is_vision_available() else None
def lowercase_ ( self : Union[str, Any] ) ->Dict:
snake_case__ : Union[str, Any] = PoolFormerImageProcessingTester(self )
@property
def lowercase_ ( self : int ) ->Dict:
return self.image_processor_tester.prepare_image_processor_dict()
def lowercase_ ( self : Union[str, Any] ) ->Optional[int]:
snake_case__ : List[str] = self.image_processing_class(**self.image_processor_dict )
self.assertTrue(hasattr(_snake_case, 'do_resize_and_center_crop' ) )
self.assertTrue(hasattr(_snake_case, 'size' ) )
self.assertTrue(hasattr(_snake_case, 'crop_pct' ) )
self.assertTrue(hasattr(_snake_case, 'do_normalize' ) )
self.assertTrue(hasattr(_snake_case, 'image_mean' ) )
self.assertTrue(hasattr(_snake_case, 'image_std' ) )
def lowercase_ ( self : List[str] ) ->List[str]:
snake_case__ : Union[str, Any] = self.image_processing_class.from_dict(self.image_processor_dict )
self.assertEqual(image_processor.size, {'shortest_edge': 3_0} )
self.assertEqual(image_processor.crop_size, {'height': 3_0, 'width': 3_0} )
snake_case__ : int = self.image_processing_class.from_dict(self.image_processor_dict, size=4_2, crop_size=8_4 )
self.assertEqual(image_processor.size, {'shortest_edge': 4_2} )
self.assertEqual(image_processor.crop_size, {'height': 8_4, 'width': 8_4} )
def lowercase_ ( self : List[Any] ) ->List[Any]:
pass
def lowercase_ ( self : List[str] ) ->str:
# Initialize image_processing
snake_case__ : Union[str, Any] = self.image_processing_class(**self.image_processor_dict )
# create random PIL images
snake_case__ : List[str] = prepare_image_inputs(self.image_processor_tester, equal_resolution=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case, Image.Image )
# Test not batched input
snake_case__ : Optional[int] = image_processing(image_inputs[0], return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
# Test batched
snake_case__ : str = image_processing(_snake_case, return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
def lowercase_ ( self : int ) ->List[Any]:
# Initialize image_processing
snake_case__ : Optional[int] = self.image_processing_class(**self.image_processor_dict )
# create random numpy tensors
snake_case__ : Dict = prepare_image_inputs(self.image_processor_tester, equal_resolution=_snake_case, numpify=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case, np.ndarray )
# Test not batched input
snake_case__ : Dict = image_processing(image_inputs[0], return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
# Test batched
snake_case__ : List[Any] = image_processing(_snake_case, return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
def lowercase_ ( self : List[str] ) ->List[str]:
# Initialize image_processing
snake_case__ : Tuple = self.image_processing_class(**self.image_processor_dict )
# create random PyTorch tensors
snake_case__ : List[str] = prepare_image_inputs(self.image_processor_tester, equal_resolution=_snake_case, torchify=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case, torch.Tensor )
# Test not batched input
snake_case__ : Tuple = image_processing(image_inputs[0], return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
# Test batched
snake_case__ : Optional[Any] = image_processing(_snake_case, return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
| 277 | 0 |
from typing import Optional, Tuple, Union
import torch
from einops import rearrange, reduce
from diffusers import DDIMScheduler, DDPMScheduler, DiffusionPipeline, ImagePipelineOutput, UNetaDConditionModel
from diffusers.schedulers.scheduling_ddim import DDIMSchedulerOutput
from diffusers.schedulers.scheduling_ddpm import DDPMSchedulerOutput
_a = 8
def _a ( SCREAMING_SNAKE_CASE : Optional[int] , SCREAMING_SNAKE_CASE : Optional[int]=BITS ) -> int:
"""simple docstring"""
__lowerCAmelCase: Optional[Any] = x.device
__lowerCAmelCase: List[Any] = (x * 2_55).int().clamp(0 , 2_55 )
__lowerCAmelCase: Any = 2 ** torch.arange(bits - 1 , -1 , -1 , device=SCREAMING_SNAKE_CASE )
__lowerCAmelCase: Tuple = rearrange(SCREAMING_SNAKE_CASE , 'd -> d 1 1' )
__lowerCAmelCase: Tuple = rearrange(SCREAMING_SNAKE_CASE , 'b c h w -> b c 1 h w' )
__lowerCAmelCase: Dict = ((x & mask) != 0).float()
__lowerCAmelCase: Dict = rearrange(SCREAMING_SNAKE_CASE , 'b c d h w -> b (c d) h w' )
__lowerCAmelCase: str = bits * 2 - 1
return bits
def _a ( SCREAMING_SNAKE_CASE : Optional[Any] , SCREAMING_SNAKE_CASE : int=BITS ) -> List[str]:
"""simple docstring"""
__lowerCAmelCase: str = x.device
__lowerCAmelCase: int = (x > 0).int()
__lowerCAmelCase: Tuple = 2 ** torch.arange(bits - 1 , -1 , -1 , device=SCREAMING_SNAKE_CASE , dtype=torch.intaa )
__lowerCAmelCase: Union[str, Any] = rearrange(SCREAMING_SNAKE_CASE , 'd -> d 1 1' )
__lowerCAmelCase: int = rearrange(SCREAMING_SNAKE_CASE , 'b (c d) h w -> b c d h w' , d=8 )
__lowerCAmelCase: List[Any] = reduce(x * mask , 'b c d h w -> b c h w' , 'sum' )
return (dec / 2_55).clamp(0.0 , 1.0 )
def _a ( self : List[str] , SCREAMING_SNAKE_CASE : torch.FloatTensor , SCREAMING_SNAKE_CASE : int , SCREAMING_SNAKE_CASE : torch.FloatTensor , SCREAMING_SNAKE_CASE : float = 0.0 , SCREAMING_SNAKE_CASE : bool = True , SCREAMING_SNAKE_CASE : Optional[Any]=None , SCREAMING_SNAKE_CASE : bool = True , ) -> Dict:
"""simple docstring"""
if self.num_inference_steps is None:
raise ValueError(
'Number of inference steps is \'None\', you need to run \'set_timesteps\' after creating the scheduler' )
# See formulas (12) and (16) of DDIM paper https://arxiv.org/pdf/2010.02502.pdf
# Ideally, read DDIM paper in-detail understanding
# Notation (<variable name> -> <name in paper>
# - pred_noise_t -> e_theta(x_t, t)
# - pred_original_sample -> f_theta(x_t, t) or x_0
# - std_dev_t -> sigma_t
# - eta -> η
# - pred_sample_direction -> "direction pointing to x_t"
# - pred_prev_sample -> "x_t-1"
# 1. get previous step value (=t-1)
__lowerCAmelCase: List[Any] = timestep - self.config.num_train_timesteps // self.num_inference_steps
# 2. compute alphas, betas
__lowerCAmelCase: List[Any] = self.alphas_cumprod[timestep]
__lowerCAmelCase: Dict = self.alphas_cumprod[prev_timestep] if prev_timestep >= 0 else self.final_alpha_cumprod
__lowerCAmelCase: int = 1 - alpha_prod_t
# 3. compute predicted original sample from predicted noise also called
# "predicted x_0" of formula (12) from https://arxiv.org/pdf/2010.02502.pdf
__lowerCAmelCase: Dict = (sample - beta_prod_t ** 0.5 * model_output) / alpha_prod_t ** 0.5
# 4. Clip "predicted x_0"
__lowerCAmelCase: Union[str, Any] = self.bit_scale
if self.config.clip_sample:
__lowerCAmelCase: List[Any] = torch.clamp(SCREAMING_SNAKE_CASE , -scale , SCREAMING_SNAKE_CASE )
# 5. compute variance: "sigma_t(η)" -> see formula (16)
# σ_t = sqrt((1 − α_t−1)/(1 − α_t)) * sqrt(1 − α_t/α_t−1)
__lowerCAmelCase: int = self._get_variance(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE )
__lowerCAmelCase: int = eta * variance ** 0.5
if use_clipped_model_output:
# the model_output is always re-derived from the clipped x_0 in Glide
__lowerCAmelCase: Optional[Any] = (sample - alpha_prod_t ** 0.5 * pred_original_sample) / beta_prod_t ** 0.5
# 6. compute "direction pointing to x_t" of formula (12) from https://arxiv.org/pdf/2010.02502.pdf
__lowerCAmelCase: Tuple = (1 - alpha_prod_t_prev - std_dev_t**2) ** 0.5 * model_output
# 7. compute x_t without "random noise" of formula (12) from https://arxiv.org/pdf/2010.02502.pdf
__lowerCAmelCase: Tuple = alpha_prod_t_prev ** 0.5 * pred_original_sample + pred_sample_direction
if eta > 0:
# randn_like does not support generator https://github.com/pytorch/pytorch/issues/27072
__lowerCAmelCase: Union[str, Any] = model_output.device if torch.is_tensor(SCREAMING_SNAKE_CASE ) else 'cpu'
__lowerCAmelCase: Any = torch.randn(model_output.shape , dtype=model_output.dtype , generator=SCREAMING_SNAKE_CASE ).to(SCREAMING_SNAKE_CASE )
__lowerCAmelCase: Optional[Any] = self._get_variance(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE ) ** 0.5 * eta * noise
__lowerCAmelCase: int = prev_sample + variance
if not return_dict:
return (prev_sample,)
return DDIMSchedulerOutput(prev_sample=SCREAMING_SNAKE_CASE , pred_original_sample=SCREAMING_SNAKE_CASE )
def _a ( self : List[str] , SCREAMING_SNAKE_CASE : torch.FloatTensor , SCREAMING_SNAKE_CASE : int , SCREAMING_SNAKE_CASE : torch.FloatTensor , SCREAMING_SNAKE_CASE : Tuple="epsilon" , SCREAMING_SNAKE_CASE : Union[str, Any]=None , SCREAMING_SNAKE_CASE : bool = True , ) -> Union[str, Any]:
"""simple docstring"""
__lowerCAmelCase: List[Any] = timestep
if model_output.shape[1] == sample.shape[1] * 2 and self.variance_type in ["learned", "learned_range"]:
__lowerCAmelCase: Optional[Any] = torch.split(SCREAMING_SNAKE_CASE , sample.shape[1] , dim=1 )
else:
__lowerCAmelCase: Tuple = None
# 1. compute alphas, betas
__lowerCAmelCase: Optional[int] = self.alphas_cumprod[t]
__lowerCAmelCase: int = self.alphas_cumprod[t - 1] if t > 0 else self.one
__lowerCAmelCase: List[str] = 1 - alpha_prod_t
__lowerCAmelCase: List[str] = 1 - alpha_prod_t_prev
# 2. compute predicted original sample from predicted noise also called
# "predicted x_0" of formula (15) from https://arxiv.org/pdf/2006.11239.pdf
if prediction_type == "epsilon":
__lowerCAmelCase: int = (sample - beta_prod_t ** 0.5 * model_output) / alpha_prod_t ** 0.5
elif prediction_type == "sample":
__lowerCAmelCase: Dict = model_output
else:
raise ValueError(f'''Unsupported prediction_type {prediction_type}.''' )
# 3. Clip "predicted x_0"
__lowerCAmelCase: List[str] = self.bit_scale
if self.config.clip_sample:
__lowerCAmelCase: Union[str, Any] = torch.clamp(SCREAMING_SNAKE_CASE , -scale , SCREAMING_SNAKE_CASE )
# 4. Compute coefficients for pred_original_sample x_0 and current sample x_t
# See formula (7) from https://arxiv.org/pdf/2006.11239.pdf
__lowerCAmelCase: Union[str, Any] = (alpha_prod_t_prev ** 0.5 * self.betas[t]) / beta_prod_t
__lowerCAmelCase: str = self.alphas[t] ** 0.5 * beta_prod_t_prev / beta_prod_t
# 5. Compute predicted previous sample µ_t
# See formula (7) from https://arxiv.org/pdf/2006.11239.pdf
__lowerCAmelCase: Optional[int] = pred_original_sample_coeff * pred_original_sample + current_sample_coeff * sample
# 6. Add noise
__lowerCAmelCase: Tuple = 0
if t > 0:
__lowerCAmelCase: Optional[Any] = torch.randn(
model_output.size() , dtype=model_output.dtype , layout=model_output.layout , generator=SCREAMING_SNAKE_CASE ).to(model_output.device )
__lowerCAmelCase: str = (self._get_variance(SCREAMING_SNAKE_CASE , predicted_variance=SCREAMING_SNAKE_CASE ) ** 0.5) * noise
__lowerCAmelCase: Any = pred_prev_sample + variance
if not return_dict:
return (pred_prev_sample,)
return DDPMSchedulerOutput(prev_sample=SCREAMING_SNAKE_CASE , pred_original_sample=SCREAMING_SNAKE_CASE )
class A_ ( lowerCAmelCase_ ):
def __init__( self : Union[str, Any] , UpperCAmelCase : UNetaDConditionModel , UpperCAmelCase : Union[DDIMScheduler, DDPMScheduler] , UpperCAmelCase : Optional[float] = 1.0 , ) -> int:
super().__init__()
__lowerCAmelCase: Optional[int] = bit_scale
__lowerCAmelCase: int = (
ddim_bit_scheduler_step if isinstance(_snake_case , _snake_case ) else ddpm_bit_scheduler_step
)
self.register_modules(unet=_snake_case , scheduler=_snake_case )
@torch.no_grad()
def __call__( self : List[str] , UpperCAmelCase : Optional[int] = 2_5_6 , UpperCAmelCase : Optional[int] = 2_5_6 , UpperCAmelCase : Optional[int] = 5_0 , UpperCAmelCase : Optional[torch.Generator] = None , UpperCAmelCase : Optional[int] = 1 , UpperCAmelCase : Optional[str] = "pil" , UpperCAmelCase : bool = True , **UpperCAmelCase : Tuple , ) -> Union[Tuple, ImagePipelineOutput]:
__lowerCAmelCase: int = torch.randn(
(batch_size, self.unet.config.in_channels, height, width) , generator=_snake_case , )
__lowerCAmelCase: Tuple = decimal_to_bits(_snake_case ) * self.bit_scale
__lowerCAmelCase: Optional[Any] = latents.to(self.device )
self.scheduler.set_timesteps(_snake_case )
for t in self.progress_bar(self.scheduler.timesteps ):
# predict the noise residual
__lowerCAmelCase: Union[str, Any] = self.unet(_snake_case , _snake_case ).sample
# compute the previous noisy sample x_t -> x_t-1
__lowerCAmelCase: int = self.scheduler.step(_snake_case , _snake_case , _snake_case ).prev_sample
__lowerCAmelCase: Optional[int] = bits_to_decimal(_snake_case )
if output_type == "pil":
__lowerCAmelCase: Any = self.numpy_to_pil(_snake_case )
if not return_dict:
return (image,)
return ImagePipelineOutput(images=_snake_case )
| 322 |
from collections import deque
from .hash_table import HashTable
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
def __init__( self : Optional[Any], *_snake_case : Optional[Any], **_snake_case : List[Any] ) ->Optional[int]:
super().__init__(*_snake_case, **_snake_case )
def lowercase_ ( self : Optional[Any], _snake_case : Tuple, _snake_case : Dict ) ->Dict:
snake_case__ : int = deque([] ) if self.values[key] is None else self.values[key]
self.values[key].appendleft(_snake_case )
snake_case__ : Dict = self.values[key]
def lowercase_ ( self : Any ) ->Optional[Any]:
return (
sum(self.charge_factor - len(_snake_case ) for slot in self.values )
/ self.size_table
* self.charge_factor
)
def lowercase_ ( self : Union[str, Any], _snake_case : str, _snake_case : Optional[int]=None ) ->Optional[Any]:
if not (
len(self.values[key] ) == self.charge_factor and self.values.count(_snake_case ) == 0
):
return key
return super()._collision_resolution(_snake_case, _snake_case )
| 277 | 0 |
import argparse
import json
import os
import torch
from torch import nn
from transformers import NllbMoeConfig, NllbMoeModel
from transformers.modeling_utils import dtype_byte_size
from transformers.utils import WEIGHTS_INDEX_NAME, WEIGHTS_NAME
def _A ( SCREAMING_SNAKE_CASE : List[str] ):
"""simple docstring"""
a__ : Tuple =[
'encoder.version',
'decoder.version',
'model.encoder.version',
'model.decoder.version',
'decoder.output_projection.weight',
'_float_tensor',
'encoder.embed_positions._float_tensor',
'decoder.embed_positions._float_tensor',
]
for k in ignore_keys:
state_dict.pop(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE )
def _A ( SCREAMING_SNAKE_CASE : str ):
"""simple docstring"""
a__ : Union[str, Any] =emb.weight.shape
a__ : str =nn.Linear(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE , bias=SCREAMING_SNAKE_CASE )
a__ : str =emb.weight.data
return lin_layer
def _A ( SCREAMING_SNAKE_CASE : Optional[int] , SCREAMING_SNAKE_CASE : Union[str, Any]=None ):
"""simple docstring"""
a__ : Any ={}
for old_key in state_dict.keys():
a__ : Tuple =old_key
if "moe_layer.experts." in key:
if expert_idx is not None:
a__ : int =key.replace("moe_layer.experts.0" , f'''ffn.experts.expert_{expert_idx}''' )
else:
a__ : Any =key.replace("moe_layer.experts." , "ffn.experts.expert_" )
if "gate" in key:
a__ : Dict =key.replace(".moe_layer.gate.wg" , ".ffn.router.classifier" )
if "fc2" and "experts" not in key:
a__ : str =key.replace(".fc2." , ".ffn.fc2." )
if "fc1" and "experts" not in key:
a__ : str =key.replace(".fc1." , ".ffn.fc1." )
if ".encoder_attn." in key:
a__ : Tuple =key.replace(".encoder_attn." , ".cross_attention." )
if "encoder_attn_layer_norm" in key:
a__ : Tuple =key.replace("encoder_attn_layer_norm" , "cross_attention_layer_norm" )
if "final_layer_norm" in key:
a__ : Optional[int] =key.replace("final_layer_norm" , "ff_layer_norm" )
a__ : Dict =state_dict[old_key]
return new_dict
def _A ( SCREAMING_SNAKE_CASE : List[Any] , SCREAMING_SNAKE_CASE : Tuple , SCREAMING_SNAKE_CASE : List[Any] , SCREAMING_SNAKE_CASE : List[str] , SCREAMING_SNAKE_CASE : str = WEIGHTS_NAME ):
"""simple docstring"""
a__ : Dict =[]
a__ : str =0
os.makedirs(SCREAMING_SNAKE_CASE , exist_ok=SCREAMING_SNAKE_CASE )
for expert in range(SCREAMING_SNAKE_CASE ):
a__ : Tuple =switch_checkpoint_path + f'''-rank-{expert}.pt'''
if os.path.isfile(SCREAMING_SNAKE_CASE ):
a__ : Optional[Any] =torch.load(SCREAMING_SNAKE_CASE )['model']
remove_ignore_keys_(SCREAMING_SNAKE_CASE )
a__ : Optional[Any] =rename_fairseq_keys(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE )
a__ : Dict =os.path.join(
SCREAMING_SNAKE_CASE , weights_name.replace(".bin" , f'''-{len(SCREAMING_SNAKE_CASE )+1:05d}-of-???.bin''' ) )
torch.save(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE )
sharded_state_dicts.append(expert_state.keys() )
total_size += sum([value.numel() for key, value in expert_state.items()] ) * dtype_byte_size(
expert_state[list(SCREAMING_SNAKE_CASE )[0]].dtype )
# Add the last block
a__ : Tuple =os.path.join(SCREAMING_SNAKE_CASE , weights_name.replace(".bin" , f'''-{len(SCREAMING_SNAKE_CASE )+1:05d}-of-???.bin''' ) )
a__ : Union[str, Any] =torch.load(switch_checkpoint_path + "-shared.pt" )['model']
remove_ignore_keys_(SCREAMING_SNAKE_CASE )
a__ : str =rename_fairseq_keys(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE )
a__ : Any =shared_weights['decoder.embed_tokens.weight']
sharded_state_dicts.append(shared_weights.keys() )
# If we only have the shared weights (dummy model/experts saved on the same file)
if len(SCREAMING_SNAKE_CASE ) == 1:
a__ : Any =os.path.join(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE )
torch.save(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE )
return {weights_name: sharded_state_dicts[0]}, None
else:
torch.save(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE )
# Otherwise, let's build the index
a__ : Tuple ={}
for idx, shard in enumerate(SCREAMING_SNAKE_CASE ):
a__ : Optional[int] =weights_name.replace(".bin" , f'''-{idx+1:05d}-of-{len(SCREAMING_SNAKE_CASE ):05d}.bin''' )
a__ : List[Any] =os.path.join(SCREAMING_SNAKE_CASE , weights_name.replace(".bin" , f'''-{idx+1:05d}-of-???.bin''' ) )
os.rename(SCREAMING_SNAKE_CASE , os.path.join(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE ) )
for key in shard:
a__ : Any =shard_file
# Add the metadata
a__ : int ={'total_size': total_size}
a__ : Dict ={'metadata': metadata, 'weight_map': weight_map}
with open(os.path.join(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE ) , "w" , encoding="utf-8" ) as f:
a__ : Any =json.dumps(SCREAMING_SNAKE_CASE , indent=2 , sort_keys=SCREAMING_SNAKE_CASE ) + '\n'
f.write(SCREAMING_SNAKE_CASE )
return metadata, index
if __name__ == "__main__":
UpperCAmelCase : int = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"""--nllb_moe_checkpoint_path""",
default="""/home/arthur_huggingface_co/fairseq/weights/checkpoints/model_moe_54b/checkpoint_2_300000""",
type=str,
required=False,
help="""Path to a directory containing a folder per layer. Follows the original Google format.""",
)
parser.add_argument("""--dtype""", default="""float32""", type=str, required=False, help="""dtype of the saved model""")
parser.add_argument(
"""--pytorch_dump_folder_path""",
default="""/home/arthur_huggingface_co/fairseq/weights/checkpoints/hf-converted-moe-54b""",
type=str,
required=False,
help="""Path to the output pytorch model.""",
)
UpperCAmelCase : Optional[Any] = parser.parse_args()
UpperCAmelCase : Optional[Any] = shard_on_the_fly(
args.nllb_moe_checkpoint_path,
args.pytorch_dump_folder_path,
128,
args.dtype,
)
UpperCAmelCase : List[str] = NllbMoeConfig.from_pretrained(
"""facebook/nllb-200-3.3B""", encoder_sparse_step=4, decoder_sparse_step=4, num_experts=128
)
config.save_pretrained(args.pytorch_dump_folder_path)
UpperCAmelCase : int = NllbMoeModel.from_pretrained(args.pytorch_dump_folder_path)
print("""Done""")
model.save_pretrained(args.pytorch_dump_folder_path)
| 95 |
def lowercase_ (A : Union[str, Any] , A : List[str] , A : int , A : Optional[int] ):
global f # a global dp table for knapsack
if f[i][j] < 0:
if j < wt[i - 1]:
snake_case__ : Union[str, Any] = mf_knapsack(i - 1 , A , A , A )
else:
snake_case__ : Any = max(
mf_knapsack(i - 1 , A , A , A ) , mf_knapsack(i - 1 , A , A , j - wt[i - 1] ) + val[i - 1] , )
snake_case__ : Optional[int] = val
return f[i][j]
def lowercase_ (A : Optional[int] , A : Union[str, Any] , A : str , A : Dict ):
snake_case__ : int = [[0] * (w + 1) for _ in range(n + 1 )]
for i in range(1 , n + 1 ):
for w_ in range(1 , w + 1 ):
if wt[i - 1] <= w_:
snake_case__ : Union[str, Any] = max(val[i - 1] + dp[i - 1][w_ - wt[i - 1]] , dp[i - 1][w_] )
else:
snake_case__ : str = dp[i - 1][w_]
return dp[n][w_], dp
def lowercase_ (A : int , A : list , A : list ):
if not (isinstance(A , (list, tuple) ) and isinstance(A , (list, tuple) )):
raise ValueError(
'Both the weights and values vectors must be either lists or tuples' )
snake_case__ : Dict = len(A )
if num_items != len(A ):
snake_case__ : str = (
'The number of weights must be the same as the number of values.\n'
F'''But got {num_items} weights and {len(A )} values'''
)
raise ValueError(A )
for i in range(A ):
if not isinstance(wt[i] , A ):
snake_case__ : Optional[int] = (
'All weights must be integers but got weight of '
F'''type {type(wt[i] )} at index {i}'''
)
raise TypeError(A )
snake_case__ , snake_case__ : Optional[int] = knapsack(A , A , A , A )
snake_case__ : set = set()
_construct_solution(A , A , A , A , A )
return optimal_val, example_optional_set
def lowercase_ (A : list , A : list , A : int , A : int , A : set ):
# for the current item i at a maximum weight j to be part of an optimal subset,
# the optimal value at (i, j) must be greater than the optimal value at (i-1, j).
# where i - 1 means considering only the previous items at the given maximum weight
if i > 0 and j > 0:
if dp[i - 1][j] == dp[i][j]:
_construct_solution(A , A , i - 1 , A , A )
else:
optimal_set.add(A )
_construct_solution(A , A , i - 1 , j - wt[i - 1] , A )
if __name__ == "__main__":
a_ :Any = [3, 2, 4, 4]
a_ :List[Any] = [4, 3, 2, 3]
a_ :Union[str, Any] = 4
a_ :List[str] = 6
a_ :Union[str, Any] = [[0] * (w + 1)] + [[0] + [-1] * (w + 1) for _ in range(n + 1)]
a_ , a_ :List[Any] = knapsack(w, wt, val, n)
print(optimal_solution)
print(mf_knapsack(n, wt, val, w)) # switched the n and w
# testing the dynamic programming problem with example
# the optimal subset for the above example are items 3 and 4
a_ , a_ :Any = knapsack_with_example_solution(w, wt, val)
assert optimal_solution == 8
assert optimal_subset == {3, 4}
print("optimal_value = ", optimal_solution)
print("An optimal subset corresponding to the optimal value", optimal_subset)
| 277 | 0 |
from collections import OrderedDict
from typing import TYPE_CHECKING, Any, Mapping, Optional, Union
from ...configuration_utils import PretrainedConfig
from ...onnx import OnnxConfig
from ...utils import logging
if TYPE_CHECKING:
from ... import FeatureExtractionMixin, PreTrainedTokenizerBase, TensorType
__a = logging.get_logger(__name__)
__a = {
"microsoft/deberta-v2-xlarge": "https://huggingface.co/microsoft/deberta-v2-xlarge/resolve/main/config.json",
"microsoft/deberta-v2-xxlarge": "https://huggingface.co/microsoft/deberta-v2-xxlarge/resolve/main/config.json",
"microsoft/deberta-v2-xlarge-mnli": (
"https://huggingface.co/microsoft/deberta-v2-xlarge-mnli/resolve/main/config.json"
),
"microsoft/deberta-v2-xxlarge-mnli": (
"https://huggingface.co/microsoft/deberta-v2-xxlarge-mnli/resolve/main/config.json"
),
}
class lowercase__( lowerCAmelCase_ ):
"""simple docstring"""
a :Any = 'deberta-v2'
def __init__( self : Union[str, Any] , SCREAMING_SNAKE_CASE_ : Dict=1_2_8_1_0_0 , SCREAMING_SNAKE_CASE_ : Any=1_5_3_6 , SCREAMING_SNAKE_CASE_ : Tuple=2_4 , SCREAMING_SNAKE_CASE_ : int=2_4 , SCREAMING_SNAKE_CASE_ : Optional[int]=6_1_4_4 , SCREAMING_SNAKE_CASE_ : Optional[int]="gelu" , SCREAMING_SNAKE_CASE_ : Optional[int]=0.1 , SCREAMING_SNAKE_CASE_ : List[str]=0.1 , SCREAMING_SNAKE_CASE_ : str=5_1_2 , SCREAMING_SNAKE_CASE_ : Optional[int]=0 , SCREAMING_SNAKE_CASE_ : Optional[int]=0.02 , SCREAMING_SNAKE_CASE_ : Dict=1e-7 , SCREAMING_SNAKE_CASE_ : int=False , SCREAMING_SNAKE_CASE_ : Any=-1 , SCREAMING_SNAKE_CASE_ : List[str]=0 , SCREAMING_SNAKE_CASE_ : Tuple=True , SCREAMING_SNAKE_CASE_ : Any=None , SCREAMING_SNAKE_CASE_ : Union[str, Any]=0 , SCREAMING_SNAKE_CASE_ : Tuple="gelu" , **SCREAMING_SNAKE_CASE_ : Union[str, Any] , ) -> Optional[int]:
super().__init__(**_snake_case )
lowercase_ = hidden_size
lowercase_ = num_hidden_layers
lowercase_ = num_attention_heads
lowercase_ = intermediate_size
lowercase_ = hidden_act
lowercase_ = hidden_dropout_prob
lowercase_ = attention_probs_dropout_prob
lowercase_ = max_position_embeddings
lowercase_ = type_vocab_size
lowercase_ = initializer_range
lowercase_ = relative_attention
lowercase_ = max_relative_positions
lowercase_ = pad_token_id
lowercase_ = position_biased_input
# Backwards compatibility
if type(_snake_case ) == str:
lowercase_ = [x.strip() for x in pos_att_type.lower().split('''|''' )]
lowercase_ = pos_att_type
lowercase_ = vocab_size
lowercase_ = layer_norm_eps
lowercase_ = kwargs.get('''pooler_hidden_size''' , _snake_case )
lowercase_ = pooler_dropout
lowercase_ = pooler_hidden_act
class lowercase__( lowerCAmelCase_ ):
"""simple docstring"""
@property
def _lowercase ( self : Optional[int] ) -> Mapping[str, Mapping[int, str]]:
if self.task == "multiple-choice":
lowercase_ = {0: 'batch', 1: 'choice', 2: 'sequence'}
else:
lowercase_ = {0: 'batch', 1: 'sequence'}
if self._config.type_vocab_size > 0:
return OrderedDict(
[('''input_ids''', dynamic_axis), ('''attention_mask''', dynamic_axis), ('''token_type_ids''', dynamic_axis)] )
else:
return OrderedDict([('''input_ids''', dynamic_axis), ('''attention_mask''', dynamic_axis)] )
@property
def _lowercase ( self : Dict ) -> int:
return 1_2
def _lowercase ( self : Tuple , SCREAMING_SNAKE_CASE_ : Union["PreTrainedTokenizerBase", "FeatureExtractionMixin"] , SCREAMING_SNAKE_CASE_ : int = -1 , SCREAMING_SNAKE_CASE_ : int = -1 , SCREAMING_SNAKE_CASE_ : int = -1 , SCREAMING_SNAKE_CASE_ : bool = False , SCREAMING_SNAKE_CASE_ : Optional["TensorType"] = None , SCREAMING_SNAKE_CASE_ : int = 3 , SCREAMING_SNAKE_CASE_ : int = 4_0 , SCREAMING_SNAKE_CASE_ : int = 4_0 , SCREAMING_SNAKE_CASE_ : "PreTrainedTokenizerBase" = None , ) -> Mapping[str, Any]:
lowercase_ = super().generate_dummy_inputs(preprocessor=_snake_case , framework=_snake_case )
if self._config.type_vocab_size == 0 and "token_type_ids" in dummy_inputs:
del dummy_inputs["token_type_ids"]
return dummy_inputs
| 30 |
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_flax_available, is_torch_available
a_ :int = {
"configuration_longt5": ["LONGT5_PRETRAINED_CONFIG_ARCHIVE_MAP", "LongT5Config", "LongT5OnnxConfig"],
}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :List[str] = [
"LONGT5_PRETRAINED_MODEL_ARCHIVE_LIST",
"LongT5EncoderModel",
"LongT5ForConditionalGeneration",
"LongT5Model",
"LongT5PreTrainedModel",
]
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :int = [
"FlaxLongT5ForConditionalGeneration",
"FlaxLongT5Model",
"FlaxLongT5PreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_longta import LONGT5_PRETRAINED_CONFIG_ARCHIVE_MAP, LongTaConfig, LongTaOnnxConfig
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_longta import (
LONGT5_PRETRAINED_MODEL_ARCHIVE_LIST,
LongTaEncoderModel,
LongTaForConditionalGeneration,
LongTaModel,
LongTaPreTrainedModel,
)
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_flax_longta import (
FlaxLongTaForConditionalGeneration,
FlaxLongTaModel,
FlaxLongTaPreTrainedModel,
)
else:
import sys
a_ :Optional[int] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)
| 277 | 0 |
"""simple docstring"""
import warnings
from functools import wraps
from typing import Callable
def lowercase ( lowerCAmelCase__ : Callable ) -> Tuple:
@wraps(lowerCAmelCase__ )
def _inner_fn(*lowerCAmelCase__ : Optional[Any] , **lowerCAmelCase__ : str ):
warnings.warn(
(f'''\'{fn.__name__}\' is experimental and might be subject to breaking changes in the future.''') , lowerCAmelCase__ , )
return fn(*lowerCAmelCase__ , **lowerCAmelCase__ )
return _inner_fn
| 45 |
import argparse
import json
import os
import torch
from torch import nn
from transformers import NllbMoeConfig, NllbMoeModel
from transformers.modeling_utils import dtype_byte_size
from transformers.utils import WEIGHTS_INDEX_NAME, WEIGHTS_NAME
def lowercase_ (A : List[str] ):
snake_case__ : Tuple = [
'encoder.version',
'decoder.version',
'model.encoder.version',
'model.decoder.version',
'decoder.output_projection.weight',
'_float_tensor',
'encoder.embed_positions._float_tensor',
'decoder.embed_positions._float_tensor',
]
for k in ignore_keys:
state_dict.pop(A , A )
def lowercase_ (A : str ):
snake_case__ , snake_case__ : Union[str, Any] = emb.weight.shape
snake_case__ : str = nn.Linear(A , A , bias=A )
snake_case__ : str = emb.weight.data
return lin_layer
def lowercase_ (A : Optional[int] , A : Union[str, Any]=None ):
snake_case__ : Any = {}
for old_key in state_dict.keys():
snake_case__ : Tuple = old_key
if "moe_layer.experts." in key:
if expert_idx is not None:
snake_case__ : int = key.replace('moe_layer.experts.0' , F'''ffn.experts.expert_{expert_idx}''' )
else:
snake_case__ : Any = key.replace('moe_layer.experts.' , 'ffn.experts.expert_' )
if "gate" in key:
snake_case__ : Dict = key.replace('.moe_layer.gate.wg' , '.ffn.router.classifier' )
if "fc2" and "experts" not in key:
snake_case__ : str = key.replace('.fc2.' , '.ffn.fc2.' )
if "fc1" and "experts" not in key:
snake_case__ : str = key.replace('.fc1.' , '.ffn.fc1.' )
if ".encoder_attn." in key:
snake_case__ : Tuple = key.replace('.encoder_attn.' , '.cross_attention.' )
if "encoder_attn_layer_norm" in key:
snake_case__ : Tuple = key.replace('encoder_attn_layer_norm' , 'cross_attention_layer_norm' )
if "final_layer_norm" in key:
snake_case__ : Optional[int] = key.replace('final_layer_norm' , 'ff_layer_norm' )
snake_case__ : Dict = state_dict[old_key]
return new_dict
def lowercase_ (A : List[Any] , A : Tuple , A : List[Any] , A : List[str] , A : str = WEIGHTS_NAME ):
snake_case__ : Dict = []
snake_case__ : str = 0
os.makedirs(A , exist_ok=A )
for expert in range(A ):
snake_case__ : Tuple = switch_checkpoint_path + F'''-rank-{expert}.pt'''
if os.path.isfile(A ):
snake_case__ : Optional[Any] = torch.load(A )['model']
remove_ignore_keys_(A )
snake_case__ : Optional[Any] = rename_fairseq_keys(A , A )
snake_case__ : Dict = os.path.join(
A , weights_name.replace('.bin' , F'''-{len(A )+1:05d}-of-???.bin''' ) )
torch.save(A , A )
sharded_state_dicts.append(expert_state.keys() )
total_size += sum([value.numel() for key, value in expert_state.items()] ) * dtype_byte_size(
expert_state[list(A )[0]].dtype )
# Add the last block
snake_case__ : Tuple = os.path.join(A , weights_name.replace('.bin' , F'''-{len(A )+1:05d}-of-???.bin''' ) )
snake_case__ : Union[str, Any] = torch.load(switch_checkpoint_path + '-shared.pt' )['model']
remove_ignore_keys_(A )
snake_case__ : str = rename_fairseq_keys(A , A )
snake_case__ : Any = shared_weights['decoder.embed_tokens.weight']
sharded_state_dicts.append(shared_weights.keys() )
# If we only have the shared weights (dummy model/experts saved on the same file)
if len(A ) == 1:
snake_case__ : Any = os.path.join(A , A )
torch.save(A , A )
return {weights_name: sharded_state_dicts[0]}, None
else:
torch.save(A , A )
# Otherwise, let's build the index
snake_case__ : Tuple = {}
for idx, shard in enumerate(A ):
snake_case__ : Optional[int] = weights_name.replace('.bin' , F'''-{idx+1:05d}-of-{len(A ):05d}.bin''' )
snake_case__ : List[Any] = os.path.join(A , weights_name.replace('.bin' , F'''-{idx+1:05d}-of-???.bin''' ) )
os.rename(A , os.path.join(A , A ) )
for key in shard:
snake_case__ : Any = shard_file
# Add the metadata
snake_case__ : int = {'total_size': total_size}
snake_case__ : Dict = {'metadata': metadata, 'weight_map': weight_map}
with open(os.path.join(A , A ) , 'w' , encoding='utf-8' ) as f:
snake_case__ : Any = json.dumps(A , indent=2 , sort_keys=A ) + '\n'
f.write(A )
return metadata, index
if __name__ == "__main__":
a_ :int = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--nllb_moe_checkpoint_path",
default="/home/arthur_huggingface_co/fairseq/weights/checkpoints/model_moe_54b/checkpoint_2_300000",
type=str,
required=False,
help="Path to a directory containing a folder per layer. Follows the original Google format.",
)
parser.add_argument("--dtype", default="float32", type=str, required=False, help="dtype of the saved model")
parser.add_argument(
"--pytorch_dump_folder_path",
default="/home/arthur_huggingface_co/fairseq/weights/checkpoints/hf-converted-moe-54b",
type=str,
required=False,
help="Path to the output pytorch model.",
)
a_ :Optional[Any] = parser.parse_args()
a_ , a_ :Optional[Any] = shard_on_the_fly(
args.nllb_moe_checkpoint_path,
args.pytorch_dump_folder_path,
128,
args.dtype,
)
a_ :List[str] = NllbMoeConfig.from_pretrained(
"facebook/nllb-200-3.3B", encoder_sparse_step=4, decoder_sparse_step=4, num_experts=128
)
config.save_pretrained(args.pytorch_dump_folder_path)
a_ :int = NllbMoeModel.from_pretrained(args.pytorch_dump_folder_path)
print("Done")
model.save_pretrained(args.pytorch_dump_folder_path)
| 277 | 0 |
'''simple docstring'''
# Copyright 2023 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from typing import TYPE_CHECKING
# rely on isort to merge the imports
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tokenizers_available, is_torch_available
_lowerCAmelCase = {"configuration_mra": ["MRA_PRETRAINED_CONFIG_ARCHIVE_MAP", "MraConfig"]}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
_lowerCAmelCase = [
"MRA_PRETRAINED_MODEL_ARCHIVE_LIST",
"MraForMaskedLM",
"MraForMultipleChoice",
"MraForQuestionAnswering",
"MraForSequenceClassification",
"MraForTokenClassification",
"MraLayer",
"MraModel",
"MraPreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_mra import MRA_PRETRAINED_CONFIG_ARCHIVE_MAP, MraConfig
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_mra import (
MRA_PRETRAINED_MODEL_ARCHIVE_LIST,
MraForMaskedLM,
MraForMultipleChoice,
MraForQuestionAnswering,
MraForSequenceClassification,
MraForTokenClassification,
MraLayer,
MraModel,
MraPreTrainedModel,
)
else:
import sys
_lowerCAmelCase = _LazyModule(__name__, globals()['''__file__'''], _import_structure)
| 298 |
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_sentencepiece_available,
is_tokenizers_available,
is_torch_available,
)
a_ :Optional[Any] = {"configuration_reformer": ["REFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP", "ReformerConfig"]}
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :str = ["ReformerTokenizer"]
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :int = ["ReformerTokenizerFast"]
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :List[str] = [
"REFORMER_PRETRAINED_MODEL_ARCHIVE_LIST",
"ReformerAttention",
"ReformerForMaskedLM",
"ReformerForQuestionAnswering",
"ReformerForSequenceClassification",
"ReformerLayer",
"ReformerModel",
"ReformerModelWithLMHead",
"ReformerPreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_reformer import REFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, ReformerConfig
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_reformer import ReformerTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_reformer_fast import ReformerTokenizerFast
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_reformer import (
REFORMER_PRETRAINED_MODEL_ARCHIVE_LIST,
ReformerAttention,
ReformerForMaskedLM,
ReformerForQuestionAnswering,
ReformerForSequenceClassification,
ReformerLayer,
ReformerModel,
ReformerModelWithLMHead,
ReformerPreTrainedModel,
)
else:
import sys
a_ :Optional[Any] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)
| 277 | 0 |
from math import sqrt
def lowerCAmelCase_ ( snake_case_ ):
assert isinstance(snake_case_,snake_case_ ) and (
number >= 0
), "'number' must been an int and positive"
_A : Optional[Any] = True
# 0 and 1 are none primes.
if number <= 1:
_A : Tuple = False
for divisor in range(2,int(round(sqrt(snake_case_ ) ) ) + 1 ):
# if 'number' divisible by 'divisor' then sets 'status'
# of false and break up the loop.
if number % divisor == 0:
_A : Optional[Any] = False
break
# precondition
assert isinstance(snake_case_,snake_case_ ), "'status' must been from type bool"
return status
def lowerCAmelCase_ ( snake_case_ ):
assert isinstance(snake_case_,snake_case_ ) and (n > 2), "'N' must been an int and > 2"
# beginList: contains all natural numbers from 2 up to N
_A : Optional[Any] = list(range(2,n + 1 ) )
_A : List[Any] = [] # this list will be returns.
# actual sieve of erathostenes
for i in range(len(snake_case_ ) ):
for j in range(i + 1,len(snake_case_ ) ):
if (begin_list[i] != 0) and (begin_list[j] % begin_list[i] == 0):
_A : int = 0
# filters actual prime numbers.
_A : Union[str, Any] = [x for x in begin_list if x != 0]
# precondition
assert isinstance(snake_case_,snake_case_ ), "'ans' must been from type list"
return ans
def lowerCAmelCase_ ( snake_case_ ):
assert isinstance(snake_case_,snake_case_ ) and (n > 2), "'N' must been an int and > 2"
_A : int = []
# iterates over all numbers between 2 up to N+1
# if a number is prime then appends to list 'ans'
for number in range(2,n + 1 ):
if is_prime(snake_case_ ):
ans.append(snake_case_ )
# precondition
assert isinstance(snake_case_,snake_case_ ), "'ans' must been from type list"
return ans
def lowerCAmelCase_ ( snake_case_ ):
assert isinstance(snake_case_,snake_case_ ) and number >= 0, "'number' must been an int and >= 0"
_A : Optional[int] = [] # this list will be returns of the function.
# potential prime number factors.
_A : List[Any] = 2
_A : Dict = number
if number == 0 or number == 1:
ans.append(snake_case_ )
# if 'number' not prime then builds the prime factorization of 'number'
elif not is_prime(snake_case_ ):
while quotient != 1:
if is_prime(snake_case_ ) and (quotient % factor == 0):
ans.append(snake_case_ )
quotient /= factor
else:
factor += 1
else:
ans.append(snake_case_ )
# precondition
assert isinstance(snake_case_,snake_case_ ), "'ans' must been from type list"
return ans
def lowerCAmelCase_ ( snake_case_ ):
assert isinstance(snake_case_,snake_case_ ) and (
number >= 0
), "'number' bust been an int and >= 0"
_A : Union[str, Any] = 0
# prime factorization of 'number'
_A : Tuple = prime_factorization(snake_case_ )
_A : Tuple = max(snake_case_ )
# precondition
assert isinstance(snake_case_,snake_case_ ), "'ans' must been from type int"
return ans
def lowerCAmelCase_ ( snake_case_ ):
assert isinstance(snake_case_,snake_case_ ) and (
number >= 0
), "'number' bust been an int and >= 0"
_A : Tuple = 0
# prime factorization of 'number'
_A : Union[str, Any] = prime_factorization(snake_case_ )
_A : Union[str, Any] = min(snake_case_ )
# precondition
assert isinstance(snake_case_,snake_case_ ), "'ans' must been from type int"
return ans
def lowerCAmelCase_ ( snake_case_ ):
assert isinstance(snake_case_,snake_case_ ), "'number' must been an int"
assert isinstance(number % 2 == 0,snake_case_ ), "compare bust been from type bool"
return number % 2 == 0
def lowerCAmelCase_ ( snake_case_ ):
assert isinstance(snake_case_,snake_case_ ), "'number' must been an int"
assert isinstance(number % 2 != 0,snake_case_ ), "compare bust been from type bool"
return number % 2 != 0
def lowerCAmelCase_ ( snake_case_ ):
assert (
isinstance(snake_case_,snake_case_ ) and (number > 2) and is_even(snake_case_ )
), "'number' must been an int, even and > 2"
_A : Dict = [] # this list will returned
# creates a list of prime numbers between 2 up to 'number'
_A : Union[str, Any] = get_prime_numbers(snake_case_ )
_A : Union[str, Any] = len(snake_case_ )
# run variable for while-loops.
_A : Dict = 0
_A : Union[str, Any] = None
# exit variable. for break up the loops
_A : str = True
while i < len_pn and loop:
_A : Union[str, Any] = i + 1
while j < len_pn and loop:
if prime_numbers[i] + prime_numbers[j] == number:
_A : List[Any] = False
ans.append(prime_numbers[i] )
ans.append(prime_numbers[j] )
j += 1
i += 1
# precondition
assert (
isinstance(snake_case_,snake_case_ )
and (len(snake_case_ ) == 2)
and (ans[0] + ans[1] == number)
and is_prime(ans[0] )
and is_prime(ans[1] )
), "'ans' must contains two primes. And sum of elements must been eq 'number'"
return ans
def lowerCAmelCase_ ( snake_case_,snake_case_ ):
assert (
isinstance(snake_case_,snake_case_ )
and isinstance(snake_case_,snake_case_ )
and (numbera >= 0)
and (numbera >= 0)
), "'number1' and 'number2' must been positive integer."
_A : Optional[Any] = 0
while numbera != 0:
_A : List[Any] = numbera % numbera
_A : List[Any] = numbera
_A : Dict = rest
# precondition
assert isinstance(snake_case_,snake_case_ ) and (
numbera >= 0
), "'number' must been from type int and positive"
return numbera
def lowerCAmelCase_ ( snake_case_,snake_case_ ):
assert (
isinstance(snake_case_,snake_case_ )
and isinstance(snake_case_,snake_case_ )
and (numbera >= 1)
and (numbera >= 1)
), "'number1' and 'number2' must been positive integer."
_A : Optional[int] = 1 # actual answer that will be return.
# for kgV (x,1)
if numbera > 1 and numbera > 1:
# builds the prime factorization of 'number1' and 'number2'
_A : Tuple = prime_factorization(snake_case_ )
_A : Dict = prime_factorization(snake_case_ )
elif numbera == 1 or numbera == 1:
_A : Optional[Any] = []
_A : List[str] = []
_A : Union[str, Any] = max(snake_case_,snake_case_ )
_A : Tuple = 0
_A : Tuple = 0
_A : Optional[int] = [] # captured numbers int both 'primeFac1' and 'primeFac2'
# iterates through primeFac1
for n in prime_fac_a:
if n not in done:
if n in prime_fac_a:
_A : int = prime_fac_a.count(snake_case_ )
_A : Union[str, Any] = prime_fac_a.count(snake_case_ )
for _ in range(max(snake_case_,snake_case_ ) ):
ans *= n
else:
_A : Union[str, Any] = prime_fac_a.count(snake_case_ )
for _ in range(snake_case_ ):
ans *= n
done.append(snake_case_ )
# iterates through primeFac2
for n in prime_fac_a:
if n not in done:
_A : int = prime_fac_a.count(snake_case_ )
for _ in range(snake_case_ ):
ans *= n
done.append(snake_case_ )
# precondition
assert isinstance(snake_case_,snake_case_ ) and (
ans >= 0
), "'ans' must been from type int and positive"
return ans
def lowerCAmelCase_ ( snake_case_ ):
assert isinstance(snake_case_,snake_case_ ) and (n >= 0), "'number' must been a positive int"
_A : Union[str, Any] = 0
_A : Optional[int] = 2 # this variable holds the answer
while index < n:
index += 1
ans += 1 # counts to the next number
# if ans not prime then
# runs to the next prime number.
while not is_prime(snake_case_ ):
ans += 1
# precondition
assert isinstance(snake_case_,snake_case_ ) and is_prime(
snake_case_ ), "'ans' must been a prime number and from type int"
return ans
def lowerCAmelCase_ ( snake_case_,snake_case_ ):
assert (
is_prime(snake_case_ ) and is_prime(snake_case_ ) and (p_number_a < p_number_a)
), "The arguments must been prime numbers and 'pNumber1' < 'pNumber2'"
_A : str = p_number_a + 1 # jump to the next number
_A : Tuple = [] # this list will be returns.
# if number is not prime then
# fetch the next prime number.
while not is_prime(snake_case_ ):
number += 1
while number < p_number_a:
ans.append(snake_case_ )
number += 1
# fetch the next prime number.
while not is_prime(snake_case_ ):
number += 1
# precondition
assert (
isinstance(snake_case_,snake_case_ )
and ans[0] != p_number_a
and ans[len(snake_case_ ) - 1] != p_number_a
), "'ans' must been a list without the arguments"
# 'ans' contains not 'pNumber1' and 'pNumber2' !
return ans
def lowerCAmelCase_ ( snake_case_ ):
assert isinstance(snake_case_,snake_case_ ) and (n >= 1), "'n' must been int and >= 1"
_A : Optional[int] = [] # will be returned.
for divisor in range(1,n + 1 ):
if n % divisor == 0:
ans.append(snake_case_ )
# precondition
assert ans[0] == 1 and ans[len(snake_case_ ) - 1] == n, "Error in function getDivisiors(...)"
return ans
def lowerCAmelCase_ ( snake_case_ ):
assert isinstance(snake_case_,snake_case_ ) and (
number > 1
), "'number' must been an int and >= 1"
_A : Optional[Any] = get_divisors(snake_case_ )
# precondition
assert (
isinstance(snake_case_,snake_case_ )
and (divisors[0] == 1)
and (divisors[len(snake_case_ ) - 1] == number)
), "Error in help-function getDivisiors(...)"
# summed all divisors up to 'number' (exclusive), hence [:-1]
return sum(divisors[:-1] ) == number
def lowerCAmelCase_ ( snake_case_,snake_case_ ):
assert (
isinstance(snake_case_,snake_case_ )
and isinstance(snake_case_,snake_case_ )
and (denominator != 0)
), "The arguments must been from type int and 'denominator' != 0"
# build the greatest common divisor of numerator and denominator.
_A : Union[str, Any] = gcd(abs(snake_case_ ),abs(snake_case_ ) )
# precondition
assert (
isinstance(snake_case_,snake_case_ )
and (numerator % gcd_of_fraction == 0)
and (denominator % gcd_of_fraction == 0)
), "Error in function gcd(...,...)"
return (numerator // gcd_of_fraction, denominator // gcd_of_fraction)
def lowerCAmelCase_ ( snake_case_ ):
assert isinstance(snake_case_,snake_case_ ) and (n >= 0), "'n' must been a int and >= 0"
_A : List[Any] = 1 # this will be return.
for factor in range(1,n + 1 ):
ans *= factor
return ans
def lowerCAmelCase_ ( snake_case_ ):
assert isinstance(snake_case_,snake_case_ ) and (n >= 0), "'n' must been an int and >= 0"
_A : Any = 0
_A : Tuple = 1
_A : Dict = 1 # this will be return
for _ in range(n - 1 ):
_A : Optional[int] = ans
ans += fiba
_A : Optional[int] = tmp
return ans
| 26 |
import itertools
import random
import unittest
import numpy as np
from transformers import BatchFeature, SpeechTaFeatureExtractor
from transformers.testing_utils import require_torch
from transformers.utils.import_utils import is_torch_available
from ...test_sequence_feature_extraction_common import SequenceFeatureExtractionTestMixin
if is_torch_available():
import torch
a_ :Any = random.Random()
def lowercase_ (A : int , A : Union[str, Any]=1.0 , A : List[str]=None , A : Any=None ):
if rng is None:
snake_case__ : List[str] = global_rng
snake_case__ : int = []
for batch_idx in range(shape[0] ):
values.append([] )
for _ in range(shape[1] ):
values[-1].append(rng.random() * scale )
return values
@require_torch
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
def __init__( self : Optional[Any], _snake_case : List[str], _snake_case : Tuple=7, _snake_case : Union[str, Any]=4_0_0, _snake_case : Any=2_0_0_0, _snake_case : Dict=1, _snake_case : Optional[Any]=0.0, _snake_case : List[Any]=1_6_0_0_0, _snake_case : List[Any]=True, _snake_case : List[Any]=8_0, _snake_case : Dict=1_6, _snake_case : str=6_4, _snake_case : Tuple="hann_window", _snake_case : Union[str, Any]=8_0, _snake_case : Optional[Any]=7_6_0_0, _snake_case : str=1e-10, _snake_case : Any=True, ) ->Union[str, Any]:
snake_case__ : Optional[int] = parent
snake_case__ : Optional[Any] = batch_size
snake_case__ : List[Any] = min_seq_length
snake_case__ : List[Any] = max_seq_length
snake_case__ : Any = (self.max_seq_length - self.min_seq_length) // (self.batch_size - 1)
snake_case__ : Tuple = feature_size
snake_case__ : List[Any] = padding_value
snake_case__ : Any = sampling_rate
snake_case__ : Dict = do_normalize
snake_case__ : Union[str, Any] = num_mel_bins
snake_case__ : Any = hop_length
snake_case__ : Any = win_length
snake_case__ : Any = win_function
snake_case__ : Optional[int] = fmin
snake_case__ : int = fmax
snake_case__ : Union[str, Any] = mel_floor
snake_case__ : Union[str, Any] = return_attention_mask
def lowercase_ ( self : Optional[int] ) ->List[str]:
return {
"feature_size": self.feature_size,
"padding_value": self.padding_value,
"sampling_rate": self.sampling_rate,
"do_normalize": self.do_normalize,
"num_mel_bins": self.num_mel_bins,
"hop_length": self.hop_length,
"win_length": self.win_length,
"win_function": self.win_function,
"fmin": self.fmin,
"fmax": self.fmax,
"mel_floor": self.mel_floor,
"return_attention_mask": self.return_attention_mask,
}
def lowercase_ ( self : Any, _snake_case : Optional[Any]=False, _snake_case : List[str]=False ) ->Union[str, Any]:
def _flatten(_snake_case : List[str] ):
return list(itertools.chain(*_snake_case ) )
if equal_length:
snake_case__ : Any = floats_list((self.batch_size, self.max_seq_length) )
else:
# make sure that inputs increase in size
snake_case__ : int = [
_flatten(floats_list((x, self.feature_size) ) )
for x in range(self.min_seq_length, self.max_seq_length, self.seq_length_diff )
]
if numpify:
snake_case__ : Any = [np.asarray(_snake_case ) for x in speech_inputs]
return speech_inputs
def lowercase_ ( self : Union[str, Any], _snake_case : str=False, _snake_case : Dict=False ) ->List[str]:
if equal_length:
snake_case__ : Optional[Any] = [floats_list((self.max_seq_length, self.num_mel_bins) ) for _ in range(self.batch_size )]
else:
# make sure that inputs increase in size
snake_case__ : List[str] = [
floats_list((x, self.num_mel_bins) )
for x in range(self.min_seq_length, self.max_seq_length, self.seq_length_diff )
]
if numpify:
snake_case__ : int = [np.asarray(_snake_case ) for x in speech_inputs]
return speech_inputs
@require_torch
class snake_case__ ( lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = SpeechTaFeatureExtractor
def lowercase_ ( self : int ) ->Union[str, Any]:
snake_case__ : List[str] = SpeechTaFeatureExtractionTester(self )
def lowercase_ ( self : Any, _snake_case : Dict ) ->Any:
self.assertTrue(np.all(np.mean(_snake_case, axis=0 ) < 1e-3 ) )
self.assertTrue(np.all(np.abs(np.var(_snake_case, axis=0 ) - 1 ) < 1e-3 ) )
def lowercase_ ( self : List[Any] ) ->Union[str, Any]:
# Tests that all call wrap to encode_plus and batch_encode_plus
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
# create three inputs of length 800, 1000, and 1200
snake_case__ : int = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Tuple = [np.asarray(_snake_case ) for speech_input in speech_inputs]
# Test not batched input
snake_case__ : str = feat_extract(speech_inputs[0], return_tensors='np' ).input_values
snake_case__ : List[str] = feat_extract(np_speech_inputs[0], return_tensors='np' ).input_values
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test batched
snake_case__ : Any = feat_extract(_snake_case, return_tensors='np' ).input_values
snake_case__ : Union[str, Any] = feat_extract(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
def lowercase_ ( self : int ) ->Optional[int]:
snake_case__ : List[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Tuple = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : int = ['longest', 'max_length', 'do_not_pad']
snake_case__ : List[str] = [None, 1_6_0_0, None]
for max_length, padding in zip(_snake_case, _snake_case ):
snake_case__ : Optional[int] = feat_extract(_snake_case, padding=_snake_case, max_length=_snake_case, return_tensors='np' )
snake_case__ : Optional[int] = processed.input_values
self._check_zero_mean_unit_variance(input_values[0][:8_0_0] )
self.assertTrue(input_values[0][8_0_0:].sum() < 1e-6 )
self._check_zero_mean_unit_variance(input_values[1][:1_0_0_0] )
self.assertTrue(input_values[0][1_0_0_0:].sum() < 1e-6 )
self._check_zero_mean_unit_variance(input_values[2][:1_2_0_0] )
def lowercase_ ( self : Union[str, Any] ) ->Optional[Any]:
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Tuple = range(8_0_0, 1_4_0_0, 2_0_0 )
snake_case__ : Optional[Any] = [floats_list((1, x) )[0] for x in lengths]
snake_case__ : Union[str, Any] = ['longest', 'max_length', 'do_not_pad']
snake_case__ : str = [None, 1_6_0_0, None]
for max_length, padding in zip(_snake_case, _snake_case ):
snake_case__ : List[str] = feat_extract(_snake_case, max_length=_snake_case, padding=_snake_case )
snake_case__ : Tuple = processed.input_values
self._check_zero_mean_unit_variance(input_values[0][:8_0_0] )
self._check_zero_mean_unit_variance(input_values[1][:1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2][:1_2_0_0] )
def lowercase_ ( self : List[Any] ) ->Optional[Any]:
snake_case__ : Any = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : str = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Optional[Any] = feat_extract(
_snake_case, truncation=_snake_case, max_length=1_0_0_0, padding='max_length', return_tensors='np' )
snake_case__ : int = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1] )
self._check_zero_mean_unit_variance(input_values[2] )
def lowercase_ ( self : int ) ->Union[str, Any]:
snake_case__ : str = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Dict = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : str = feat_extract(
_snake_case, truncation=_snake_case, max_length=1_0_0_0, padding='longest', return_tensors='np' )
snake_case__ : Dict = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1, :1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2] )
# make sure that if max_length < longest -> then pad to max_length
self.assertTrue(input_values.shape == (3, 1_0_0_0) )
snake_case__ : Tuple = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : List[str] = feat_extract(
_snake_case, truncation=_snake_case, max_length=2_0_0_0, padding='longest', return_tensors='np' )
snake_case__ : Optional[Any] = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1, :1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2] )
# make sure that if max_length > longest -> then pad to longest
self.assertTrue(input_values.shape == (3, 1_2_0_0) )
def lowercase_ ( self : List[str] ) ->Dict:
snake_case__ : Dict = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : List[Any] = np.random.rand(1_0_0 ).astype(np.floataa )
snake_case__ : int = np_speech_inputs.tolist()
for inputs in [py_speech_inputs, np_speech_inputs]:
snake_case__ : int = feature_extractor.pad([{'input_values': inputs}], return_tensors='np' )
self.assertTrue(np_processed.input_values.dtype == np.floataa )
snake_case__ : Optional[int] = feature_extractor.pad([{'input_values': inputs}], return_tensors='pt' )
self.assertTrue(pt_processed.input_values.dtype == torch.floataa )
def lowercase_ ( self : Optional[int] ) ->Optional[Any]:
# Tests that all call wrap to encode_plus and batch_encode_plus
snake_case__ : str = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
# create three inputs of length 800, 1000, and 1200
snake_case__ : List[Any] = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Dict = [np.asarray(_snake_case ) for speech_input in speech_inputs]
# Test feature size
snake_case__ : Optional[int] = feature_extractor(audio_target=_snake_case, padding=_snake_case, return_tensors='np' ).input_values
self.assertTrue(input_values.ndim == 3 )
self.assertTrue(input_values.shape[-1] == feature_extractor.num_mel_bins )
# Test not batched input
snake_case__ : Dict = feature_extractor(speech_inputs[0], return_tensors='np' ).input_values
snake_case__ : Any = feature_extractor(np_speech_inputs[0], return_tensors='np' ).input_values
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test batched
snake_case__ : Dict = feature_extractor(_snake_case, return_tensors='np' ).input_values
snake_case__ : Dict = feature_extractor(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test 2-D numpy arrays are batched.
snake_case__ : Optional[Any] = [floats_list((1, x) )[0] for x in (8_0_0, 8_0_0, 8_0_0)]
snake_case__ : int = np.asarray(_snake_case )
snake_case__ : Union[str, Any] = feature_extractor(_snake_case, return_tensors='np' ).input_values
snake_case__ : Union[str, Any] = feature_extractor(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
def lowercase_ ( self : Union[str, Any] ) ->str:
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : List[str] = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : Tuple = BatchFeature({input_name: speech_inputs} )
self.assertTrue(all(len(_snake_case ) == len(_snake_case ) for x, y in zip(_snake_case, processed_features[input_name] ) ) )
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target(equal_length=_snake_case )
snake_case__ : Union[str, Any] = BatchFeature({input_name: speech_inputs}, tensor_type='np' )
snake_case__ : Dict = processed_features[input_name]
if len(batch_features_input.shape ) < 3:
snake_case__ : List[str] = batch_features_input[:, :, None]
self.assertTrue(
batch_features_input.shape
== (self.feat_extract_tester.batch_size, len(speech_inputs[0] ), self.feat_extract_tester.num_mel_bins) )
@require_torch
def lowercase_ ( self : List[str] ) ->Any:
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target(equal_length=_snake_case )
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Tuple = feat_extract.model_input_names[0]
snake_case__ : List[Any] = BatchFeature({input_name: speech_inputs}, tensor_type='pt' )
snake_case__ : Tuple = processed_features[input_name]
if len(batch_features_input.shape ) < 3:
snake_case__ : Any = batch_features_input[:, :, None]
self.assertTrue(
batch_features_input.shape
== (self.feat_extract_tester.batch_size, len(speech_inputs[0] ), self.feat_extract_tester.num_mel_bins) )
@require_torch
def lowercase_ ( self : Optional[int] ) ->Tuple:
snake_case__ : Dict = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Union[str, Any] = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : List[str] = BatchFeature({input_name: speech_inputs} )
snake_case__ : int = feat_extract.num_mel_bins # hack!
snake_case__ : Tuple = feat_extract.pad(_snake_case, padding='longest', return_tensors='np' )[input_name]
snake_case__ : Union[str, Any] = feat_extract.pad(_snake_case, padding='longest', return_tensors='pt' )[input_name]
self.assertTrue(abs(input_np.astype(np.floataa ).sum() - input_pt.numpy().astype(np.floataa ).sum() ) < 1e-2 )
def lowercase_ ( self : int ) ->Any:
snake_case__ : Any = self.feat_extract_dict
snake_case__ : List[Any] = True
snake_case__ : Union[str, Any] = self.feature_extraction_class(**_snake_case )
snake_case__ : Any = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : List[Any] = [len(_snake_case ) for x in speech_inputs]
snake_case__ : Union[str, Any] = feat_extract.model_input_names[0]
snake_case__ : Optional[int] = BatchFeature({input_name: speech_inputs} )
snake_case__ : List[str] = feat_extract.num_mel_bins # hack!
snake_case__ : str = feat_extract.pad(_snake_case, padding='longest', return_tensors='np' )
self.assertIn('attention_mask', _snake_case )
self.assertListEqual(list(processed.attention_mask.shape ), list(processed[input_name].shape[:2] ) )
self.assertListEqual(processed.attention_mask.sum(-1 ).tolist(), _snake_case )
def lowercase_ ( self : Optional[int] ) ->str:
snake_case__ : int = self.feat_extract_dict
snake_case__ : List[str] = True
snake_case__ : Tuple = self.feature_extraction_class(**_snake_case )
snake_case__ : List[str] = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : str = [len(_snake_case ) for x in speech_inputs]
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : Optional[int] = BatchFeature({input_name: speech_inputs} )
snake_case__ : Optional[Any] = min(_snake_case )
snake_case__ : Union[str, Any] = feat_extract.num_mel_bins # hack!
snake_case__ : Tuple = feat_extract.pad(
_snake_case, padding='max_length', max_length=_snake_case, truncation=_snake_case, return_tensors='np' )
self.assertIn('attention_mask', _snake_case )
self.assertListEqual(
list(processed_pad.attention_mask.shape ), [processed_pad[input_name].shape[0], max_length] )
self.assertListEqual(
processed_pad.attention_mask[:, :max_length].sum(-1 ).tolist(), [max_length for x in speech_inputs] )
def lowercase_ ( self : List[Any], _snake_case : Optional[int] ) ->Optional[Any]:
from datasets import load_dataset
snake_case__ : str = load_dataset('hf-internal-testing/librispeech_asr_dummy', 'clean', split='validation' )
# automatic decoding with librispeech
snake_case__ : Dict = ds.sort('id' ).select(range(_snake_case ) )[:num_samples]['audio']
return [x["array"] for x in speech_samples]
def lowercase_ ( self : str ) ->str:
# fmt: off
snake_case__ : List[Any] = torch.tensor(
[2.3804e-03, 2.0752e-03, 1.9836e-03, 2.1057e-03, 1.6174e-03,
3.0518e-04, 9.1553e-05, 3.3569e-04, 9.7656e-04, 1.8311e-03,
2.0142e-03, 2.1057e-03, 1.7395e-03, 4.5776e-04, -3.9673e-04,
4.5776e-04, 1.0071e-03, 9.1553e-05, 4.8828e-04, 1.1597e-03,
7.3242e-04, 9.4604e-04, 1.8005e-03, 1.8311e-03, 8.8501e-04,
4.2725e-04, 4.8828e-04, 7.3242e-04, 1.0986e-03, 2.1057e-03] )
# fmt: on
snake_case__ : Union[str, Any] = self._load_datasamples(1 )
snake_case__ : Optional[int] = SpeechTaFeatureExtractor()
snake_case__ : List[Any] = feature_extractor(_snake_case, return_tensors='pt' ).input_values
self.assertEquals(input_values.shape, (1, 9_3_6_8_0) )
self.assertTrue(torch.allclose(input_values[0, :3_0], _snake_case, atol=1e-6 ) )
def lowercase_ ( self : Any ) ->str:
# fmt: off
snake_case__ : Optional[Any] = torch.tensor(
[-2.6_8_7_0, -3.0_1_0_4, -3.1_3_5_6, -3.5_3_5_2, -3.0_0_4_4, -3.0_3_5_3, -3.4_7_1_9, -3.6_7_7_7,
-3.1_5_2_0, -2.9_4_3_5, -2.6_5_5_3, -2.8_7_9_5, -2.9_9_4_4, -2.5_9_2_1, -3.0_2_7_9, -3.0_3_8_6,
-3.0_8_6_4, -3.1_2_9_1, -3.2_3_5_3, -2.7_4_4_4, -2.6_8_3_1, -2.7_2_8_7, -3.1_7_6_1, -3.1_5_7_1,
-3.2_7_2_6, -3.0_5_8_2, -3.1_0_0_7, -3.4_5_3_3, -3.4_6_9_5, -3.0_9_9_8] )
# fmt: on
snake_case__ : List[str] = self._load_datasamples(1 )
snake_case__ : str = SpeechTaFeatureExtractor()
snake_case__ : Optional[Any] = feature_extractor(audio_target=_snake_case, return_tensors='pt' ).input_values
self.assertEquals(input_values.shape, (1, 3_6_6, 8_0) )
self.assertTrue(torch.allclose(input_values[0, 0, :3_0], _snake_case, atol=1e-4 ) )
| 277 | 0 |
import copy
import os
import tempfile
from unittest import TestCase
from unittest.mock import patch
import numpy as np
import pyarrow as pa
import pyarrow.parquet as pq
import pytest
from datasets.arrow_writer import ArrowWriter, OptimizedTypedSequence, ParquetWriter, TypedSequence
from datasets.features import ArrayaD, ClassLabel, Features, Image, Value
from datasets.features.features import ArrayaDExtensionType, cast_to_python_objects
from datasets.keyhash import DuplicatedKeysError, InvalidKeyError
from .utils import require_pil
class __lowerCAmelCase ( lowerCAmelCase_ ):
def _lowerCamelCase ( self : List[Any]) -> str:
"""simple docstring"""
_UpperCAmelCase = pa.array(TypedSequence([1, 2, 3]))
self.assertEqual(arr.type , pa.intaa())
def _lowerCamelCase ( self : Dict) -> str:
"""simple docstring"""
with self.assertRaises(_snake_case):
_UpperCAmelCase = pa.array(TypedSequence([1, 2, 3]) , type=pa.intaa())
def _lowerCamelCase ( self : List[Any]) -> Dict:
"""simple docstring"""
with self.assertRaises(_snake_case):
_UpperCAmelCase = pa.array(TypedSequence([1, 2, 3] , try_type=Value('bool') , type=Value('int64')))
def _lowerCamelCase ( self : int) -> Optional[int]:
"""simple docstring"""
_UpperCAmelCase = pa.array(TypedSequence([1, 2, 3] , type=Value('int32')))
self.assertEqual(arr.type , pa.intaa())
def _lowerCamelCase ( self : List[str]) -> Optional[Any]:
"""simple docstring"""
with self.assertRaises((TypeError, pa.lib.ArrowInvalid)):
_UpperCAmelCase = pa.array(TypedSequence(['foo', 'bar'] , type=Value('int64')))
def _lowerCamelCase ( self : Tuple) -> Any:
"""simple docstring"""
_UpperCAmelCase = pa.array(TypedSequence([1, 2, 3] , try_type=Value('int32')))
self.assertEqual(arr.type , pa.intaa())
def _lowerCamelCase ( self : Optional[int]) -> List[str]:
"""simple docstring"""
_UpperCAmelCase = pa.array(TypedSequence(['foo', 'bar'] , try_type=Value('int64')))
self.assertEqual(arr.type , pa.string())
def _lowerCamelCase ( self : str) -> List[Any]:
"""simple docstring"""
_UpperCAmelCase = pa.array(TypedSequence([[[1, 2, 3]]] , type=ArrayaD((1, 3) , 'int64')))
self.assertEqual(arr.type , ArrayaDExtensionType((1, 3) , 'int64'))
def _lowerCamelCase ( self : Any) -> List[Any]:
"""simple docstring"""
with self.assertRaises((TypeError, pa.lib.ArrowInvalid)):
_UpperCAmelCase = pa.array(TypedSequence(['foo', 'bar'] , type=ArrayaD((1, 3) , 'int64')))
def _lowerCamelCase ( self : Union[str, Any]) -> Tuple:
"""simple docstring"""
_UpperCAmelCase = pa.array(TypedSequence([[[1, 2, 3]]] , try_type=ArrayaD((1, 3) , 'int64')))
self.assertEqual(arr.type , ArrayaDExtensionType((1, 3) , 'int64'))
def _lowerCamelCase ( self : int) -> Union[str, Any]:
"""simple docstring"""
_UpperCAmelCase = pa.array(TypedSequence(['foo', 'bar'] , try_type=ArrayaD((1, 3) , 'int64')))
self.assertEqual(arr.type , pa.string())
@require_pil
def _lowerCamelCase ( self : Any) -> Optional[int]:
"""simple docstring"""
import PIL.Image
_UpperCAmelCase = PIL.Image.fromarray(np.arange(10 , dtype=np.uinta).reshape(2 , 5))
with patch(
'datasets.arrow_writer.cast_to_python_objects' , side_effect=_snake_case) as mock_cast_to_python_objects:
_UpperCAmelCase = pa.array(TypedSequence([{'path': None, 'bytes': B'image_bytes'}, pil_image] , type=Image()))
_UpperCAmelCase = mock_cast_to_python_objects.call_args_list[-1]
self.assertIn('optimize_list_casting' , _snake_case)
self.assertFalse(kwargs['optimize_list_casting'])
def A ( _UpperCAmelCase : Dict , _UpperCAmelCase : int ) -> Dict:
'''simple docstring'''
_UpperCAmelCase = pa.BufferReader(_UpperCAmelCase ) if isinstance(_UpperCAmelCase , pa.Buffer ) else pa.memory_map(_UpperCAmelCase )
_UpperCAmelCase = pa.ipc.open_stream(_UpperCAmelCase )
_UpperCAmelCase = f.read_all()
assert len(pa_table.to_batches() ) == expected_num_chunks
assert pa_table.to_pydict() == {"col_1": ["foo", "bar"], "col_2": [1, 2]}
del pa_table
@pytest.mark.parametrize('writer_batch_size' , [None, 1, 10] )
@pytest.mark.parametrize(
'fields' , [None, {'col_1': pa.string(), 'col_2': pa.intaa()}, {'col_1': pa.string(), 'col_2': pa.intaa()}] )
def A ( _UpperCAmelCase : List[str] , _UpperCAmelCase : List[str] ) -> List[Any]:
'''simple docstring'''
_UpperCAmelCase = pa.BufferOutputStream()
_UpperCAmelCase = pa.schema(_UpperCAmelCase ) if fields else None
with ArrowWriter(stream=_UpperCAmelCase , schema=_UpperCAmelCase , writer_batch_size=_UpperCAmelCase ) as writer:
writer.write({'col_1': 'foo', 'col_2': 1} )
writer.write({'col_1': 'bar', 'col_2': 2} )
_UpperCAmelCase = writer.finalize()
assert num_examples == 2
assert num_bytes > 0
if not fields:
_UpperCAmelCase = {'col_1': pa.string(), 'col_2': pa.intaa()}
assert writer._schema == pa.schema(_UpperCAmelCase , metadata=writer._schema.metadata )
_check_output(output.getvalue() , expected_num_chunks=num_examples if writer_batch_size == 1 else 1 )
def A ( ) -> Optional[int]:
'''simple docstring'''
_UpperCAmelCase = pa.BufferOutputStream()
_UpperCAmelCase = Features({'labels': ClassLabel(names=['neg', 'pos'] )} )
with ArrowWriter(stream=_UpperCAmelCase , features=_UpperCAmelCase ) as writer:
writer.write({'labels': 0} )
writer.write({'labels': 1} )
_UpperCAmelCase = writer.finalize()
assert num_examples == 2
assert num_bytes > 0
assert writer._schema == features.arrow_schema
assert writer._schema.metadata == features.arrow_schema.metadata
_UpperCAmelCase = pa.BufferReader(output.getvalue() )
_UpperCAmelCase = pa.ipc.open_stream(_UpperCAmelCase )
_UpperCAmelCase = f.read_all()
_UpperCAmelCase = pa_table.schema
assert pa_table.num_rows == 2
assert schema == features.arrow_schema
assert schema.metadata == features.arrow_schema.metadata
assert features == Features.from_arrow_schema(_UpperCAmelCase )
@pytest.mark.parametrize('writer_batch_size' , [None, 1, 10] )
def A ( _UpperCAmelCase : int ) -> Dict:
'''simple docstring'''
_UpperCAmelCase = pa.BufferOutputStream()
with ArrowWriter(
stream=_UpperCAmelCase , writer_batch_size=_UpperCAmelCase , hash_salt='split_name' , check_duplicates=_UpperCAmelCase , ) as writer:
with pytest.raises(_UpperCAmelCase ):
writer.write({'col_1': 'foo', 'col_2': 1} , key=[1, 2] )
_UpperCAmelCase = writer.finalize()
@pytest.mark.parametrize('writer_batch_size' , [None, 2, 10] )
def A ( _UpperCAmelCase : Optional[int] ) -> List[Any]:
'''simple docstring'''
_UpperCAmelCase = pa.BufferOutputStream()
with ArrowWriter(
stream=_UpperCAmelCase , writer_batch_size=_UpperCAmelCase , hash_salt='split_name' , check_duplicates=_UpperCAmelCase , ) as writer:
with pytest.raises(_UpperCAmelCase ):
writer.write({'col_1': 'foo', 'col_2': 1} , key=10 )
writer.write({'col_1': 'bar', 'col_2': 2} , key=10 )
_UpperCAmelCase = writer.finalize()
@pytest.mark.parametrize('writer_batch_size' , [None, 2, 10] )
def A ( _UpperCAmelCase : List[Any] ) -> Optional[int]:
'''simple docstring'''
_UpperCAmelCase = pa.BufferOutputStream()
with ArrowWriter(
stream=_UpperCAmelCase , writer_batch_size=_UpperCAmelCase , hash_salt='split_name' , check_duplicates=_UpperCAmelCase , ) as writer:
writer.write({'col_1': 'foo', 'col_2': 1} , key=1 )
writer.write({'col_1': 'bar', 'col_2': 2} , key=2 )
_UpperCAmelCase = writer.finalize()
assert num_examples == 2
assert num_bytes > 0
_check_output(output.getvalue() , expected_num_chunks=num_examples if writer_batch_size == 1 else 1 )
@pytest.mark.parametrize('writer_batch_size' , [None, 1, 10] )
@pytest.mark.parametrize(
'fields' , [None, {'col_1': pa.string(), 'col_2': pa.intaa()}, {'col_1': pa.string(), 'col_2': pa.intaa()}] )
def A ( _UpperCAmelCase : Dict , _UpperCAmelCase : Any ) -> List[Any]:
'''simple docstring'''
_UpperCAmelCase = pa.BufferOutputStream()
_UpperCAmelCase = pa.schema(_UpperCAmelCase ) if fields else None
with ArrowWriter(stream=_UpperCAmelCase , schema=_UpperCAmelCase , writer_batch_size=_UpperCAmelCase ) as writer:
writer.write_batch({'col_1': ['foo', 'bar'], 'col_2': [1, 2]} )
writer.write_batch({'col_1': [], 'col_2': []} )
_UpperCAmelCase = writer.finalize()
assert num_examples == 2
assert num_bytes > 0
if not fields:
_UpperCAmelCase = {'col_1': pa.string(), 'col_2': pa.intaa()}
assert writer._schema == pa.schema(_UpperCAmelCase , metadata=writer._schema.metadata )
_check_output(output.getvalue() , expected_num_chunks=num_examples if writer_batch_size == 1 else 1 )
@pytest.mark.parametrize('writer_batch_size' , [None, 1, 10] )
@pytest.mark.parametrize(
'fields' , [None, {'col_1': pa.string(), 'col_2': pa.intaa()}, {'col_1': pa.string(), 'col_2': pa.intaa()}] )
def A ( _UpperCAmelCase : Tuple , _UpperCAmelCase : str ) -> Any:
'''simple docstring'''
_UpperCAmelCase = pa.BufferOutputStream()
_UpperCAmelCase = pa.schema(_UpperCAmelCase ) if fields else None
with ArrowWriter(stream=_UpperCAmelCase , schema=_UpperCAmelCase , writer_batch_size=_UpperCAmelCase ) as writer:
writer.write_table(pa.Table.from_pydict({'col_1': ['foo', 'bar'], 'col_2': [1, 2]} ) )
_UpperCAmelCase = writer.finalize()
assert num_examples == 2
assert num_bytes > 0
if not fields:
_UpperCAmelCase = {'col_1': pa.string(), 'col_2': pa.intaa()}
assert writer._schema == pa.schema(_UpperCAmelCase , metadata=writer._schema.metadata )
_check_output(output.getvalue() , expected_num_chunks=num_examples if writer_batch_size == 1 else 1 )
@pytest.mark.parametrize('writer_batch_size' , [None, 1, 10] )
@pytest.mark.parametrize(
'fields' , [None, {'col_1': pa.string(), 'col_2': pa.intaa()}, {'col_1': pa.string(), 'col_2': pa.intaa()}] )
def A ( _UpperCAmelCase : List[str] , _UpperCAmelCase : Union[str, Any] ) -> Tuple:
'''simple docstring'''
_UpperCAmelCase = pa.BufferOutputStream()
_UpperCAmelCase = pa.schema(_UpperCAmelCase ) if fields else None
with ArrowWriter(stream=_UpperCAmelCase , schema=_UpperCAmelCase , writer_batch_size=_UpperCAmelCase ) as writer:
writer.write_row(pa.Table.from_pydict({'col_1': ['foo'], 'col_2': [1]} ) )
writer.write_row(pa.Table.from_pydict({'col_1': ['bar'], 'col_2': [2]} ) )
_UpperCAmelCase = writer.finalize()
assert num_examples == 2
assert num_bytes > 0
if not fields:
_UpperCAmelCase = {'col_1': pa.string(), 'col_2': pa.intaa()}
assert writer._schema == pa.schema(_UpperCAmelCase , metadata=writer._schema.metadata )
_check_output(output.getvalue() , expected_num_chunks=num_examples if writer_batch_size == 1 else 1 )
def A ( ) -> int:
'''simple docstring'''
with tempfile.TemporaryDirectory() as tmp_dir:
_UpperCAmelCase = {'col_1': pa.string(), 'col_2': pa.intaa()}
_UpperCAmelCase = os.path.join(_UpperCAmelCase , 'test.arrow' )
with ArrowWriter(path=_UpperCAmelCase , schema=pa.schema(_UpperCAmelCase ) ) as writer:
writer.write_batch({'col_1': ['foo', 'bar'], 'col_2': [1, 2]} )
_UpperCAmelCase = writer.finalize()
assert num_examples == 2
assert num_bytes > 0
assert writer._schema == pa.schema(_UpperCAmelCase , metadata=writer._schema.metadata )
_check_output(_UpperCAmelCase , 1 )
def A ( _UpperCAmelCase : Union[str, Any] ) -> Dict:
'''simple docstring'''
if pa.types.is_list(_UpperCAmelCase ):
return get_base_dtype(arr_type.value_type )
else:
return arr_type
def A ( _UpperCAmelCase : List[Any] , _UpperCAmelCase : Union[str, Any] ) -> Union[str, Any]:
'''simple docstring'''
if isinstance(lst[0] , _UpperCAmelCase ):
change_first_primitive_element_in_list(lst[0] , _UpperCAmelCase )
else:
_UpperCAmelCase = value
@pytest.mark.parametrize('optimized_int_type, expected_dtype' , [(None, pa.intaa()), (Value('int32' ), pa.intaa())] )
@pytest.mark.parametrize('sequence' , [[1, 2, 3], [[1, 2, 3]], [[[1, 2, 3]]]] )
def A ( _UpperCAmelCase : Tuple , _UpperCAmelCase : List[str] , _UpperCAmelCase : List[Any] ) -> Union[str, Any]:
'''simple docstring'''
_UpperCAmelCase = pa.array(TypedSequence(_UpperCAmelCase , optimized_int_type=_UpperCAmelCase ) )
assert get_base_dtype(arr.type ) == expected_dtype
@pytest.mark.parametrize(
'col, expected_dtype' , [
('attention_mask', pa.inta()),
('special_tokens_mask', pa.inta()),
('token_type_ids', pa.inta()),
('input_ids', pa.intaa()),
('other', pa.intaa()),
] , )
@pytest.mark.parametrize('sequence' , [[1, 2, 3], [[1, 2, 3]], [[[1, 2, 3]]]] )
def A ( _UpperCAmelCase : List[Any] , _UpperCAmelCase : Union[str, Any] , _UpperCAmelCase : str ) -> Optional[Any]:
'''simple docstring'''
# in range
_UpperCAmelCase = pa.array(OptimizedTypedSequence(_UpperCAmelCase , col=_UpperCAmelCase ) )
assert get_base_dtype(arr.type ) == expected_dtype
# not in range
if col != "other":
# avoids errors due to in-place modifications
_UpperCAmelCase = copy.deepcopy(_UpperCAmelCase )
_UpperCAmelCase = np.iinfo(expected_dtype.to_pandas_dtype() ).max + 1
change_first_primitive_element_in_list(_UpperCAmelCase , _UpperCAmelCase )
_UpperCAmelCase = pa.array(OptimizedTypedSequence(_UpperCAmelCase , col=_UpperCAmelCase ) )
assert get_base_dtype(arr.type ) == pa.intaa()
@pytest.mark.parametrize('raise_exception' , [False, True] )
def A ( _UpperCAmelCase : Union[str, Any] , _UpperCAmelCase : List[str] ) -> Optional[Any]:
'''simple docstring'''
_UpperCAmelCase = str(tmp_path / 'dataset-train.arrow' )
try:
with ArrowWriter(path=_UpperCAmelCase ) as writer:
if raise_exception:
raise pa.lib.ArrowInvalid()
else:
writer.stream.close()
except pa.lib.ArrowInvalid:
pass
finally:
assert writer.stream.closed
def A ( _UpperCAmelCase : int ) -> Tuple:
'''simple docstring'''
_UpperCAmelCase = 'mock://dataset-train.arrow'
with ArrowWriter(path=_UpperCAmelCase , storage_options=mockfs.storage_options ) as writer:
assert isinstance(writer._fs , type(_UpperCAmelCase ) )
assert writer._fs.storage_options == mockfs.storage_options
writer.write({'col_1': 'foo', 'col_2': 1} )
writer.write({'col_1': 'bar', 'col_2': 2} )
_UpperCAmelCase = writer.finalize()
assert num_examples == 2
assert num_bytes > 0
assert mockfs.exists(_UpperCAmelCase )
def A ( ) -> Any:
'''simple docstring'''
_UpperCAmelCase = pa.BufferOutputStream()
with ParquetWriter(stream=_UpperCAmelCase ) as writer:
writer.write({'col_1': 'foo', 'col_2': 1} )
writer.write({'col_1': 'bar', 'col_2': 2} )
_UpperCAmelCase = writer.finalize()
assert num_examples == 2
assert num_bytes > 0
_UpperCAmelCase = pa.BufferReader(output.getvalue() )
_UpperCAmelCase = pq.read_table(_UpperCAmelCase )
assert pa_table.to_pydict() == {"col_1": ["foo", "bar"], "col_2": [1, 2]}
@require_pil
@pytest.mark.parametrize('embed_local_files' , [False, True] )
def A ( _UpperCAmelCase : List[Any] , _UpperCAmelCase : Optional[Any] ) -> List[Any]:
'''simple docstring'''
import PIL.Image
_UpperCAmelCase = str(tmp_path / 'test_image_rgb.jpg' )
PIL.Image.fromarray(np.zeros((5, 5) , dtype=np.uinta ) ).save(_UpperCAmelCase , format='png' )
_UpperCAmelCase = pa.BufferOutputStream()
with ParquetWriter(
stream=_UpperCAmelCase , features=Features({'image': Image()} ) , embed_local_files=_UpperCAmelCase ) as writer:
writer.write({'image': image_path} )
writer.finalize()
_UpperCAmelCase = pa.BufferReader(output.getvalue() )
_UpperCAmelCase = pq.read_table(_UpperCAmelCase )
_UpperCAmelCase = pa_table.to_pydict()
if embed_local_files:
assert isinstance(out['image'][0]['path'] , _UpperCAmelCase )
with open(_UpperCAmelCase , 'rb' ) as f:
assert out["image"][0]["bytes"] == f.read()
else:
assert out["image"][0]["path"] == image_path
assert out["image"][0]["bytes"] is None
def A ( ) -> Dict:
'''simple docstring'''
_UpperCAmelCase = pa.schema([pa.field('col_1' , pa.string() , nullable=_UpperCAmelCase )] )
_UpperCAmelCase = pa.BufferOutputStream()
with ArrowWriter(stream=_UpperCAmelCase ) as writer:
writer._build_writer(inferred_schema=_UpperCAmelCase )
assert writer._schema == pa.schema([pa.field('col_1' , pa.string() )] )
| 339 |
# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from ..models.auto import AutoModelForSeqaSeqLM, AutoTokenizer
from .base import PipelineTool
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """philschmid/bart-large-cnn-samsum"""
_SCREAMING_SNAKE_CASE = (
"""This is a tool that summarizes an English text. It takes an input `text` containing the text to summarize, """
"""and returns a summary of the text."""
)
_SCREAMING_SNAKE_CASE = """summarizer"""
_SCREAMING_SNAKE_CASE = AutoTokenizer
_SCREAMING_SNAKE_CASE = AutoModelForSeqaSeqLM
_SCREAMING_SNAKE_CASE = ["""text"""]
_SCREAMING_SNAKE_CASE = ["""text"""]
def lowercase_ ( self : Optional[Any], _snake_case : str ) ->Any:
return self.pre_processor(_snake_case, return_tensors='pt', truncation=_snake_case )
def lowercase_ ( self : int, _snake_case : List[Any] ) ->Any:
return self.model.generate(**_snake_case )[0]
def lowercase_ ( self : int, _snake_case : int ) ->str:
return self.pre_processor.decode(_snake_case, skip_special_tokens=_snake_case, clean_up_tokenization_spaces=_snake_case )
| 277 | 0 |
# We ignore warnings about stepping the scheduler since we step it ourselves during gradient accumulation
import warnings
from .state import AcceleratorState, GradientState
warnings.filterwarnings("ignore", category=UserWarning, module="torch.optim.lr_scheduler")
class UpperCamelCase_ :
'''simple docstring'''
def __init__( self , a , a , a = True , a = False ) -> int:
snake_case_ = scheduler
snake_case_ = optimizers if isinstance(_snake_case , (list, tuple) ) else [optimizers]
snake_case_ = split_batches
snake_case_ = step_with_optimizer
snake_case_ = GradientState()
def _UpperCamelCase ( self , *a , **a ) -> Tuple:
if not self.step_with_optimizer:
# No link between scheduler and optimizer -> just step
self.scheduler.step(*_snake_case , **_snake_case )
return
# Otherwise, first make sure the optimizer was stepped.
if not self.gradient_state.sync_gradients:
if self.gradient_state.adjust_scheduler:
self.scheduler._step_count += 1
return
for opt in self.optimizers:
if opt.step_was_skipped:
return
if self.split_batches:
# Split batches -> the training dataloader batch size is not changed so one step per training step
self.scheduler.step(*_snake_case , **_snake_case )
else:
# Otherwise the training dataloader batch size was multiplied by `num_processes`, so we need to do
# num_processes steps per training step
snake_case_ = AcceleratorState().num_processes
for _ in range(_snake_case ):
# Special case when using OneCycle and `drop_last` was not used
if hasattr(self.scheduler , 'total_steps' ):
if self.scheduler._step_count <= self.scheduler.total_steps:
self.scheduler.step(*_snake_case , **_snake_case )
else:
self.scheduler.step(*_snake_case , **_snake_case )
def _UpperCamelCase ( self ) -> List[str]:
return self.scheduler.get_last_lr()
def _UpperCamelCase ( self ) -> int:
return self.scheduler.state_dict()
def _UpperCamelCase ( self , a ) -> Optional[int]:
self.scheduler.load_state_dict(_snake_case )
def _UpperCamelCase ( self ) -> Union[str, Any]:
return self.scheduler.get_lr()
def _UpperCamelCase ( self , *a , **a ) -> Optional[Any]:
return self.scheduler.print_lr(*_snake_case , **_snake_case )
| 178 |
import argparse
import torch
from transformers import LxmertConfig, LxmertForPreTraining, load_tf_weights_in_lxmert
from transformers.utils import logging
logging.set_verbosity_info()
def lowercase_ (A : str , A : List[Any] , A : Any ):
# Initialise PyTorch model
snake_case__ : List[Any] = LxmertConfig.from_json_file(A )
print(F'''Building PyTorch model from configuration: {config}''' )
snake_case__ : List[str] = LxmertForPreTraining(A )
# Load weights from tf checkpoint
load_tf_weights_in_lxmert(A , A , A )
# Save pytorch-model
print(F'''Save PyTorch model to {pytorch_dump_path}''' )
torch.save(model.state_dict() , A )
if __name__ == "__main__":
a_ :Union[str, Any] = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--tf_checkpoint_path", default=None, type=str, required=True, help="Path to the TensorFlow checkpoint path."
)
parser.add_argument(
"--config_file",
default=None,
type=str,
required=True,
help="The config json file corresponding to the pre-trained model. \nThis specifies the model architecture.",
)
parser.add_argument(
"--pytorch_dump_path", default=None, type=str, required=True, help="Path to the output PyTorch model."
)
a_ :Optional[int] = parser.parse_args()
convert_tf_checkpoint_to_pytorch(args.tf_checkpoint_path, args.config_file, args.pytorch_dump_path)
| 277 | 0 |
"""simple docstring"""
__UpperCamelCase = 8.3_144_598
def lowercase (SCREAMING_SNAKE_CASE_ : float , SCREAMING_SNAKE_CASE_ : float ) -> List[str]:
if temperature < 0:
raise Exception('Temperature cannot be less than 0 K' )
if molar_mass <= 0:
raise Exception('Molar mass cannot be less than or equal to 0 kg/mol' )
else:
return (3 * UNIVERSAL_GAS_CONSTANT * temperature / molar_mass) ** 0.5
if __name__ == "__main__":
import doctest
# run doctest
doctest.testmod()
# example
__UpperCamelCase = 300
__UpperCamelCase = 28
__UpperCamelCase = rms_speed_of_molecule(temperature, molar_mass)
print(f'''Vrms of Nitrogen gas at 300 K is {vrms} m/s''')
| 113 |
import argparse
import fairseq
import torch
from torch import nn
from transformers import (
MBartaaTokenizer,
MBartConfig,
MBartForCausalLM,
SpeechEncoderDecoderConfig,
SpeechEncoderDecoderModel,
WavaVecaConfig,
WavaVecaFeatureExtractor,
WavaVecaModel,
logging,
)
logging.set_verbosity_info()
a_ :Tuple = logging.get_logger(__name__)
a_ :List[Any] = {
"post_extract_proj": "feature_projection.projection",
"encoder.pos_conv.0": "encoder.pos_conv_embed.conv",
"self_attn.k_proj": "encoder.layers.*.attention.k_proj",
"self_attn.v_proj": "encoder.layers.*.attention.v_proj",
"self_attn.q_proj": "encoder.layers.*.attention.q_proj",
"self_attn.out_proj": "encoder.layers.*.attention.out_proj",
"self_attn_layer_norm": "encoder.layers.*.layer_norm",
"fc1": "encoder.layers.*.feed_forward.intermediate_dense",
"fc2": "encoder.layers.*.feed_forward.output_dense",
"final_layer_norm": "encoder.layers.*.final_layer_norm",
"encoder.layer_norm": "encoder.layer_norm",
"w2v_model.layer_norm": "feature_projection.layer_norm",
"quantizer.weight_proj": "quantizer.weight_proj",
"quantizer.vars": "quantizer.codevectors",
"project_q": "project_q",
"final_proj": "project_hid",
"w2v_encoder.proj": "lm_head",
"mask_emb": "masked_spec_embed",
}
a_ :Optional[int] = [
"lm_head",
"quantizer.weight_proj",
"quantizer.codevectors",
"project_q",
"project_hid",
]
def lowercase_ (A : Union[str, Any] , A : str , A : Dict , A : Optional[Any] , A : Optional[Any] ):
for attribute in key.split('.' ):
snake_case__ : Any = getattr(A , A )
if weight_type is not None:
snake_case__ : Optional[Any] = getattr(A , A ).shape
else:
snake_case__ : Optional[int] = hf_pointer.shape
assert hf_shape == value.shape, (
F'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
F''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
snake_case__ : Tuple = value
elif weight_type == "weight_g":
snake_case__ : Tuple = value
elif weight_type == "weight_v":
snake_case__ : List[Any] = value
elif weight_type == "bias":
snake_case__ : List[Any] = value
else:
snake_case__ : Optional[Any] = value
logger.info(F'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def lowercase_ (A : str , A : Any ):
snake_case__ : Union[str, Any] = []
snake_case__ : Union[str, Any] = fairseq_model.state_dict()
snake_case__ : Union[str, Any] = hf_model.feature_extractor
snake_case__ : Any = hf_model.adapter
for name, value in fairseq_dict.items():
snake_case__ : Any = False
if "conv_layers" in name:
load_conv_layer(
A , A , A , A , hf_model.config.feat_extract_norm == 'group' , )
snake_case__ : List[Any] = True
elif any(x in name for x in ['adaptor', 'w2v_encoder.proj.', 'w2v_proj_ln.'] ):
load_adapter(A , A , A , A )
snake_case__ : Optional[Any] = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split('w2v_model.' )[-1] == name.split('.' )[0]:
snake_case__ : Tuple = True
if "*" in mapped_key:
snake_case__ : List[Any] = name.split(A )[0].split('.' )[-2]
snake_case__ : Optional[int] = mapped_key.replace('*' , A )
if "weight_g" in name:
snake_case__ : Optional[int] = 'weight_g'
elif "weight_v" in name:
snake_case__ : Optional[Any] = 'weight_v'
elif "bias" in name:
snake_case__ : Union[str, Any] = 'bias'
elif "weight" in name:
snake_case__ : Optional[int] = 'weight'
else:
snake_case__ : Tuple = None
set_recursively(A , A , A , A , A )
continue
if not is_used:
unused_weights.append(A )
logger.warning(F'''Unused weights: {unused_weights}''' )
def lowercase_ (A : Union[str, Any] , A : Any , A : str , A : str , A : int ):
snake_case__ : str = full_name.split('conv_layers.' )[-1]
snake_case__ : Optional[int] = name.split('.' )
snake_case__ : Tuple = int(items[0] )
snake_case__ : Any = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
snake_case__ : Union[str, Any] = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
snake_case__ : Union[str, Any] = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
F'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
snake_case__ : Optional[int] = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
snake_case__ : Optional[Any] = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(A )
def lowercase_ (A : Optional[Any] , A : Any , A : Tuple , A : Any ):
snake_case__ : List[str] = full_name.split('adaptor.' )[-1]
snake_case__ : Tuple = name.split('.' )
if items[1].isdigit():
snake_case__ : Optional[int] = int(items[1] )
else:
snake_case__ : Any = None
if "adaptor" not in full_name:
if "proj_ln" in full_name:
# has to be layer norm
if "bias" in name:
assert (
value.shape == adapter.proj_layer_norm.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj_layer_norm.bias.data.shape} was found.'''
snake_case__ : List[Any] = value
logger.info(F'''Adapter proj layer norm bias was initialized from {full_name}.''' )
if "weight" in name:
assert (
value.shape == adapter.proj_layer_norm.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj_layer_norm.weight.data.shape} was found.'''
snake_case__ : int = value
else:
# has to be projection layer
if "bias" in name:
assert (
value.shape == adapter.proj.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj.bias.data.shape} was found.'''
snake_case__ : str = value
logger.info(F'''Adapter proj layer bias was initialized from {full_name}.''' )
if "weight" in name:
assert (
value.shape == adapter.proj.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj.weight.data.shape} was found.'''
snake_case__ : Dict = value
logger.info(F'''Adapter proj layer weight was initialized from {full_name}.''' )
elif isinstance(A , A ):
if "bias" in name:
assert (
value.shape == adapter.layers[layer_id].conv.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.layers[layer_id].conv.bias.data.shape} was found.'''
snake_case__ : List[str] = value
logger.info(F'''Adapter layer {layer_id} bias was initialized from {full_name}.''' )
elif "weight" in name:
assert (
value.shape == adapter.layers[layer_id].conv.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.layers[layer_id].conv.weight.data.shape} was found.'''
snake_case__ : List[str] = value
logger.info(F'''Adapter layer {layer_id} bias was initialized from {full_name}.''' )
else:
unused_weights.append(A )
def lowercase_ (A : int ):
snake_case__ , snake_case__ : Union[str, Any] = emb.weight.shape
snake_case__ : int = nn.Linear(A , A , bias=A )
snake_case__ : Optional[Any] = emb.weight.data
return lin_layer
@torch.no_grad()
def lowercase_ (A : Tuple , A : Tuple , A : Any , A : Optional[Any] , A : int , A : Optional[Any] , A : Union[str, Any] , A : Union[str, Any] , A : Optional[Any] , A : List[Any] , A : Union[str, Any] , ):
snake_case__ : Optional[Any] = WavaVecaConfig.from_pretrained(
A , add_adapter=A , adapter_stride=A , adapter_kernel_size=A , use_auth_token=A , output_hidden_size=A , )
snake_case__ : Dict = MBartConfig.from_pretrained(A )
# load model
snake_case__ , snake_case__ , snake_case__ : Any = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] , arg_overrides={
'config_yaml': config_yaml_path,
'data': '/'.join(dict_path.split('/' )[:-1] ),
'w2v_path': checkpoint_path,
'load_pretrained_decoder_from': None,
} , )
snake_case__ : List[Any] = model[0].eval()
# load feature extractor
snake_case__ : str = WavaVecaFeatureExtractor.from_pretrained(A , use_auth_token=A )
# set weights for wav2vec2 encoder
snake_case__ : List[str] = WavaVecaModel(A )
recursively_load_weights_wavaveca(model.encoder , A )
# load decoder weights
snake_case__ : Any = MBartForCausalLM(A )
snake_case__ , snake_case__ : int = hf_decoder.model.decoder.load_state_dict(model.decoder.state_dict() , strict=A )
logger.warning(F'''The following keys are missing when loading the decoder weights: {missing_keys}''' )
logger.warning(F'''The following keys are unexpected when loading the decoder weights: {unexpected_keys}''' )
snake_case__ : Union[str, Any] = SpeechEncoderDecoderModel(encoder=A , decoder=A )
snake_case__ : str = False
snake_case__ : int = MBartaaTokenizer(A )
tokenizer.save_pretrained(A )
snake_case__ : Any = hf_wavavec.config.to_dict()
snake_case__ : Tuple = tokenizer.pad_token_id
snake_case__ : Union[str, Any] = tokenizer.bos_token_id
snake_case__ : Dict = tokenizer.eos_token_id
snake_case__ : Optional[int] = 'mbart50'
snake_case__ : Union[str, Any] = 'wav2vec2'
snake_case__ : List[str] = tokenizer.eos_token_id
snake_case__ : Union[str, Any] = 2_5_0_0_0_4
snake_case__ : int = tokenizer.eos_token_id
snake_case__ : Union[str, Any] = SpeechEncoderDecoderConfig.from_dict(A )
hf_wavavec.save_pretrained(A )
feature_extractor.save_pretrained(A )
if __name__ == "__main__":
a_ :str = argparse.ArgumentParser()
parser.add_argument("--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model.")
parser.add_argument("--checkpoint_path", default=None, type=str, help="Path to fairseq checkpoint")
parser.add_argument("--dict_path", default=None, type=str, help="Path to dict of fine-tuned model")
parser.add_argument("--config_yaml_path", default=None, type=str, help="Path to yaml file of fine-tuned model")
parser.add_argument(
"--encoder_config_path",
default="facebook/wav2vec2-xls-r-1b",
type=str,
help="Path to hf encoder wav2vec2 checkpoint config",
)
parser.add_argument(
"--decoder_config_path",
default="facebook/mbart-large-50-one-to-many-mmt",
type=str,
help="Path to hf decoder checkpoint config",
)
parser.add_argument("--add_adapter", default=True, type=bool, help="whethere to add model adapter layers")
parser.add_argument("--adapter_stride", default=2, type=int, help="stride of adapter layers")
parser.add_argument("--adapter_kernel_size", default=3, type=int, help="kernel size of adapter layers")
parser.add_argument("--encoder_output_dim", default=1_024, type=int, help="encoder output dim")
parser.add_argument("--start_token_id", default=250_004, type=int, help="`decoder_start_token_id` of model config")
a_ :Union[str, Any] = parser.parse_args()
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.dict_path,
args.config_yaml_path,
encoder_config_path=args.encoder_config_path,
decoder_config_path=args.decoder_config_path,
add_adapter=args.add_adapter,
adapter_kernel_size=args.adapter_kernel_size,
adapter_stride=args.adapter_stride,
decoder_start_token_id=args.start_token_id,
encoder_output_dim=args.encoder_output_dim,
)
| 277 | 0 |
'''simple docstring'''
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tokenizers_available, is_torch_available
_A : Union[str, Any] = {
"configuration_squeezebert": [
"SQUEEZEBERT_PRETRAINED_CONFIG_ARCHIVE_MAP",
"SqueezeBertConfig",
"SqueezeBertOnnxConfig",
],
"tokenization_squeezebert": ["SqueezeBertTokenizer"],
}
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
_A : Union[str, Any] = ["SqueezeBertTokenizerFast"]
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
_A : List[str] = [
"SQUEEZEBERT_PRETRAINED_MODEL_ARCHIVE_LIST",
"SqueezeBertForMaskedLM",
"SqueezeBertForMultipleChoice",
"SqueezeBertForQuestionAnswering",
"SqueezeBertForSequenceClassification",
"SqueezeBertForTokenClassification",
"SqueezeBertModel",
"SqueezeBertModule",
"SqueezeBertPreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_squeezebert import (
SQUEEZEBERT_PRETRAINED_CONFIG_ARCHIVE_MAP,
SqueezeBertConfig,
SqueezeBertOnnxConfig,
)
from .tokenization_squeezebert import SqueezeBertTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_squeezebert_fast import SqueezeBertTokenizerFast
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_squeezebert import (
SQUEEZEBERT_PRETRAINED_MODEL_ARCHIVE_LIST,
SqueezeBertForMaskedLM,
SqueezeBertForMultipleChoice,
SqueezeBertForQuestionAnswering,
SqueezeBertForSequenceClassification,
SqueezeBertForTokenClassification,
SqueezeBertModel,
SqueezeBertModule,
SqueezeBertPreTrainedModel,
)
else:
import sys
_A : Dict = _LazyModule(__name__, globals()['''__file__'''], _import_structure, module_spec=__spec__)
| 229 |
from collections import OrderedDict
from typing import TYPE_CHECKING, Any, Mapping, Optional, Union
from ...configuration_utils import PretrainedConfig
from ...onnx import OnnxConfig
from ...utils import logging
if TYPE_CHECKING:
from ... import FeatureExtractionMixin, PreTrainedTokenizerBase, TensorType
a_ :Tuple = logging.get_logger(__name__)
a_ :Union[str, Any] = {
"microsoft/deberta-v2-xlarge": "https://huggingface.co/microsoft/deberta-v2-xlarge/resolve/main/config.json",
"microsoft/deberta-v2-xxlarge": "https://huggingface.co/microsoft/deberta-v2-xxlarge/resolve/main/config.json",
"microsoft/deberta-v2-xlarge-mnli": (
"https://huggingface.co/microsoft/deberta-v2-xlarge-mnli/resolve/main/config.json"
),
"microsoft/deberta-v2-xxlarge-mnli": (
"https://huggingface.co/microsoft/deberta-v2-xxlarge-mnli/resolve/main/config.json"
),
}
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """deberta-v2"""
def __init__( self : Union[str, Any], _snake_case : Dict=1_2_8_1_0_0, _snake_case : Any=1_5_3_6, _snake_case : Tuple=2_4, _snake_case : int=2_4, _snake_case : Optional[int]=6_1_4_4, _snake_case : Optional[int]="gelu", _snake_case : Optional[int]=0.1, _snake_case : List[str]=0.1, _snake_case : str=5_1_2, _snake_case : Optional[int]=0, _snake_case : Optional[int]=0.0_2, _snake_case : Dict=1e-7, _snake_case : int=False, _snake_case : Any=-1, _snake_case : List[str]=0, _snake_case : Tuple=True, _snake_case : Any=None, _snake_case : Union[str, Any]=0, _snake_case : Tuple="gelu", **_snake_case : Union[str, Any], ) ->Optional[int]:
super().__init__(**_snake_case )
snake_case__ : Dict = hidden_size
snake_case__ : Optional[int] = num_hidden_layers
snake_case__ : Any = num_attention_heads
snake_case__ : List[Any] = intermediate_size
snake_case__ : List[Any] = hidden_act
snake_case__ : Union[str, Any] = hidden_dropout_prob
snake_case__ : Dict = attention_probs_dropout_prob
snake_case__ : List[str] = max_position_embeddings
snake_case__ : List[str] = type_vocab_size
snake_case__ : Optional[Any] = initializer_range
snake_case__ : Optional[int] = relative_attention
snake_case__ : Tuple = max_relative_positions
snake_case__ : Union[str, Any] = pad_token_id
snake_case__ : Optional[int] = position_biased_input
# Backwards compatibility
if type(_snake_case ) == str:
snake_case__ : int = [x.strip() for x in pos_att_type.lower().split('|' )]
snake_case__ : List[str] = pos_att_type
snake_case__ : Union[str, Any] = vocab_size
snake_case__ : Optional[int] = layer_norm_eps
snake_case__ : Optional[int] = kwargs.get('pooler_hidden_size', _snake_case )
snake_case__ : int = pooler_dropout
snake_case__ : str = pooler_hidden_act
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
@property
def lowercase_ ( self : Optional[int] ) ->Mapping[str, Mapping[int, str]]:
if self.task == "multiple-choice":
snake_case__ : List[Any] = {0: 'batch', 1: 'choice', 2: 'sequence'}
else:
snake_case__ : int = {0: 'batch', 1: 'sequence'}
if self._config.type_vocab_size > 0:
return OrderedDict(
[('input_ids', dynamic_axis), ('attention_mask', dynamic_axis), ('token_type_ids', dynamic_axis)] )
else:
return OrderedDict([('input_ids', dynamic_axis), ('attention_mask', dynamic_axis)] )
@property
def lowercase_ ( self : Dict ) ->int:
return 1_2
def lowercase_ ( self : Tuple, _snake_case : Union["PreTrainedTokenizerBase", "FeatureExtractionMixin"], _snake_case : int = -1, _snake_case : int = -1, _snake_case : int = -1, _snake_case : bool = False, _snake_case : Optional["TensorType"] = None, _snake_case : int = 3, _snake_case : int = 4_0, _snake_case : int = 4_0, _snake_case : "PreTrainedTokenizerBase" = None, ) ->Mapping[str, Any]:
snake_case__ : Union[str, Any] = super().generate_dummy_inputs(preprocessor=_snake_case, framework=_snake_case )
if self._config.type_vocab_size == 0 and "token_type_ids" in dummy_inputs:
del dummy_inputs["token_type_ids"]
return dummy_inputs
| 277 | 0 |
"""simple docstring"""
from typing import Dict
import numpy as np
import torch
from . import residue_constants as rc
from .tensor_utils import tensor_tree_map, tree_map
def lowercase ( a__ : Dict[str, torch.Tensor] ) -> List[Any]:
_UpperCamelCase = []
_UpperCamelCase = []
_UpperCamelCase = []
for rt in rc.restypes:
_UpperCamelCase = rc.restype_name_to_atomaa_names[rc.restype_atoa[rt]]
restype_atomaa_to_atomaa_list.append([(rc.atom_order[name] if name else 0) for name in atom_names] )
_UpperCamelCase = {name: i for i, name in enumerate(a__ )}
restype_atomaa_to_atomaa_list.append(
[(atom_name_to_idxaa[name] if name in atom_name_to_idxaa else 0) for name in rc.atom_types] )
restype_atomaa_mask_list.append([(1.0 if name else 0.0) for name in atom_names] )
# Add dummy mapping for restype 'UNK'
restype_atomaa_to_atomaa_list.append([0] * 14 )
restype_atomaa_to_atomaa_list.append([0] * 37 )
restype_atomaa_mask_list.append([0.0] * 14 )
_UpperCamelCase = torch.tensor(
a__ , dtype=torch.intaa , device=protein['''aatype'''].device , )
_UpperCamelCase = torch.tensor(
a__ , dtype=torch.intaa , device=protein['''aatype'''].device , )
_UpperCamelCase = torch.tensor(
a__ , dtype=torch.floataa , device=protein['''aatype'''].device , )
_UpperCamelCase = protein['aatype'].to(torch.long )
# create the mapping for (residx, atom14) --> atom37, i.e. an array
# with shape (num_res, 14) containing the atom37 indices for this protein
_UpperCamelCase = restype_atomaa_to_atomaa[protein_aatype]
_UpperCamelCase = restype_atomaa_mask[protein_aatype]
_UpperCamelCase = residx_atomaa_mask
_UpperCamelCase = residx_atomaa_to_atomaa.long()
# create the gather indices for mapping back
_UpperCamelCase = restype_atomaa_to_atomaa[protein_aatype]
_UpperCamelCase = residx_atomaa_to_atomaa.long()
# create the corresponding mask
_UpperCamelCase = torch.zeros([21, 37] , dtype=torch.floataa , device=protein['''aatype'''].device )
for restype, restype_letter in enumerate(rc.restypes ):
_UpperCamelCase = rc.restype_atoa[restype_letter]
_UpperCamelCase = rc.residue_atoms[restype_name]
for atom_name in atom_names:
_UpperCamelCase = rc.atom_order[atom_name]
_UpperCamelCase = 1
_UpperCamelCase = restype_atomaa_mask[protein_aatype]
_UpperCamelCase = residx_atomaa_mask
return protein
def lowercase ( a__ : Dict[str, torch.Tensor] ) -> Any:
_UpperCamelCase = tree_map(lambda a__ : torch.tensor(a__ , device=batch['''aatype'''].device ) , a__ , np.ndarray )
_UpperCamelCase = tensor_tree_map(lambda a__ : np.array(a__ ) , make_atomaa_masks(a__ ) )
return out
| 256 |
import argparse
import json
import pickle
from pathlib import Path
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import MaskFormerConfig, MaskFormerForInstanceSegmentation, MaskFormerImageProcessor, SwinConfig
from transformers.utils import logging
logging.set_verbosity_info()
a_ :str = logging.get_logger(__name__)
def lowercase_ (A : str ):
snake_case__ : Tuple = SwinConfig.from_pretrained(
'microsoft/swin-tiny-patch4-window7-224' , out_features=['stage1', 'stage2', 'stage3', 'stage4'] )
snake_case__ : List[Any] = MaskFormerConfig(backbone_config=A )
snake_case__ : Union[str, Any] = 'huggingface/label-files'
if "ade20k-full" in model_name:
# this should be ok
snake_case__ : Dict = 8_4_7
snake_case__ : List[str] = 'maskformer-ade20k-full-id2label.json'
elif "ade" in model_name:
# this should be ok
snake_case__ : Union[str, Any] = 1_5_0
snake_case__ : Any = 'ade20k-id2label.json'
elif "coco-stuff" in model_name:
# this should be ok
snake_case__ : List[str] = 1_7_1
snake_case__ : Union[str, Any] = 'maskformer-coco-stuff-id2label.json'
elif "coco" in model_name:
# TODO
snake_case__ : Dict = 1_3_3
snake_case__ : str = 'coco-panoptic-id2label.json'
elif "cityscapes" in model_name:
# this should be ok
snake_case__ : List[str] = 1_9
snake_case__ : Union[str, Any] = 'cityscapes-id2label.json'
elif "vistas" in model_name:
# this should be ok
snake_case__ : Tuple = 6_5
snake_case__ : List[str] = 'mapillary-vistas-id2label.json'
snake_case__ : Dict = json.load(open(hf_hub_download(A , A , repo_type='dataset' ) , 'r' ) )
snake_case__ : List[str] = {int(A ): v for k, v in idalabel.items()}
return config
def lowercase_ (A : Any ):
snake_case__ : Optional[int] = []
# stem
# fmt: off
rename_keys.append(('backbone.patch_embed.proj.weight', 'model.pixel_level_module.encoder.model.embeddings.patch_embeddings.projection.weight') )
rename_keys.append(('backbone.patch_embed.proj.bias', 'model.pixel_level_module.encoder.model.embeddings.patch_embeddings.projection.bias') )
rename_keys.append(('backbone.patch_embed.norm.weight', 'model.pixel_level_module.encoder.model.embeddings.norm.weight') )
rename_keys.append(('backbone.patch_embed.norm.bias', 'model.pixel_level_module.encoder.model.embeddings.norm.bias') )
# stages
for i in range(len(config.backbone_config.depths ) ):
for j in range(config.backbone_config.depths[i] ):
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm1.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_before.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm1.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_before.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.relative_position_bias_table''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.self.relative_position_bias_table''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.relative_position_index''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.self.relative_position_index''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.proj.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.output.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.proj.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.output.dense.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm2.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_after.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm2.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_after.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc1.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.intermediate.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc1.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.intermediate.dense.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc2.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.output.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc2.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.output.dense.bias''') )
if i < 3:
rename_keys.append((F'''backbone.layers.{i}.downsample.reduction.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.reduction.weight''') )
rename_keys.append((F'''backbone.layers.{i}.downsample.norm.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.norm.weight''') )
rename_keys.append((F'''backbone.layers.{i}.downsample.norm.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.norm.bias''') )
rename_keys.append((F'''backbone.norm{i}.weight''', F'''model.pixel_level_module.encoder.hidden_states_norms.{i}.weight''') )
rename_keys.append((F'''backbone.norm{i}.bias''', F'''model.pixel_level_module.encoder.hidden_states_norms.{i}.bias''') )
# FPN
rename_keys.append(('sem_seg_head.layer_4.weight', 'model.pixel_level_module.decoder.fpn.stem.0.weight') )
rename_keys.append(('sem_seg_head.layer_4.norm.weight', 'model.pixel_level_module.decoder.fpn.stem.1.weight') )
rename_keys.append(('sem_seg_head.layer_4.norm.bias', 'model.pixel_level_module.decoder.fpn.stem.1.bias') )
for source_index, target_index in zip(range(3 , 0 , -1 ) , range(0 , 3 ) ):
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.0.weight''') )
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.norm.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.1.weight''') )
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.norm.bias''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.1.bias''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.0.weight''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.norm.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.1.weight''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.norm.bias''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.1.bias''') )
rename_keys.append(('sem_seg_head.mask_features.weight', 'model.pixel_level_module.decoder.mask_projection.weight') )
rename_keys.append(('sem_seg_head.mask_features.bias', 'model.pixel_level_module.decoder.mask_projection.bias') )
# Transformer decoder
for idx in range(config.decoder_config.decoder_layers ):
# self-attention out projection
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.out_proj.weight''', F'''model.transformer_module.decoder.layers.{idx}.self_attn.out_proj.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.out_proj.bias''', F'''model.transformer_module.decoder.layers.{idx}.self_attn.out_proj.bias''') )
# cross-attention out projection
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.out_proj.weight''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn.out_proj.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.out_proj.bias''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn.out_proj.bias''') )
# MLP 1
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear1.weight''', F'''model.transformer_module.decoder.layers.{idx}.fc1.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear1.bias''', F'''model.transformer_module.decoder.layers.{idx}.fc1.bias''') )
# MLP 2
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear2.weight''', F'''model.transformer_module.decoder.layers.{idx}.fc2.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear2.bias''', F'''model.transformer_module.decoder.layers.{idx}.fc2.bias''') )
# layernorm 1 (self-attention layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm1.weight''', F'''model.transformer_module.decoder.layers.{idx}.self_attn_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm1.bias''', F'''model.transformer_module.decoder.layers.{idx}.self_attn_layer_norm.bias''') )
# layernorm 2 (cross-attention layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm2.weight''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm2.bias''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn_layer_norm.bias''') )
# layernorm 3 (final layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm3.weight''', F'''model.transformer_module.decoder.layers.{idx}.final_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm3.bias''', F'''model.transformer_module.decoder.layers.{idx}.final_layer_norm.bias''') )
rename_keys.append(('sem_seg_head.predictor.transformer.decoder.norm.weight', 'model.transformer_module.decoder.layernorm.weight') )
rename_keys.append(('sem_seg_head.predictor.transformer.decoder.norm.bias', 'model.transformer_module.decoder.layernorm.bias') )
# heads on top
rename_keys.append(('sem_seg_head.predictor.query_embed.weight', 'model.transformer_module.queries_embedder.weight') )
rename_keys.append(('sem_seg_head.predictor.input_proj.weight', 'model.transformer_module.input_projection.weight') )
rename_keys.append(('sem_seg_head.predictor.input_proj.bias', 'model.transformer_module.input_projection.bias') )
rename_keys.append(('sem_seg_head.predictor.class_embed.weight', 'class_predictor.weight') )
rename_keys.append(('sem_seg_head.predictor.class_embed.bias', 'class_predictor.bias') )
for i in range(3 ):
rename_keys.append((F'''sem_seg_head.predictor.mask_embed.layers.{i}.weight''', F'''mask_embedder.{i}.0.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.mask_embed.layers.{i}.bias''', F'''mask_embedder.{i}.0.bias''') )
# fmt: on
return rename_keys
def lowercase_ (A : Tuple , A : Tuple , A : Optional[Any] ):
snake_case__ : Optional[int] = dct.pop(A )
snake_case__ : Union[str, Any] = val
def lowercase_ (A : Optional[Any] , A : Tuple ):
snake_case__ : Optional[int] = [int(backbone_config.embed_dim * 2**i ) for i in range(len(backbone_config.depths ) )]
for i in range(len(backbone_config.depths ) ):
snake_case__ : Optional[int] = num_features[i]
for j in range(backbone_config.depths[i] ):
# fmt: off
# read in weights + bias of input projection layer (in original implementation, this is a single matrix + bias)
snake_case__ : int = state_dict.pop(F'''backbone.layers.{i}.blocks.{j}.attn.qkv.weight''' )
snake_case__ : Tuple = state_dict.pop(F'''backbone.layers.{i}.blocks.{j}.attn.qkv.bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : str = in_proj_weight[:dim, :]
snake_case__ : int = in_proj_bias[: dim]
snake_case__ : List[Any] = in_proj_weight[
dim : dim * 2, :
]
snake_case__ : List[str] = in_proj_bias[
dim : dim * 2
]
snake_case__ : List[Any] = in_proj_weight[
-dim :, :
]
snake_case__ : Dict = in_proj_bias[-dim :]
# fmt: on
def lowercase_ (A : List[str] , A : List[Any] ):
# fmt: off
snake_case__ : str = config.decoder_config.hidden_size
for idx in range(config.decoder_config.decoder_layers ):
# read in weights + bias of self-attention input projection layer (in the original implementation, this is a single matrix + bias)
snake_case__ : List[Any] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.in_proj_weight''' )
snake_case__ : int = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.in_proj_bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : Any = in_proj_weight[: hidden_size, :]
snake_case__ : Tuple = in_proj_bias[:config.hidden_size]
snake_case__ : List[str] = in_proj_weight[hidden_size : hidden_size * 2, :]
snake_case__ : Dict = in_proj_bias[hidden_size : hidden_size * 2]
snake_case__ : Any = in_proj_weight[-hidden_size :, :]
snake_case__ : int = in_proj_bias[-hidden_size :]
# read in weights + bias of cross-attention input projection layer (in the original implementation, this is a single matrix + bias)
snake_case__ : List[Any] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.in_proj_weight''' )
snake_case__ : List[str] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.in_proj_bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : Optional[int] = in_proj_weight[: hidden_size, :]
snake_case__ : Optional[Any] = in_proj_bias[:config.hidden_size]
snake_case__ : int = in_proj_weight[hidden_size : hidden_size * 2, :]
snake_case__ : List[str] = in_proj_bias[hidden_size : hidden_size * 2]
snake_case__ : List[str] = in_proj_weight[-hidden_size :, :]
snake_case__ : str = in_proj_bias[-hidden_size :]
# fmt: on
def lowercase_ ():
snake_case__ : Any = 'http://images.cocodataset.org/val2017/000000039769.jpg'
snake_case__ : int = Image.open(requests.get(A , stream=A ).raw )
return im
@torch.no_grad()
def lowercase_ (A : str , A : str , A : str , A : bool = False ):
snake_case__ : Optional[int] = get_maskformer_config(A )
# load original state_dict
with open(A , 'rb' ) as f:
snake_case__ : List[Any] = pickle.load(A )
snake_case__ : Optional[int] = data['model']
# for name, param in state_dict.items():
# print(name, param.shape)
# rename keys
snake_case__ : List[str] = create_rename_keys(A )
for src, dest in rename_keys:
rename_key(A , A , A )
read_in_swin_q_k_v(A , config.backbone_config )
read_in_decoder_q_k_v(A , A )
# update to torch tensors
for key, value in state_dict.items():
snake_case__ : int = torch.from_numpy(A )
# load 🤗 model
snake_case__ : str = MaskFormerForInstanceSegmentation(A )
model.eval()
for name, param in model.named_parameters():
print(A , param.shape )
snake_case__ , snake_case__ : Union[str, Any] = model.load_state_dict(A , strict=A )
assert missing_keys == [
"model.pixel_level_module.encoder.model.layernorm.weight",
"model.pixel_level_module.encoder.model.layernorm.bias",
]
assert len(A ) == 0, F'''Unexpected keys: {unexpected_keys}'''
# verify results
snake_case__ : Optional[Any] = prepare_img()
if "vistas" in model_name:
snake_case__ : int = 6_5
elif "cityscapes" in model_name:
snake_case__ : Dict = 6_5_5_3_5
else:
snake_case__ : Tuple = 2_5_5
snake_case__ : Optional[int] = True if 'ade' in model_name else False
snake_case__ : Dict = MaskFormerImageProcessor(ignore_index=A , reduce_labels=A )
snake_case__ : Any = image_processor(A , return_tensors='pt' )
snake_case__ : Any = model(**A )
print('Logits:' , outputs.class_queries_logits[0, :3, :3] )
if model_name == "maskformer-swin-tiny-ade":
snake_case__ : Tuple = torch.tensor(
[[3.6353, -4.4770, -2.6065], [0.5081, -4.2394, -3.5343], [2.1909, -5.0353, -1.9323]] )
assert torch.allclose(outputs.class_queries_logits[0, :3, :3] , A , atol=1e-4 )
print('Looks ok!' )
if pytorch_dump_folder_path is not None:
print(F'''Saving model and image processor to {pytorch_dump_folder_path}''' )
Path(A ).mkdir(exist_ok=A )
model.save_pretrained(A )
image_processor.save_pretrained(A )
if push_to_hub:
print('Pushing model and image processor to the hub...' )
model.push_to_hub(F'''nielsr/{model_name}''' )
image_processor.push_to_hub(F'''nielsr/{model_name}''' )
if __name__ == "__main__":
a_ :Optional[int] = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--model_name",
default="maskformer-swin-tiny-ade",
type=str,
help=("Name of the MaskFormer model you'd like to convert",),
)
parser.add_argument(
"--checkpoint_path",
default="/Users/nielsrogge/Documents/MaskFormer_checkpoints/MaskFormer-Swin-tiny-ADE20k/model.pkl",
type=str,
help="Path to the original state dict (.pth file).",
)
parser.add_argument(
"--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model directory."
)
parser.add_argument(
"--push_to_hub", action="store_true", help="Whether or not to push the converted model to the 🤗 hub."
)
a_ :Dict = parser.parse_args()
convert_maskformer_checkpoint(
args.model_name, args.checkpoint_path, args.pytorch_dump_folder_path, args.push_to_hub
)
| 277 | 0 |
from __future__ import annotations
from math import ceil, floor, sqrt
def _a ( SCREAMING_SNAKE_CASE : int = 2_00_00_00 ) -> Optional[int]:
"""simple docstring"""
__lowerCAmelCase: list[int] = [0]
__lowerCAmelCase: int
for idx in range(1 , ceil(sqrt(target * 2 ) * 1.1 ) ):
triangle_numbers.append(triangle_numbers[-1] + idx )
# we want this to be as close as possible to target
__lowerCAmelCase: int = 0
# the area corresponding to the grid that gives the product closest to target
__lowerCAmelCase: int = 0
# an estimate of b, using the quadratic formula
__lowerCAmelCase: float
# the largest integer less than b_estimate
__lowerCAmelCase: int
# the largest integer less than b_estimate
__lowerCAmelCase: int
# the triangle number corresponding to b_floor
__lowerCAmelCase: int
# the triangle number corresponding to b_ceil
__lowerCAmelCase: int
for idx_a, triangle_a in enumerate(triangle_numbers[1:] , 1 ):
__lowerCAmelCase: Dict = (-1 + sqrt(1 + 8 * target / triangle_a )) / 2
__lowerCAmelCase: Optional[Any] = floor(SCREAMING_SNAKE_CASE )
__lowerCAmelCase: List[Any] = ceil(SCREAMING_SNAKE_CASE )
__lowerCAmelCase: Union[str, Any] = triangle_numbers[b_floor]
__lowerCAmelCase: Any = triangle_numbers[b_ceil]
if abs(target - triangle_b_first_guess * triangle_a ) < abs(
target - best_product ):
__lowerCAmelCase: List[str] = triangle_b_first_guess * triangle_a
__lowerCAmelCase: Dict = idx_a * b_floor
if abs(target - triangle_b_second_guess * triangle_a ) < abs(
target - best_product ):
__lowerCAmelCase: int = triangle_b_second_guess * triangle_a
__lowerCAmelCase: List[Any] = idx_a * b_ceil
return area
if __name__ == "__main__":
print(f"{solution() = }")
| 322 |
import unittest
from transformers import is_torch_available
from transformers.testing_utils import require_torch, slow, torch_device
from ...generation.test_utils import GenerationTesterMixin
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers import (
OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST,
OpenAIGPTConfig,
OpenAIGPTDoubleHeadsModel,
OpenAIGPTForSequenceClassification,
OpenAIGPTLMHeadModel,
OpenAIGPTModel,
)
class snake_case__ :
"""simple docstring"""
def __init__( self : List[str], _snake_case : Any, _snake_case : int=1_3, _snake_case : Optional[int]=7, _snake_case : int=True, _snake_case : Optional[Any]=True, _snake_case : Optional[Any]=True, _snake_case : Union[str, Any]=9_9, _snake_case : Optional[Any]=3_2, _snake_case : Tuple=5, _snake_case : str=4, _snake_case : Any=3_7, _snake_case : int="gelu", _snake_case : Optional[Any]=0.1, _snake_case : str=0.1, _snake_case : str=5_1_2, _snake_case : Dict=1_6, _snake_case : str=2, _snake_case : Union[str, Any]=0.0_2, _snake_case : Optional[int]=3, _snake_case : Union[str, Any]=4, _snake_case : Tuple=None, ) ->Optional[Any]:
snake_case__ : Optional[int] = parent
snake_case__ : List[Any] = batch_size
snake_case__ : Tuple = seq_length
snake_case__ : str = is_training
snake_case__ : Optional[int] = use_token_type_ids
snake_case__ : Any = use_labels
snake_case__ : Dict = vocab_size
snake_case__ : str = hidden_size
snake_case__ : Union[str, Any] = num_hidden_layers
snake_case__ : List[str] = num_attention_heads
snake_case__ : Union[str, Any] = intermediate_size
snake_case__ : List[Any] = hidden_act
snake_case__ : int = hidden_dropout_prob
snake_case__ : str = attention_probs_dropout_prob
snake_case__ : Any = max_position_embeddings
snake_case__ : Union[str, Any] = type_vocab_size
snake_case__ : Optional[Any] = type_sequence_label_size
snake_case__ : Optional[int] = initializer_range
snake_case__ : Optional[int] = num_labels
snake_case__ : str = num_choices
snake_case__ : int = scope
snake_case__ : List[str] = self.vocab_size - 1
def lowercase_ ( self : Union[str, Any] ) ->Tuple:
snake_case__ : List[str] = ids_tensor([self.batch_size, self.seq_length], self.vocab_size )
snake_case__ : List[str] = None
if self.use_token_type_ids:
snake_case__ : Optional[Any] = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size )
snake_case__ : Tuple = None
snake_case__ : str = None
snake_case__ : List[Any] = None
if self.use_labels:
snake_case__ : Dict = ids_tensor([self.batch_size], self.type_sequence_label_size )
snake_case__ : int = ids_tensor([self.batch_size, self.seq_length], self.num_labels )
snake_case__ : List[str] = ids_tensor([self.batch_size], self.num_choices )
snake_case__ : Union[str, Any] = OpenAIGPTConfig(
vocab_size=self.vocab_size, n_embd=self.hidden_size, n_layer=self.num_hidden_layers, n_head=self.num_attention_heads, n_positions=self.max_position_embeddings, pad_token_id=self.pad_token_id, )
snake_case__ : List[str] = ids_tensor([self.num_hidden_layers, self.num_attention_heads], 2 )
return (
config,
input_ids,
head_mask,
token_type_ids,
sequence_labels,
token_labels,
choice_labels,
)
def lowercase_ ( self : Any, _snake_case : List[str], _snake_case : Any, _snake_case : List[Any], _snake_case : Tuple, *_snake_case : Optional[Any] ) ->Tuple:
snake_case__ : Union[str, Any] = OpenAIGPTModel(config=_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Optional[Any] = model(_snake_case, token_type_ids=_snake_case, head_mask=_snake_case )
snake_case__ : Union[str, Any] = model(_snake_case, token_type_ids=_snake_case )
snake_case__ : Optional[Any] = model(_snake_case )
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size) )
def lowercase_ ( self : Optional[int], _snake_case : Optional[Any], _snake_case : Union[str, Any], _snake_case : Optional[int], _snake_case : List[Any], *_snake_case : Dict ) ->Optional[int]:
snake_case__ : Optional[Any] = OpenAIGPTLMHeadModel(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Tuple = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.loss.shape, () )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size) )
def lowercase_ ( self : int, _snake_case : Tuple, _snake_case : List[str], _snake_case : List[Any], _snake_case : List[Any], *_snake_case : List[Any] ) ->Optional[int]:
snake_case__ : List[str] = OpenAIGPTDoubleHeadsModel(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Optional[Any] = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.loss.shape, () )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size) )
def lowercase_ ( self : Optional[int], _snake_case : Tuple, _snake_case : Dict, _snake_case : List[str], _snake_case : Optional[Any], *_snake_case : Union[str, Any] ) ->str:
snake_case__ : List[str] = self.num_labels
snake_case__ : Dict = OpenAIGPTForSequenceClassification(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : List[str] = ids_tensor([self.batch_size], self.type_sequence_label_size )
snake_case__ : List[str] = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels) )
def lowercase_ ( self : Dict ) ->int:
snake_case__ : List[Any] = self.prepare_config_and_inputs()
(
(
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) ,
) : Optional[Any] = config_and_inputs
snake_case__ : str = {
'input_ids': input_ids,
'token_type_ids': token_type_ids,
'head_mask': head_mask,
}
return config, inputs_dict
@require_torch
class snake_case__ ( lowerCAmelCase_ , lowerCAmelCase_ , lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = (
(OpenAIGPTModel, OpenAIGPTLMHeadModel, OpenAIGPTDoubleHeadsModel, OpenAIGPTForSequenceClassification)
if is_torch_available()
else ()
)
_SCREAMING_SNAKE_CASE = (
(OpenAIGPTLMHeadModel,) if is_torch_available() else ()
) # TODO (PVP): Add Double HeadsModel when generate() function is changed accordingly
_SCREAMING_SNAKE_CASE = (
{
"""feature-extraction""": OpenAIGPTModel,
"""text-classification""": OpenAIGPTForSequenceClassification,
"""text-generation""": OpenAIGPTLMHeadModel,
"""zero-shot""": OpenAIGPTForSequenceClassification,
}
if is_torch_available()
else {}
)
def lowercase_ ( self : Optional[int], _snake_case : Union[str, Any], _snake_case : int, _snake_case : Tuple, _snake_case : Tuple, _snake_case : List[str] ) ->Optional[Any]:
if pipeline_test_casse_name == "ZeroShotClassificationPipelineTests":
# Get `tokenizer does not have a padding token` error for both fast/slow tokenizers.
# `OpenAIGPTConfig` was never used in pipeline tests, either because of a missing checkpoint or because a
# tiny config could not be created.
return True
return False
def lowercase_ ( self : Optional[Any], _snake_case : Union[str, Any], _snake_case : List[str], _snake_case : Any=False ) ->Tuple:
snake_case__ : Optional[int] = super()._prepare_for_class(_snake_case, _snake_case, return_labels=_snake_case )
if return_labels:
if model_class.__name__ == "OpenAIGPTDoubleHeadsModel":
snake_case__ : Union[str, Any] = torch.zeros(
(self.model_tester.batch_size, self.model_tester.num_choices, self.model_tester.seq_length), dtype=torch.long, device=_snake_case, )
snake_case__ : List[Any] = inputs_dict['labels']
snake_case__ : List[Any] = inputs_dict['labels']
snake_case__ : Any = torch.zeros(
(self.model_tester.batch_size, self.model_tester.num_choices), dtype=torch.long, device=_snake_case, )
snake_case__ : Tuple = torch.zeros(
self.model_tester.batch_size, dtype=torch.long, device=_snake_case )
return inputs_dict
def lowercase_ ( self : Union[str, Any] ) ->List[str]:
snake_case__ : List[str] = OpenAIGPTModelTester(self )
snake_case__ : Any = ConfigTester(self, config_class=_snake_case, n_embd=3_7 )
def lowercase_ ( self : Optional[int] ) ->str:
self.config_tester.run_common_tests()
def lowercase_ ( self : int ) ->Tuple:
snake_case__ : str = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_openai_gpt_model(*_snake_case )
def lowercase_ ( self : Tuple ) ->List[str]:
snake_case__ : Optional[int] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_lm_head_model(*_snake_case )
def lowercase_ ( self : Dict ) ->int:
snake_case__ : Optional[Any] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_double_lm_head_model(*_snake_case )
def lowercase_ ( self : int ) ->str:
snake_case__ : List[Any] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_openai_gpt_for_sequence_classification(*_snake_case )
@slow
def lowercase_ ( self : Optional[Any] ) ->str:
for model_name in OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
snake_case__ : Optional[int] = OpenAIGPTModel.from_pretrained(_snake_case )
self.assertIsNotNone(_snake_case )
@require_torch
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
@slow
def lowercase_ ( self : Tuple ) ->Optional[int]:
snake_case__ : Union[str, Any] = OpenAIGPTLMHeadModel.from_pretrained('openai-gpt' )
model.to(_snake_case )
snake_case__ : Tuple = torch.tensor([[4_8_1, 4_7_3_5, 5_4_4]], dtype=torch.long, device=_snake_case ) # the president is
snake_case__ : int = [
4_8_1,
4_7_3_5,
5_4_4,
2_4_6,
9_6_3,
8_7_0,
7_6_2,
2_3_9,
2_4_4,
4_0_4_7_7,
2_4_4,
2_4_9,
7_1_9,
8_8_1,
4_8_7,
5_4_4,
2_4_0,
2_4_4,
6_0_3,
4_8_1,
] # the president is a very good man. " \n " i\'m sure he is, " said the
snake_case__ : Optional[int] = model.generate(_snake_case, do_sample=_snake_case )
self.assertListEqual(output_ids[0].tolist(), _snake_case )
| 277 | 0 |
from __future__ import annotations
import unittest
from transformers import BlenderbotSmallConfig, BlenderbotSmallTokenizer, is_tf_available
from transformers.testing_utils import require_tf, require_tokenizers, slow
from transformers.utils import cached_property
from ...test_configuration_common import ConfigTester
from ...test_modeling_tf_common import TFModelTesterMixin, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_tf_available():
import tensorflow as tf
from transformers import TFAutoModelForSeqaSeqLM, TFBlenderbotSmallForConditionalGeneration, TFBlenderbotSmallModel
@require_tf
class __lowerCAmelCase :
_lowercase : str = BlenderbotSmallConfig
_lowercase : Optional[int] = {}
_lowercase : int = """gelu"""
def __init__( self , lowerCAmelCase__ , lowerCAmelCase__=1_3 , lowerCAmelCase__=7 , lowerCAmelCase__=True , lowerCAmelCase__=False , lowerCAmelCase__=9_9 , lowerCAmelCase__=3_2 , lowerCAmelCase__=2 , lowerCAmelCase__=4 , lowerCAmelCase__=3_7 , lowerCAmelCase__=0.1 , lowerCAmelCase__=0.1 , lowerCAmelCase__=2_0 , lowerCAmelCase__=2 , lowerCAmelCase__=1 , lowerCAmelCase__=0 , ) -> str:
'''simple docstring'''
a__ : List[str] =parent
a__ : Union[str, Any] =batch_size
a__ : Optional[Any] =seq_length
a__ : str =is_training
a__ : Any =use_labels
a__ : Optional[Any] =vocab_size
a__ : List[Any] =hidden_size
a__ : List[str] =num_hidden_layers
a__ : List[str] =num_attention_heads
a__ : Tuple =intermediate_size
a__ : Optional[int] =hidden_dropout_prob
a__ : List[Any] =attention_probs_dropout_prob
a__ : int =max_position_embeddings
a__ : Optional[Any] =eos_token_id
a__ : List[Any] =pad_token_id
a__ : Any =bos_token_id
def _lowercase ( self ) -> Optional[int]:
'''simple docstring'''
a__ : int =ids_tensor([self.batch_size, self.seq_length - 1] , self.vocab_size )
a__ : int =tf.expand_dims(tf.constant([self.eos_token_id] * self.batch_size ) , 1 )
a__ : List[str] =tf.concat([input_ids, eos_tensor] , axis=1 )
a__ : Any =ids_tensor([self.batch_size, self.seq_length] , self.vocab_size )
a__ : Dict =self.config_cls(
vocab_size=self.vocab_size , d_model=self.hidden_size , encoder_layers=self.num_hidden_layers , decoder_layers=self.num_hidden_layers , encoder_attention_heads=self.num_attention_heads , decoder_attention_heads=self.num_attention_heads , encoder_ffn_dim=self.intermediate_size , decoder_ffn_dim=self.intermediate_size , dropout=self.hidden_dropout_prob , attention_dropout=self.attention_probs_dropout_prob , max_position_embeddings=self.max_position_embeddings , eos_token_ids=[2] , bos_token_id=self.bos_token_id , pad_token_id=self.pad_token_id , decoder_start_token_id=self.pad_token_id , **self.config_updates , )
a__ : str =prepare_blenderbot_small_inputs_dict(_snake_case , _snake_case , _snake_case )
return config, inputs_dict
def _lowercase ( self , lowerCAmelCase__ , lowerCAmelCase__ ) -> Tuple:
'''simple docstring'''
a__ : str =TFBlenderbotSmallModel(config=_snake_case ).get_decoder()
a__ : List[str] =inputs_dict['input_ids']
a__ : Tuple =input_ids[:1, :]
a__ : str =inputs_dict['attention_mask'][:1, :]
a__ : Optional[Any] =inputs_dict['head_mask']
a__ : Union[str, Any] =1
# first forward pass
a__ : Dict =model(_snake_case , attention_mask=_snake_case , head_mask=_snake_case , use_cache=_snake_case )
a__ : List[Any] =outputs.to_tuple()
# create hypothetical next token and extent to next_input_ids
a__ : Optional[int] =ids_tensor((self.batch_size, 3) , config.vocab_size )
a__ : Any =tf.cast(ids_tensor((self.batch_size, 3) , 2 ) , tf.inta )
# append to next input_ids and
a__ : int =tf.concat([input_ids, next_tokens] , axis=-1 )
a__ : List[str] =tf.concat([attention_mask, next_attn_mask] , axis=-1 )
a__ : Tuple =model(_snake_case , attention_mask=_snake_case )[0]
a__ : int =model(_snake_case , attention_mask=_snake_case , past_key_values=_snake_case )[0]
self.parent.assertEqual(next_tokens.shape[1] , output_from_past.shape[1] )
# select random slice
a__ : List[str] =int(ids_tensor((1,) , output_from_past.shape[-1] ) )
a__ : Tuple =output_from_no_past[:, -3:, random_slice_idx]
a__ : List[Any] =output_from_past[:, :, random_slice_idx]
# test that outputs are equal for slice
tf.debugging.assert_near(_snake_case , _snake_case , rtol=1E-3 )
def _A ( SCREAMING_SNAKE_CASE : Dict , SCREAMING_SNAKE_CASE : Any , SCREAMING_SNAKE_CASE : Optional[int] , SCREAMING_SNAKE_CASE : Dict=None , SCREAMING_SNAKE_CASE : int=None , SCREAMING_SNAKE_CASE : List[str]=None , SCREAMING_SNAKE_CASE : Union[str, Any]=None , SCREAMING_SNAKE_CASE : Optional[Any]=None , ):
"""simple docstring"""
if attention_mask is None:
a__ : Union[str, Any] =tf.cast(tf.math.not_equal(SCREAMING_SNAKE_CASE , config.pad_token_id ) , tf.inta )
if decoder_attention_mask is None:
a__ : List[str] =tf.concat(
[
tf.ones(decoder_input_ids[:, :1].shape , dtype=tf.inta ),
tf.cast(tf.math.not_equal(decoder_input_ids[:, 1:] , config.pad_token_id ) , tf.inta ),
] , axis=-1 , )
if head_mask is None:
a__ : Tuple =tf.ones((config.encoder_layers, config.encoder_attention_heads) )
if decoder_head_mask is None:
a__ : int =tf.ones((config.decoder_layers, config.decoder_attention_heads) )
if cross_attn_head_mask is None:
a__ : Any =tf.ones((config.decoder_layers, config.decoder_attention_heads) )
return {
"input_ids": input_ids,
"decoder_input_ids": decoder_input_ids,
"attention_mask": attention_mask,
"decoder_attention_mask": decoder_attention_mask,
"head_mask": head_mask,
"decoder_head_mask": decoder_head_mask,
"cross_attn_head_mask": cross_attn_head_mask,
}
@require_tf
class __lowerCAmelCase ( lowerCAmelCase_ , lowerCAmelCase_ , unittest.TestCase):
_lowercase : List[str] = (
(TFBlenderbotSmallForConditionalGeneration, TFBlenderbotSmallModel) if is_tf_available() else ()
)
_lowercase : Optional[Any] = (TFBlenderbotSmallForConditionalGeneration,) if is_tf_available() else ()
_lowercase : Dict = (
{
"""conversational""": TFBlenderbotSmallForConditionalGeneration,
"""feature-extraction""": TFBlenderbotSmallModel,
"""summarization""": TFBlenderbotSmallForConditionalGeneration,
"""text2text-generation""": TFBlenderbotSmallForConditionalGeneration,
"""translation""": TFBlenderbotSmallForConditionalGeneration,
}
if is_tf_available()
else {}
)
_lowercase : Tuple = True
_lowercase : List[str] = False
_lowercase : List[Any] = False
def _lowercase ( self ) -> List[str]:
'''simple docstring'''
a__ : Optional[int] =TFBlenderbotSmallModelTester(self )
a__ : List[Any] =ConfigTester(self , config_class=_snake_case )
def _lowercase ( self ) -> Any:
'''simple docstring'''
self.config_tester.run_common_tests()
def _lowercase ( self ) -> Optional[Any]:
'''simple docstring'''
a__ : List[Any] =self.model_tester.prepare_config_and_inputs_for_common()
self.model_tester.check_decoder_model_past_large_inputs(*_snake_case )
@require_tokenizers
@require_tf
class __lowerCAmelCase ( unittest.TestCase):
_lowercase : Dict = [
"""Social anxiety\nWow, I am never shy. Do you have anxiety?\nYes. I end up sweating and blushing and feel like """
""" i'm going to throw up.\nand why is that?"""
]
_lowercase : int = """facebook/blenderbot_small-90M"""
@cached_property
def _lowercase ( self ) -> List[Any]:
'''simple docstring'''
return BlenderbotSmallTokenizer.from_pretrained("facebook/blenderbot-90M" )
@cached_property
def _lowercase ( self ) -> List[str]:
'''simple docstring'''
a__ : Tuple =TFAutoModelForSeqaSeqLM.from_pretrained(self.model_name )
return model
@slow
def _lowercase ( self ) -> Optional[int]:
'''simple docstring'''
a__ : Any =self.tokenizer(self.src_text , return_tensors="tf" )
a__ : List[str] =self.model.generate(
model_inputs.input_ids , attention_mask=model_inputs.attention_mask , num_beams=2 , use_cache=_snake_case , )
a__ : List[str] =self.tokenizer.batch_decode(generated_ids.numpy() , skip_special_tokens=_snake_case )[0]
assert generated_words in (
"i don't know. i just feel like i'm going to throw up. it's not fun.",
"i'm not sure. i just feel like i've been feeling like i have to be in a certain place",
"i'm not sure. i just feel like i've been in a bad situation.",
)
| 95 |
import os
import unittest
from transformers.models.transfo_xl.tokenization_transfo_xl import VOCAB_FILES_NAMES, TransfoXLTokenizer
from ...test_tokenization_common import TokenizerTesterMixin
class snake_case__ ( lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = TransfoXLTokenizer
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
def lowercase_ ( self : Optional[int] ) ->Any:
super().setUp()
snake_case__ : Tuple = [
'<unk>',
'[CLS]',
'[SEP]',
'want',
'unwanted',
'wa',
'un',
'running',
',',
'low',
'l',
]
snake_case__ : Any = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['vocab_file'] )
with open(self.vocab_file, 'w', encoding='utf-8' ) as vocab_writer:
vocab_writer.write(''.join([x + '\n' for x in vocab_tokens] ) )
def lowercase_ ( self : Union[str, Any], **_snake_case : List[Any] ) ->Dict:
snake_case__ : str = True
return TransfoXLTokenizer.from_pretrained(self.tmpdirname, **_snake_case )
def lowercase_ ( self : Optional[Any], _snake_case : str ) ->Dict:
snake_case__ : List[Any] = '<unk> UNwanted , running'
snake_case__ : List[Any] = '<unk> unwanted, running'
return input_text, output_text
def lowercase_ ( self : List[Any] ) ->Tuple:
snake_case__ : Dict = TransfoXLTokenizer(vocab_file=self.vocab_file, lower_case=_snake_case )
snake_case__ : str = tokenizer.tokenize('<unk> UNwanted , running' )
self.assertListEqual(_snake_case, ['<unk>', 'unwanted', ',', 'running'] )
self.assertListEqual(tokenizer.convert_tokens_to_ids(_snake_case ), [0, 4, 8, 7] )
def lowercase_ ( self : List[str] ) ->List[Any]:
snake_case__ : str = TransfoXLTokenizer(lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHeLLo ! how \n Are yoU ? ' ), ['hello', '!', 'how', 'are', 'you', '?'] )
def lowercase_ ( self : Optional[int] ) ->Optional[Any]:
snake_case__ : Optional[int] = TransfoXLTokenizer(lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHeLLo ! how \n Are yoU ? ' ), ['HeLLo', '!', 'how', 'Are', 'yoU', '?'] )
def lowercase_ ( self : Optional[int] ) ->Union[str, Any]:
snake_case__ : List[Any] = TransfoXLTokenizer(lower_case=_snake_case )
snake_case__ : Dict = 'Hello (bracket) and side-scrolled [and] Henry\'s $5,000 with 3.34 m. What\'s up!?'
snake_case__ : List[Any] = [
'Hello',
'(',
'bracket',
')',
'and',
'side',
'@-@',
'scrolled',
'[',
'and',
']',
'Henry',
'\'s',
'$',
'5',
'@,@',
'000',
'with',
'3',
'@.@',
'34',
'm',
'.',
'What',
'\'s',
'up',
'!',
'?',
]
self.assertListEqual(tokenizer.tokenize(_snake_case ), _snake_case )
self.assertEqual(tokenizer.convert_tokens_to_string(_snake_case ), _snake_case )
def lowercase_ ( self : Dict ) ->Any:
snake_case__ : Dict = self.get_tokenizer()
snake_case__ : Optional[Any] = len(_snake_case )
tokenizer.add_tokens(['new1', 'new2'] )
tokenizer.move_added_token('new1', 1 )
# Check that moved token is not copied (duplicate)
self.assertEqual(len(_snake_case ), original_len + 2 )
# Check that token is moved to specified id
self.assertEqual(tokenizer.encode('new1' ), [1] )
self.assertEqual(tokenizer.decode([1] ), 'new1' )
| 277 | 0 |
import gc
import unittest
import numpy as np
import torch
from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer
from diffusers import (
AutoencoderKL,
DDIMScheduler,
EulerAncestralDiscreteScheduler,
LMSDiscreteScheduler,
PNDMScheduler,
StableDiffusionPanoramaPipeline,
UNetaDConditionModel,
)
from diffusers.utils import slow, torch_device
from diffusers.utils.testing_utils import enable_full_determinism, require_torch_gpu, skip_mps
from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_IMAGE_PARAMS, TEXT_TO_IMAGE_PARAMS
from ..test_pipelines_common import PipelineLatentTesterMixin, PipelineTesterMixin
enable_full_determinism()
@skip_mps
class lowercase__( lowerCAmelCase_ , lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
a :int = StableDiffusionPanoramaPipeline
a :List[str] = TEXT_TO_IMAGE_PARAMS
a :Dict = TEXT_TO_IMAGE_BATCH_PARAMS
a :Any = TEXT_TO_IMAGE_IMAGE_PARAMS
a :List[Any] = TEXT_TO_IMAGE_IMAGE_PARAMS
def _lowercase ( self : Union[str, Any] ) -> Any:
torch.manual_seed(0 )
lowercase_ = UNetaDConditionModel(
block_out_channels=(3_2, 6_4) , layers_per_block=1 , sample_size=3_2 , in_channels=4 , out_channels=4 , down_block_types=('''DownBlock2D''', '''CrossAttnDownBlock2D''') , up_block_types=('''CrossAttnUpBlock2D''', '''UpBlock2D''') , cross_attention_dim=3_2 , )
lowercase_ = DDIMScheduler()
torch.manual_seed(0 )
lowercase_ = AutoencoderKL(
block_out_channels=[3_2, 6_4] , in_channels=3 , out_channels=3 , down_block_types=['''DownEncoderBlock2D''', '''DownEncoderBlock2D'''] , up_block_types=['''UpDecoderBlock2D''', '''UpDecoderBlock2D'''] , latent_channels=4 , )
torch.manual_seed(0 )
lowercase_ = CLIPTextConfig(
bos_token_id=0 , eos_token_id=2 , hidden_size=3_2 , intermediate_size=3_7 , layer_norm_eps=1e-05 , num_attention_heads=4 , num_hidden_layers=5 , pad_token_id=1 , vocab_size=1_0_0_0 , )
lowercase_ = CLIPTextModel(_snake_case )
lowercase_ = CLIPTokenizer.from_pretrained('''hf-internal-testing/tiny-random-clip''' )
lowercase_ = {
'unet': unet,
'scheduler': scheduler,
'vae': vae,
'text_encoder': text_encoder,
'tokenizer': tokenizer,
'safety_checker': None,
'feature_extractor': None,
}
return components
def _lowercase ( self : Optional[Any] , SCREAMING_SNAKE_CASE_ : List[Any] , SCREAMING_SNAKE_CASE_ : Tuple=0 ) -> Union[str, Any]:
lowercase_ = torch.manual_seed(_snake_case )
lowercase_ = {
'prompt': 'a photo of the dolomites',
'generator': generator,
# Setting height and width to None to prevent OOMs on CPU.
'height': None,
'width': None,
'num_inference_steps': 1,
'guidance_scale': 6.0,
'output_type': 'numpy',
}
return inputs
def _lowercase ( self : Optional[Any] ) -> str:
lowercase_ = 'cpu' # ensure determinism for the device-dependent torch.Generator
lowercase_ = self.get_dummy_components()
lowercase_ = StableDiffusionPanoramaPipeline(**_snake_case )
lowercase_ = sd_pipe.to(_snake_case )
sd_pipe.set_progress_bar_config(disable=_snake_case )
lowercase_ = self.get_dummy_inputs(_snake_case )
lowercase_ = sd_pipe(**_snake_case ).images
lowercase_ = image[0, -3:, -3:, -1]
assert image.shape == (1, 6_4, 6_4, 3)
lowercase_ = np.array([0.61_86, 0.53_74, 0.49_15, 0.41_35, 0.41_14, 0.45_63, 0.51_28, 0.49_77, 0.47_57] )
assert np.abs(image_slice.flatten() - expected_slice ).max() < 1e-2
def _lowercase ( self : Tuple ) -> Any:
super().test_inference_batch_consistent(batch_sizes=[1, 2] )
def _lowercase ( self : Optional[Any] ) -> Any:
super().test_inference_batch_single_identical(batch_size=2 , expected_max_diff=3.25e-3 )
def _lowercase ( self : str ) -> List[str]:
lowercase_ = 'cpu' # ensure determinism for the device-dependent torch.Generator
lowercase_ = self.get_dummy_components()
lowercase_ = StableDiffusionPanoramaPipeline(**_snake_case )
lowercase_ = sd_pipe.to(_snake_case )
sd_pipe.set_progress_bar_config(disable=_snake_case )
lowercase_ = self.get_dummy_inputs(_snake_case )
lowercase_ = 'french fries'
lowercase_ = sd_pipe(**_snake_case , negative_prompt=_snake_case )
lowercase_ = output.images
lowercase_ = image[0, -3:, -3:, -1]
assert image.shape == (1, 6_4, 6_4, 3)
lowercase_ = np.array([0.61_87, 0.53_75, 0.49_15, 0.41_36, 0.41_14, 0.45_63, 0.51_28, 0.49_76, 0.47_57] )
assert np.abs(image_slice.flatten() - expected_slice ).max() < 1e-2
def _lowercase ( self : Optional[Any] ) -> Optional[Any]:
lowercase_ = 'cpu' # ensure determinism for the device-dependent torch.Generator
lowercase_ = self.get_dummy_components()
lowercase_ = StableDiffusionPanoramaPipeline(**_snake_case )
lowercase_ = sd_pipe.to(_snake_case )
sd_pipe.set_progress_bar_config(disable=_snake_case )
lowercase_ = self.get_dummy_inputs(_snake_case )
lowercase_ = sd_pipe(**_snake_case , view_batch_size=2 )
lowercase_ = output.images
lowercase_ = image[0, -3:, -3:, -1]
assert image.shape == (1, 6_4, 6_4, 3)
lowercase_ = np.array([0.61_87, 0.53_75, 0.49_15, 0.41_36, 0.41_14, 0.45_63, 0.51_28, 0.49_76, 0.47_57] )
assert np.abs(image_slice.flatten() - expected_slice ).max() < 1e-2
def _lowercase ( self : Dict ) -> Any:
lowercase_ = 'cpu' # ensure determinism for the device-dependent torch.Generator
lowercase_ = self.get_dummy_components()
lowercase_ = EulerAncestralDiscreteScheduler(
beta_start=0.0_00_85 , beta_end=0.0_12 , beta_schedule='''scaled_linear''' )
lowercase_ = StableDiffusionPanoramaPipeline(**_snake_case )
lowercase_ = sd_pipe.to(_snake_case )
sd_pipe.set_progress_bar_config(disable=_snake_case )
lowercase_ = self.get_dummy_inputs(_snake_case )
lowercase_ = sd_pipe(**_snake_case ).images
lowercase_ = image[0, -3:, -3:, -1]
assert image.shape == (1, 6_4, 6_4, 3)
lowercase_ = np.array([0.40_24, 0.65_10, 0.49_01, 0.53_78, 0.58_13, 0.56_22, 0.47_95, 0.44_67, 0.49_52] )
assert np.abs(image_slice.flatten() - expected_slice ).max() < 1e-2
def _lowercase ( self : Union[str, Any] ) -> Any:
lowercase_ = 'cpu' # ensure determinism for the device-dependent torch.Generator
lowercase_ = self.get_dummy_components()
lowercase_ = PNDMScheduler(
beta_start=0.0_00_85 , beta_end=0.0_12 , beta_schedule='''scaled_linear''' , skip_prk_steps=_snake_case )
lowercase_ = StableDiffusionPanoramaPipeline(**_snake_case )
lowercase_ = sd_pipe.to(_snake_case )
sd_pipe.set_progress_bar_config(disable=_snake_case )
lowercase_ = self.get_dummy_inputs(_snake_case )
lowercase_ = sd_pipe(**_snake_case ).images
lowercase_ = image[0, -3:, -3:, -1]
assert image.shape == (1, 6_4, 6_4, 3)
lowercase_ = np.array([0.63_91, 0.62_91, 0.48_61, 0.51_34, 0.55_52, 0.45_78, 0.50_32, 0.50_23, 0.45_39] )
assert np.abs(image_slice.flatten() - expected_slice ).max() < 1e-2
@slow
@require_torch_gpu
class lowercase__( unittest.TestCase ):
"""simple docstring"""
def _lowercase ( self : Dict ) -> List[Any]:
super().tearDown()
gc.collect()
torch.cuda.empty_cache()
def _lowercase ( self : Optional[Any] , SCREAMING_SNAKE_CASE_ : Any=0 ) -> Any:
lowercase_ = torch.manual_seed(_snake_case )
lowercase_ = {
'prompt': 'a photo of the dolomites',
'generator': generator,
'num_inference_steps': 3,
'guidance_scale': 7.5,
'output_type': 'numpy',
}
return inputs
def _lowercase ( self : Any ) -> Optional[Any]:
lowercase_ = 'stabilityai/stable-diffusion-2-base'
lowercase_ = DDIMScheduler.from_pretrained(_snake_case , subfolder='''scheduler''' )
lowercase_ = StableDiffusionPanoramaPipeline.from_pretrained(_snake_case , scheduler=_snake_case , safety_checker=_snake_case )
pipe.to(_snake_case )
pipe.set_progress_bar_config(disable=_snake_case )
pipe.enable_attention_slicing()
lowercase_ = self.get_inputs()
lowercase_ = pipe(**_snake_case ).images
lowercase_ = image[0, -3:, -3:, -1].flatten()
assert image.shape == (1, 5_1_2, 2_0_4_8, 3)
lowercase_ = np.array(
[
0.36_96_83_92,
0.27_02_53_72,
0.32_44_67_66,
0.28_37_93_87,
0.36_36_32_74,
0.30_73_33_47,
0.27_10_00_27,
0.27_05_41_25,
0.25_53_60_96,
] )
assert np.abs(expected_slice - image_slice ).max() < 1e-2
def _lowercase ( self : Optional[Any] ) -> str:
lowercase_ = StableDiffusionPanoramaPipeline.from_pretrained(
'''stabilityai/stable-diffusion-2-base''' , safety_checker=_snake_case )
lowercase_ = LMSDiscreteScheduler.from_config(pipe.scheduler.config )
pipe.to(_snake_case )
pipe.set_progress_bar_config(disable=_snake_case )
pipe.enable_attention_slicing()
lowercase_ = self.get_inputs()
lowercase_ = pipe(**_snake_case ).images
lowercase_ = image[0, -3:, -3:, -1].flatten()
assert image.shape == (1, 5_1_2, 2_0_4_8, 3)
lowercase_ = np.array(
[
[
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
0.0,
]
] )
assert np.abs(expected_slice - image_slice ).max() < 1e-3
def _lowercase ( self : int ) -> Any:
lowercase_ = 0
def callback_fn(SCREAMING_SNAKE_CASE_ : int , SCREAMING_SNAKE_CASE_ : int , SCREAMING_SNAKE_CASE_ : torch.FloatTensor ) -> None:
lowercase_ = True
nonlocal number_of_steps
number_of_steps += 1
if step == 1:
lowercase_ = latents.detach().cpu().numpy()
assert latents.shape == (1, 4, 6_4, 2_5_6)
lowercase_ = latents[0, -3:, -3:, -1]
lowercase_ = np.array(
[
0.18_68_18_69,
0.33_90_78_16,
0.5_36_12_76,
0.14_43_28_65,
-0.02_85_66_11,
-0.73_94_11_23,
0.23_39_79_87,
0.47_32_26_82,
-0.37_82_31_64,
] )
assert np.abs(latents_slice.flatten() - expected_slice ).max() < 5e-2
elif step == 2:
lowercase_ = latents.detach().cpu().numpy()
assert latents.shape == (1, 4, 6_4, 2_5_6)
lowercase_ = latents[0, -3:, -3:, -1]
lowercase_ = np.array(
[
0.18_53_96_45,
0.33_98_72_48,
0.5_37_85_59,
0.14_43_71_42,
-0.02_45_52_61,
-0.7_33_83_17,
0.23_99_07_55,
0.47_35_62_72,
-0.3_78_65_05,
] )
assert np.abs(latents_slice.flatten() - expected_slice ).max() < 5e-2
lowercase_ = False
lowercase_ = 'stabilityai/stable-diffusion-2-base'
lowercase_ = DDIMScheduler.from_pretrained(_snake_case , subfolder='''scheduler''' )
lowercase_ = StableDiffusionPanoramaPipeline.from_pretrained(_snake_case , scheduler=_snake_case , safety_checker=_snake_case )
lowercase_ = pipe.to(_snake_case )
pipe.set_progress_bar_config(disable=_snake_case )
pipe.enable_attention_slicing()
lowercase_ = self.get_inputs()
pipe(**_snake_case , callback=_snake_case , callback_steps=1 )
assert callback_fn.has_been_called
assert number_of_steps == 3
def _lowercase ( self : Optional[Any] ) -> Optional[int]:
torch.cuda.empty_cache()
torch.cuda.reset_max_memory_allocated()
torch.cuda.reset_peak_memory_stats()
lowercase_ = 'stabilityai/stable-diffusion-2-base'
lowercase_ = DDIMScheduler.from_pretrained(_snake_case , subfolder='''scheduler''' )
lowercase_ = StableDiffusionPanoramaPipeline.from_pretrained(_snake_case , scheduler=_snake_case , safety_checker=_snake_case )
lowercase_ = pipe.to(_snake_case )
pipe.set_progress_bar_config(disable=_snake_case )
pipe.enable_attention_slicing(1 )
pipe.enable_sequential_cpu_offload()
lowercase_ = self.get_inputs()
lowercase_ = pipe(**_snake_case )
lowercase_ = torch.cuda.max_memory_allocated()
# make sure that less than 5.2 GB is allocated
assert mem_bytes < 5.5 * 1_0**9
| 30 |
from ...configuration_utils import PretrainedConfig
from ...utils import logging
a_ :Optional[int] = logging.get_logger(__name__)
a_ :Dict = {"openai-gpt": "https://huggingface.co/openai-gpt/resolve/main/config.json"}
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """openai-gpt"""
_SCREAMING_SNAKE_CASE = {
"""max_position_embeddings""": """n_positions""",
"""hidden_size""": """n_embd""",
"""num_attention_heads""": """n_head""",
"""num_hidden_layers""": """n_layer""",
}
def __init__( self : Optional[int], _snake_case : Dict=4_0_4_7_8, _snake_case : str=5_1_2, _snake_case : int=7_6_8, _snake_case : Tuple=1_2, _snake_case : Any=1_2, _snake_case : str="gelu", _snake_case : List[str]=0.1, _snake_case : Any=0.1, _snake_case : Dict=0.1, _snake_case : int=1e-5, _snake_case : Optional[Any]=0.0_2, _snake_case : List[Any]="cls_index", _snake_case : Any=True, _snake_case : Any=None, _snake_case : int=True, _snake_case : Optional[Any]=0.1, **_snake_case : List[Any], ) ->Optional[int]:
snake_case__ : int = vocab_size
snake_case__ : Dict = n_positions
snake_case__ : str = n_embd
snake_case__ : str = n_layer
snake_case__ : List[Any] = n_head
snake_case__ : List[Any] = afn
snake_case__ : Optional[Any] = resid_pdrop
snake_case__ : List[str] = embd_pdrop
snake_case__ : List[Any] = attn_pdrop
snake_case__ : Optional[int] = layer_norm_epsilon
snake_case__ : str = initializer_range
snake_case__ : List[str] = summary_type
snake_case__ : Optional[int] = summary_use_proj
snake_case__ : List[str] = summary_activation
snake_case__ : Optional[Any] = summary_first_dropout
snake_case__ : int = summary_proj_to_labels
super().__init__(**_snake_case )
| 277 | 0 |
"""simple docstring"""
from collections import OrderedDict
from typing import List, Mapping
from packaging import version
from ...configuration_utils import PretrainedConfig
from ...onnx import OnnxConfig
from ...utils import logging
lowercase_ = logging.get_logger(__name__)
lowercase_ = {
"google/efficientnet-b7": "https://huggingface.co/google/efficientnet-b7/resolve/main/config.json",
}
class __lowerCAmelCase ( lowerCAmelCase_ ):
'''simple docstring'''
__UpperCAmelCase : List[str] = 'efficientnet'
def __init__( self , _a = 3 , _a = 600 , _a = 2.0 , _a = 3.1 , _a = 8 , _a = [3, 3, 5, 3, 5, 5, 3] , _a = [32, 16, 24, 40, 80, 112, 192] , _a = [16, 24, 40, 80, 112, 192, 320] , _a = [] , _a = [1, 2, 2, 2, 1, 2, 1] , _a = [1, 2, 2, 3, 3, 4, 1] , _a = [1, 6, 6, 6, 6, 6, 6] , _a = 0.25 , _a = "swish" , _a = 2_560 , _a = "mean" , _a = 0.02 , _a = 0.001 , _a = 0.99 , _a = 0.5 , _a = 0.2 , **_a , ):
super().__init__(**_snake_case )
__a = num_channels
__a = image_size
__a = width_coefficient
__a = depth_coefficient
__a = depth_divisor
__a = kernel_sizes
__a = in_channels
__a = out_channels
__a = depthwise_padding
__a = strides
__a = num_block_repeats
__a = expand_ratios
__a = squeeze_expansion_ratio
__a = hidden_act
__a = hidden_dim
__a = pooling_type
__a = initializer_range
__a = batch_norm_eps
__a = batch_norm_momentum
__a = dropout_rate
__a = drop_connect_rate
__a = sum(_snake_case ) * 4
class __lowerCAmelCase ( lowerCAmelCase_ ):
'''simple docstring'''
__UpperCAmelCase : Union[str, Any] = version.parse('1.11' )
@property
def __UpperCAmelCase ( self ):
return OrderedDict(
[
('''pixel_values''', {0: '''batch''', 1: '''num_channels''', 2: '''height''', 3: '''width'''}),
] )
@property
def __UpperCAmelCase ( self ):
return 1E-5
| 45 |
import argparse
import logging
import os
from datetime import datetime
import numpy as np
import torch
from torch import nn
from torch.utils.data import DataLoader, RandomSampler, TensorDataset
from tqdm import tqdm
from transformers import GPTaLMHeadModel
a_ :Optional[Any] = logging.getLogger(__name__)
def lowercase_ (A : List[Any] , A : List[Any] ):
# save results
if os.path.exists(A ):
if os.path.exists(os.path.join(A , 'config.json' ) ) and os.path.isfile(
os.path.join(A , 'config.json' ) ):
os.remove(os.path.join(A , 'config.json' ) )
if os.path.exists(os.path.join(A , 'pytorch_model.bin' ) ) and os.path.isfile(
os.path.join(A , 'pytorch_model.bin' ) ):
os.remove(os.path.join(A , 'pytorch_model.bin' ) )
else:
os.makedirs(A )
model.save_pretrained(A )
def lowercase_ (A : Any , A : Optional[Any]=False ):
snake_case__ : str = 2
if unlogit:
snake_case__ : Dict = torch.pow(A , A )
snake_case__ : Any = p * torch.log(A )
snake_case__ : Tuple = 0
return -plogp.sum(dim=-1 )
def lowercase_ (A : List[str] ):
logger.info('lv, h >\t' + '\t'.join(F'''{x + 1}''' for x in range(len(A ) ) ) )
for row in range(len(A ) ):
if tensor.dtype != torch.long:
logger.info(F'''layer {row + 1}:\t''' + '\t'.join(F'''{x:.5f}''' for x in tensor[row].cpu().data ) )
else:
logger.info(F'''layer {row + 1}:\t''' + '\t'.join(F'''{x:d}''' for x in tensor[row].cpu().data ) )
def lowercase_ (A : Tuple , A : Optional[Any] , A : str , A : int=True , A : Optional[int]=True , A : Any=None , A : int=False ):
snake_case__ , snake_case__ : Optional[Any] = model.config.num_hidden_layers, model.config.num_attention_heads
snake_case__ : int = torch.zeros(A , A ).to(args.device )
snake_case__ : Any = torch.zeros(A , A ).to(args.device )
if head_mask is None:
snake_case__ : Dict = torch.ones(A , A ).to(args.device )
head_mask.requires_grad_(requires_grad=A )
# If actually pruned attention multi-head, set head mask to None to avoid shape mismatch
if actually_pruned:
snake_case__ : Optional[int] = None
snake_case__ : List[Any] = 0.0
snake_case__ : str = 0.0
for step, inputs in enumerate(tqdm(A , desc='Iteration' , disable=args.local_rank not in [-1, 0] ) ):
snake_case__ : Union[str, Any] = tuple(t.to(args.device ) for t in inputs )
((snake_case__) , ) : Optional[Any] = inputs
# Do a forward pass (not with torch.no_grad() since we need gradients for importance score - see below)
snake_case__ : Union[str, Any] = model(A , labels=A , head_mask=A )
# (loss), lm_logits, presents, (all hidden_states), (attentions)
snake_case__ , snake_case__ , snake_case__ : Dict = (
outputs[0],
outputs[1],
outputs[-1],
) # Loss and logits are the first, attention the last
loss.backward() # Backpropagate to populate the gradients in the head mask
total_loss += loss.detach().cpu().numpy()
if compute_entropy:
for layer, attn in enumerate(A ):
snake_case__ : Optional[Any] = entropy(attn.detach() , A )
attn_entropy[layer] += masked_entropy.sum(-1 ).sum(0 ).sum(0 ).detach()
if compute_importance:
head_importance += head_mask.grad.abs().detach()
tot_tokens += torch.ones_like(A ).float().detach().sum().data
# Normalize
attn_entropy /= tot_tokens
head_importance /= tot_tokens
# Layerwise importance normalization
if not args.dont_normalize_importance_by_layer:
snake_case__ : Union[str, Any] = 2
snake_case__ : List[Any] = torch.pow(torch.pow(A , A ).sum(-1 ) , 1 / exponent )
head_importance /= norm_by_layer.unsqueeze(-1 ) + 1e-20
if not args.dont_normalize_global_importance:
snake_case__ : Tuple = (head_importance - head_importance.min()) / (head_importance.max() - head_importance.min())
# Print matrices
if compute_entropy:
logger.info('Attention entropies' )
print_ad_tensor(A )
if compute_importance:
logger.info('Head importance scores' )
print_ad_tensor(A )
logger.info('Head ranked by importance scores' )
snake_case__ : Tuple = torch.zeros(head_importance.numel() , dtype=torch.long , device=args.device )
snake_case__ : Union[str, Any] = torch.arange(
head_importance.numel() , device=args.device )
snake_case__ : str = head_ranks.view_as(A )
print_ad_tensor(A )
return attn_entropy, head_importance, total_loss
def lowercase_ (A : Optional[int] , A : Dict , A : Optional[int] ):
snake_case__ , snake_case__ , snake_case__ : Any = compute_heads_importance(A , A , A , compute_entropy=A )
snake_case__ : Tuple = 1 / loss # instead of downsteam score use the LM loss
logger.info('Pruning: original score: %f, threshold: %f' , A , original_score * args.masking_threshold )
snake_case__ : Optional[Any] = torch.ones_like(A )
snake_case__ : Union[str, Any] = max(1 , int(new_head_mask.numel() * args.masking_amount ) )
snake_case__ : Dict = original_score
while current_score >= original_score * args.masking_threshold:
snake_case__ : int = new_head_mask.clone().detach() # save current head mask
# heads from least important to most - keep only not-masked heads
snake_case__ : List[Any] = float('Inf' )
snake_case__ : Union[str, Any] = head_importance.view(-1 ).sort()[1]
if len(A ) <= num_to_mask:
print('BREAK BY num_to_mask' )
break
# mask heads
snake_case__ : int = current_heads_to_mask[:num_to_mask]
logger.info('Heads to mask: %s' , str(current_heads_to_mask.tolist() ) )
snake_case__ : int = new_head_mask.view(-1 )
snake_case__ : int = 0.0
snake_case__ : Union[str, Any] = new_head_mask.view_as(A )
snake_case__ : List[str] = new_head_mask.clone().detach()
print_ad_tensor(A )
# Compute metric and head importance again
snake_case__ , snake_case__ , snake_case__ : Any = compute_heads_importance(
A , A , A , compute_entropy=A , head_mask=A )
snake_case__ : Dict = 1 / loss
logger.info(
'Masking: current score: %f, remaining heads %d (%.1f percents)' , A , new_head_mask.sum() , new_head_mask.sum() / new_head_mask.numel() * 1_0_0 , )
logger.info('Final head mask' )
print_ad_tensor(A )
np.save(os.path.join(args.output_dir , 'head_mask.npy' ) , head_mask.detach().cpu().numpy() )
return head_mask
def lowercase_ (A : List[str] , A : Tuple , A : Optional[Any] , A : int ):
snake_case__ : Any = datetime.now()
snake_case__ , snake_case__ , snake_case__ : str = compute_heads_importance(
A , A , A , compute_entropy=A , compute_importance=A , head_mask=A )
snake_case__ : Tuple = 1 / loss
snake_case__ : Dict = datetime.now() - before_time
snake_case__ : Union[str, Any] = sum(p.numel() for p in model.parameters() )
snake_case__ : Optional[Any] = {
layer: (1 - head_mask[layer].long()).nonzero().squeeze().tolist() for layer in range(len(A ) )
}
for k, v in heads_to_prune.items():
if isinstance(A , A ):
snake_case__ : Any = [
v,
]
assert sum(len(A ) for h in heads_to_prune.values() ) == (1 - head_mask.long()).sum().item()
model.prune_heads(A )
snake_case__ : Dict = sum(p.numel() for p in model.parameters() )
snake_case__ : Tuple = datetime.now()
snake_case__ , snake_case__ , snake_case__ : Dict = compute_heads_importance(
A , A , A , compute_entropy=A , compute_importance=A , head_mask=A , actually_pruned=A , )
snake_case__ : Any = 1 / loss
snake_case__ : int = datetime.now() - before_time
logger.info(
'Pruning: original num of params: %.2e, after pruning %.2e (%.1f percents)' , A , A , pruned_num_params / original_num_params * 1_0_0 , )
logger.info('Pruning: score with masking: %f score with pruning: %f' , A , A )
logger.info('Pruning: speed ratio (original timing / new timing): %f percents' , original_time / new_time * 1_0_0 )
save_model(A , args.output_dir )
def lowercase_ ():
snake_case__ : str = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
'--data_dir' , default=A , type=A , required=A , help='The input data dir. Should contain the .tsv files (or other data files) for the task.' , )
parser.add_argument(
'--model_name_or_path' , default=A , type=A , required=A , help='Path to pretrained model or model identifier from huggingface.co/models' , )
parser.add_argument(
'--output_dir' , default=A , type=A , required=A , help='The output directory where the model predictions and checkpoints will be written.' , )
# Other parameters
parser.add_argument(
'--config_name' , default='' , type=A , help='Pretrained config name or path if not the same as model_name_or_path' , )
parser.add_argument(
'--tokenizer_name' , default='' , type=A , help='Pretrained tokenizer name or path if not the same as model_name_or_path' , )
parser.add_argument(
'--cache_dir' , default=A , type=A , help='Where do you want to store the pre-trained models downloaded from s3' , )
parser.add_argument(
'--data_subset' , type=A , default=-1 , help='If > 0: limit the data to a subset of data_subset instances.' )
parser.add_argument(
'--overwrite_output_dir' , action='store_true' , help='Whether to overwrite data in output directory' )
parser.add_argument(
'--overwrite_cache' , action='store_true' , help='Overwrite the cached training and evaluation sets' )
parser.add_argument(
'--dont_normalize_importance_by_layer' , action='store_true' , help='Don\'t normalize importance score by layers' )
parser.add_argument(
'--dont_normalize_global_importance' , action='store_true' , help='Don\'t normalize all importance scores between 0 and 1' , )
parser.add_argument(
'--try_masking' , action='store_true' , help='Whether to try to mask head until a threshold of accuracy.' )
parser.add_argument(
'--masking_threshold' , default=0.9 , type=A , help='masking threshold in term of metrics (stop masking when metric < threshold * original metric value).' , )
parser.add_argument(
'--masking_amount' , default=0.1 , type=A , help='Amount to heads to masking at each masking step.' )
parser.add_argument('--metric_name' , default='acc' , type=A , help='Metric to use for head masking.' )
parser.add_argument(
'--max_seq_length' , default=1_2_8 , type=A , help=(
'The maximum total input sequence length after WordPiece tokenization. \n'
'Sequences longer than this will be truncated, sequences shorter padded.'
) , )
parser.add_argument('--batch_size' , default=1 , type=A , help='Batch size.' )
parser.add_argument('--seed' , type=A , default=4_2 )
parser.add_argument('--local_rank' , type=A , default=-1 , help='local_rank for distributed training on gpus' )
parser.add_argument('--no_cuda' , action='store_true' , help='Whether not to use CUDA when available' )
parser.add_argument('--server_ip' , type=A , default='' , help='Can be used for distant debugging.' )
parser.add_argument('--server_port' , type=A , default='' , help='Can be used for distant debugging.' )
snake_case__ : Optional[int] = parser.parse_args()
if args.server_ip and args.server_port:
# Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script
import ptvsd
print('Waiting for debugger attach' )
ptvsd.enable_attach(address=(args.server_ip, args.server_port) , redirect_output=A )
ptvsd.wait_for_attach()
# Setup devices and distributed training
if args.local_rank == -1 or args.no_cuda:
snake_case__ : List[Any] = torch.device('cuda' if torch.cuda.is_available() and not args.no_cuda else 'cpu' )
snake_case__ : Optional[Any] = 0 if args.no_cuda else torch.cuda.device_count()
else:
torch.cuda.set_device(args.local_rank )
snake_case__ : int = torch.device('cuda' , args.local_rank )
snake_case__ : List[str] = 1
torch.distributed.init_process_group(backend='nccl' ) # Initializes the distributed backend
# Setup logging
logging.basicConfig(level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN )
logger.info('device: {} n_gpu: {}, distributed: {}'.format(args.device , args.n_gpu , bool(args.local_rank != -1 ) ) )
snake_case__ : Any = GPTaLMHeadModel.from_pretrained(args.model_name_or_path )
# Distributed and parallel training
model.to(args.device )
if args.local_rank != -1:
snake_case__ : List[str] = nn.parallel.DistributedDataParallel(
A , device_ids=[args.local_rank] , output_device=args.local_rank , find_unused_parameters=A )
elif args.n_gpu > 1:
snake_case__ : Optional[int] = nn.DataParallel(A )
# Print/save training arguments
os.makedirs(args.output_dir , exist_ok=A )
torch.save(A , os.path.join(args.output_dir , 'run_args.bin' ) )
logger.info('Training/evaluation parameters %s' , A )
# Prepare dataset
snake_case__ : Optional[Any] = np.concatenate(
[
np.loadtxt(args.data_dir , dtype=np.intaa ),
] )
snake_case__ : List[str] = (torch.from_numpy(A ),)
snake_case__ : int = TensorDataset(*A )
snake_case__ : Union[str, Any] = RandomSampler(A )
snake_case__ : Any = DataLoader(A , sampler=A , batch_size=args.batch_size )
# Compute head entropy and importance score
compute_heads_importance(A , A , A )
# Try head masking (set heads to zero until the score goes under a threshole)
# and head pruning (remove masked heads and see the effect on the network)
if args.try_masking and args.masking_threshold > 0.0 and args.masking_threshold < 1.0:
snake_case__ : Dict = mask_heads(A , A , A )
prune_heads(A , A , A , A )
if __name__ == "__main__":
main()
| 277 | 0 |
'''simple docstring'''
def __lowerCAmelCase ( snake_case__ ):
__UpperCamelCase : int = len(snake_case__ )
for i in range(snake_case__ ):
for j in range(i + 1 , snake_case__ ):
if numbers[j] < numbers[i]:
__UpperCamelCase : Dict = numbers[j], numbers[i]
return numbers
if __name__ == "__main__":
_lowerCAmelCase = input('''Enter numbers separated by a comma:\n''').strip()
_lowerCAmelCase = [int(item) for item in user_input.split(''',''')]
print(exchange_sort(unsorted))
| 298 |
import argparse
import json
from collections import OrderedDict
from pathlib import Path
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import (
SegformerConfig,
SegformerForImageClassification,
SegformerForSemanticSegmentation,
SegformerImageProcessor,
)
from transformers.utils import logging
logging.set_verbosity_info()
a_ :Dict = logging.get_logger(__name__)
def lowercase_ (A : Optional[Any] , A : Any=False ):
snake_case__ : List[Any] = OrderedDict()
for key, value in state_dict.items():
if encoder_only and not key.startswith('head' ):
snake_case__ : str = 'segformer.encoder.' + key
if key.startswith('backbone' ):
snake_case__ : str = key.replace('backbone' , 'segformer.encoder' )
if "patch_embed" in key:
# replace for example patch_embed1 by patch_embeddings.0
snake_case__ : Optional[int] = key[key.find('patch_embed' ) + len('patch_embed' )]
snake_case__ : int = key.replace(F'''patch_embed{idx}''' , F'''patch_embeddings.{int(A )-1}''' )
if "norm" in key:
snake_case__ : Optional[int] = key.replace('norm' , 'layer_norm' )
if "segformer.encoder.layer_norm" in key:
# replace for example layer_norm1 by layer_norm.0
snake_case__ : Tuple = key[key.find('segformer.encoder.layer_norm' ) + len('segformer.encoder.layer_norm' )]
snake_case__ : Union[str, Any] = key.replace(F'''layer_norm{idx}''' , F'''layer_norm.{int(A )-1}''' )
if "layer_norm1" in key:
snake_case__ : List[Any] = key.replace('layer_norm1' , 'layer_norm_1' )
if "layer_norm2" in key:
snake_case__ : List[Any] = key.replace('layer_norm2' , 'layer_norm_2' )
if "block" in key:
# replace for example block1 by block.0
snake_case__ : List[Any] = key[key.find('block' ) + len('block' )]
snake_case__ : List[Any] = key.replace(F'''block{idx}''' , F'''block.{int(A )-1}''' )
if "attn.q" in key:
snake_case__ : int = key.replace('attn.q' , 'attention.self.query' )
if "attn.proj" in key:
snake_case__ : str = key.replace('attn.proj' , 'attention.output.dense' )
if "attn" in key:
snake_case__ : Optional[int] = key.replace('attn' , 'attention.self' )
if "fc1" in key:
snake_case__ : str = key.replace('fc1' , 'dense1' )
if "fc2" in key:
snake_case__ : Dict = key.replace('fc2' , 'dense2' )
if "linear_pred" in key:
snake_case__ : Union[str, Any] = key.replace('linear_pred' , 'classifier' )
if "linear_fuse" in key:
snake_case__ : List[str] = key.replace('linear_fuse.conv' , 'linear_fuse' )
snake_case__ : List[Any] = key.replace('linear_fuse.bn' , 'batch_norm' )
if "linear_c" in key:
# replace for example linear_c4 by linear_c.3
snake_case__ : Optional[int] = key[key.find('linear_c' ) + len('linear_c' )]
snake_case__ : Tuple = key.replace(F'''linear_c{idx}''' , F'''linear_c.{int(A )-1}''' )
if key.startswith('head' ):
snake_case__ : Tuple = key.replace('head' , 'classifier' )
snake_case__ : Optional[int] = value
return new_state_dict
def lowercase_ (A : Tuple , A : Optional[int] ):
# for each of the encoder blocks:
for i in range(config.num_encoder_blocks ):
for j in range(config.depths[i] ):
# read in weights + bias of keys and values (which is a single matrix in the original implementation)
snake_case__ : List[str] = state_dict.pop(F'''segformer.encoder.block.{i}.{j}.attention.self.kv.weight''' )
snake_case__ : Optional[Any] = state_dict.pop(F'''segformer.encoder.block.{i}.{j}.attention.self.kv.bias''' )
# next, add keys and values (in that order) to the state dict
snake_case__ : str = kv_weight[
: config.hidden_sizes[i], :
]
snake_case__ : Dict = kv_bias[: config.hidden_sizes[i]]
snake_case__ : List[str] = kv_weight[
config.hidden_sizes[i] :, :
]
snake_case__ : List[Any] = kv_bias[
config.hidden_sizes[i] :
]
def lowercase_ ():
snake_case__ : Union[str, Any] = 'http://images.cocodataset.org/val2017/000000039769.jpg'
snake_case__ : Dict = Image.open(requests.get(A , stream=A ).raw )
return image
@torch.no_grad()
def lowercase_ (A : Any , A : Union[str, Any] , A : Optional[Any] ):
snake_case__ : List[str] = SegformerConfig()
snake_case__ : Dict = False
# set attributes based on model_name
snake_case__ : Optional[int] = 'huggingface/label-files'
if "segformer" in model_name:
snake_case__ : str = model_name[len('segformer.' ) : len('segformer.' ) + 2]
if "ade" in model_name:
snake_case__ : Optional[int] = 1_5_0
snake_case__ : int = 'ade20k-id2label.json'
snake_case__ : List[Any] = (1, 1_5_0, 1_2_8, 1_2_8)
elif "city" in model_name:
snake_case__ : str = 1_9
snake_case__ : List[str] = 'cityscapes-id2label.json'
snake_case__ : Optional[Any] = (1, 1_9, 1_2_8, 1_2_8)
else:
raise ValueError(F'''Model {model_name} not supported''' )
elif "mit" in model_name:
snake_case__ : str = True
snake_case__ : Union[str, Any] = model_name[4:6]
snake_case__ : Optional[Any] = 1_0_0_0
snake_case__ : Optional[int] = 'imagenet-1k-id2label.json'
snake_case__ : List[Any] = (1, 1_0_0_0)
else:
raise ValueError(F'''Model {model_name} not supported''' )
# set config attributes
snake_case__ : str = json.load(open(hf_hub_download(A , A , repo_type='dataset' ) , 'r' ) )
snake_case__ : List[Any] = {int(A ): v for k, v in idalabel.items()}
snake_case__ : Union[str, Any] = idalabel
snake_case__ : Tuple = {v: k for k, v in idalabel.items()}
if size == "b0":
pass
elif size == "b1":
snake_case__ : List[Any] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : Tuple = 2_5_6
elif size == "b2":
snake_case__ : List[str] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : int = 7_6_8
snake_case__ : List[Any] = [3, 4, 6, 3]
elif size == "b3":
snake_case__ : Optional[Any] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : int = 7_6_8
snake_case__ : Optional[Any] = [3, 4, 1_8, 3]
elif size == "b4":
snake_case__ : str = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : Optional[Any] = 7_6_8
snake_case__ : Union[str, Any] = [3, 8, 2_7, 3]
elif size == "b5":
snake_case__ : List[str] = [6_4, 1_2_8, 3_2_0, 5_1_2]
snake_case__ : Optional[Any] = 7_6_8
snake_case__ : Any = [3, 6, 4_0, 3]
else:
raise ValueError(F'''Size {size} not supported''' )
# load image processor (only resize + normalize)
snake_case__ : Dict = SegformerImageProcessor(
image_scale=(5_1_2, 5_1_2) , keep_ratio=A , align=A , do_random_crop=A )
# prepare image
snake_case__ : List[str] = prepare_img()
snake_case__ : Dict = image_processor(images=A , return_tensors='pt' ).pixel_values
logger.info(F'''Converting model {model_name}...''' )
# load original state dict
if encoder_only:
snake_case__ : Tuple = torch.load(A , map_location=torch.device('cpu' ) )
else:
snake_case__ : int = torch.load(A , map_location=torch.device('cpu' ) )['state_dict']
# rename keys
snake_case__ : List[Any] = rename_keys(A , encoder_only=A )
if not encoder_only:
del state_dict["decode_head.conv_seg.weight"]
del state_dict["decode_head.conv_seg.bias"]
# key and value matrices need special treatment
read_in_k_v(A , A )
# create HuggingFace model and load state dict
if encoder_only:
snake_case__ : str = False
snake_case__ : List[Any] = SegformerForImageClassification(A )
else:
snake_case__ : Dict = SegformerForSemanticSegmentation(A )
model.load_state_dict(A )
model.eval()
# forward pass
snake_case__ : int = model(A )
snake_case__ : Any = outputs.logits
# set expected_slice based on model name
# ADE20k checkpoints
if model_name == "segformer.b0.512x512.ade.160k":
snake_case__ : Dict = torch.tensor(
[
[[-4.6310, -5.5232, -6.2356], [-5.1921, -6.1444, -6.5996], [-5.4424, -6.2790, -6.7574]],
[[-12.1391, -13.3122, -13.9554], [-12.8732, -13.9352, -14.3563], [-12.9438, -13.8226, -14.2513]],
[[-12.5134, -13.4686, -14.4915], [-12.8669, -14.4343, -14.7758], [-13.2523, -14.5819, -15.0694]],
] )
elif model_name == "segformer.b1.512x512.ade.160k":
snake_case__ : Optional[int] = torch.tensor(
[
[[-7.5820, -8.7231, -8.3215], [-8.0600, -10.3529, -10.0304], [-7.5208, -9.4103, -9.6239]],
[[-12.6918, -13.8994, -13.7137], [-13.3196, -15.7523, -15.4789], [-12.9343, -14.8757, -14.9689]],
[[-11.1911, -11.9421, -11.3243], [-11.3342, -13.6839, -13.3581], [-10.3909, -12.1832, -12.4858]],
] )
elif model_name == "segformer.b2.512x512.ade.160k":
snake_case__ : List[Any] = torch.tensor(
[
[[-11.8173, -14.3850, -16.3128], [-14.5648, -16.5804, -18.6568], [-14.7223, -15.7387, -18.4218]],
[[-15.7290, -17.9171, -19.4423], [-18.3105, -19.9448, -21.4661], [-17.9296, -18.6497, -20.7910]],
[[-15.0783, -17.0336, -18.2789], [-16.8771, -18.6870, -20.1612], [-16.2454, -17.1426, -19.5055]],
] )
elif model_name == "segformer.b3.512x512.ade.160k":
snake_case__ : Union[str, Any] = torch.tensor(
[
[[-9.0878, -10.2081, -10.1891], [-9.3144, -10.7941, -10.9843], [-9.2294, -10.3855, -10.5704]],
[[-12.2316, -13.9068, -13.6102], [-12.9161, -14.3702, -14.3235], [-12.5233, -13.7174, -13.7932]],
[[-14.6275, -15.2490, -14.9727], [-14.3400, -15.9687, -16.2827], [-14.1484, -15.4033, -15.8937]],
] )
elif model_name == "segformer.b4.512x512.ade.160k":
snake_case__ : Dict = torch.tensor(
[
[[-12.3144, -13.2447, -14.0802], [-13.3614, -14.5816, -15.6117], [-13.3340, -14.4433, -16.2219]],
[[-19.2781, -20.4128, -20.7506], [-20.6153, -21.6566, -22.0998], [-19.9800, -21.0430, -22.1494]],
[[-18.8739, -19.7804, -21.1834], [-20.1233, -21.6765, -23.2944], [-20.0315, -21.2641, -23.6944]],
] )
elif model_name == "segformer.b5.640x640.ade.160k":
snake_case__ : List[Any] = torch.tensor(
[
[[-9.5524, -12.0835, -11.7348], [-10.5229, -13.6446, -14.5662], [-9.5842, -12.8851, -13.9414]],
[[-15.3432, -17.5323, -17.0818], [-16.3330, -18.9255, -19.2101], [-15.1340, -17.7848, -18.3971]],
[[-12.6072, -14.9486, -14.6631], [-13.7629, -17.0907, -17.7745], [-12.7899, -16.1695, -17.1671]],
] )
# Cityscapes checkpoints
elif model_name == "segformer.b0.1024x1024.city.160k":
snake_case__ : str = torch.tensor(
[
[[-11.9295, -13.4057, -14.8106], [-13.3431, -14.8179, -15.3781], [-14.2836, -15.5942, -16.1588]],
[[-11.4906, -12.8067, -13.6564], [-13.1189, -14.0500, -14.1543], [-13.8748, -14.5136, -14.8789]],
[[0.5374, 0.1067, -0.4742], [0.1141, -0.2255, -0.7099], [-0.3000, -0.5924, -1.3105]],
] )
elif model_name == "segformer.b0.512x1024.city.160k":
snake_case__ : Tuple = torch.tensor(
[
[[-7.8217, -9.8767, -10.1717], [-9.4438, -10.9058, -11.4047], [-9.7939, -12.3495, -12.1079]],
[[-7.1514, -9.5336, -10.0860], [-9.7776, -11.6822, -11.8439], [-10.1411, -12.7655, -12.8972]],
[[0.3021, 0.0805, -0.2310], [-0.0328, -0.1605, -0.2714], [-0.1408, -0.5477, -0.6976]],
] )
elif model_name == "segformer.b0.640x1280.city.160k":
snake_case__ : Any = torch.tensor(
[
[
[-1.1_372e01, -1.2_787e01, -1.3_477e01],
[-1.2_536e01, -1.4_194e01, -1.4_409e01],
[-1.3_217e01, -1.4_888e01, -1.5_327e01],
],
[
[-1.4_791e01, -1.7_122e01, -1.8_277e01],
[-1.7_163e01, -1.9_192e01, -1.9_533e01],
[-1.7_897e01, -1.9_991e01, -2.0_315e01],
],
[
[7.6_723e-01, 4.1_921e-01, -7.7_878e-02],
[4.7_772e-01, 9.5_557e-03, -2.8_082e-01],
[3.6_032e-01, -2.4_826e-01, -5.1_168e-01],
],
] )
elif model_name == "segformer.b0.768x768.city.160k":
snake_case__ : Optional[int] = torch.tensor(
[
[[-9.4959, -11.3087, -11.7479], [-11.0025, -12.6540, -12.3319], [-11.4064, -13.0487, -12.9905]],
[[-9.8905, -11.3084, -12.0854], [-11.1726, -12.7698, -12.9583], [-11.5985, -13.3278, -14.1774]],
[[0.2213, 0.0192, -0.2466], [-0.1731, -0.4213, -0.4874], [-0.3126, -0.6541, -1.1389]],
] )
elif model_name == "segformer.b1.1024x1024.city.160k":
snake_case__ : Union[str, Any] = torch.tensor(
[
[[-13.5748, -13.9111, -12.6500], [-14.3500, -15.3683, -14.2328], [-14.7532, -16.0424, -15.6087]],
[[-17.1651, -15.8725, -12.9653], [-17.2580, -17.3718, -14.8223], [-16.6058, -16.8783, -16.7452]],
[[-3.6456, -3.0209, -1.4203], [-3.0797, -3.1959, -2.0000], [-1.8757, -1.9217, -1.6997]],
] )
elif model_name == "segformer.b2.1024x1024.city.160k":
snake_case__ : List[str] = torch.tensor(
[
[[-16.0976, -16.4856, -17.3962], [-16.6234, -19.0342, -19.7685], [-16.0900, -18.0661, -19.1180]],
[[-18.4750, -18.8488, -19.5074], [-19.4030, -22.1570, -22.5977], [-19.1191, -20.8486, -22.3783]],
[[-4.5178, -5.5037, -6.5109], [-5.0884, -7.2174, -8.0334], [-4.4156, -5.8117, -7.2970]],
] )
elif model_name == "segformer.b3.1024x1024.city.160k":
snake_case__ : List[Any] = torch.tensor(
[
[[-14.2081, -14.4732, -14.1977], [-14.5867, -16.4423, -16.6356], [-13.4441, -14.9685, -16.8696]],
[[-14.4576, -14.7073, -15.0451], [-15.0816, -17.6237, -17.9873], [-14.4213, -16.0199, -18.5992]],
[[-4.7349, -4.9588, -5.0966], [-4.3210, -6.9325, -7.2591], [-3.4312, -4.7484, -7.1917]],
] )
elif model_name == "segformer.b4.1024x1024.city.160k":
snake_case__ : str = torch.tensor(
[
[[-11.7737, -11.9526, -11.3273], [-13.6692, -14.4574, -13.8878], [-13.8937, -14.6924, -15.9345]],
[[-14.6706, -14.5330, -14.1306], [-16.1502, -16.8180, -16.4269], [-16.8338, -17.8939, -20.1746]],
[[1.0491, 0.8289, 1.0310], [1.1044, 0.5219, 0.8055], [1.0899, 0.6926, 0.5590]],
] )
elif model_name == "segformer.b5.1024x1024.city.160k":
snake_case__ : List[str] = torch.tensor(
[
[[-12.5641, -13.4777, -13.0684], [-13.9587, -15.8983, -16.6557], [-13.3109, -15.7350, -16.3141]],
[[-14.7074, -15.4352, -14.5944], [-16.6353, -18.1663, -18.6120], [-15.1702, -18.0329, -18.1547]],
[[-1.7990, -2.0951, -1.7784], [-2.6397, -3.8245, -3.9686], [-1.5264, -2.8126, -2.9316]],
] )
else:
snake_case__ : Tuple = logits.argmax(-1 ).item()
print('Predicted class:' , model.config.idalabel[predicted_class_idx] )
# verify logits
if not encoder_only:
assert logits.shape == expected_shape
assert torch.allclose(logits[0, :3, :3, :3] , A , atol=1e-2 )
# finally, save model and image processor
logger.info(F'''Saving PyTorch model and image processor to {pytorch_dump_folder_path}...''' )
Path(A ).mkdir(exist_ok=A )
model.save_pretrained(A )
image_processor.save_pretrained(A )
if __name__ == "__main__":
a_ :Optional[int] = argparse.ArgumentParser()
parser.add_argument(
"--model_name",
default="segformer.b0.512x512.ade.160k",
type=str,
help="Name of the model you'd like to convert.",
)
parser.add_argument(
"--checkpoint_path", default=None, type=str, help="Path to the original PyTorch checkpoint (.pth file)."
)
parser.add_argument(
"--pytorch_dump_folder_path", default=None, type=str, help="Path to the folder to output PyTorch model."
)
a_ :Union[str, Any] = parser.parse_args()
convert_segformer_checkpoint(args.model_name, args.checkpoint_path, args.pytorch_dump_folder_path)
| 277 | 0 |
import logging
import os
import sys
import warnings
from dataclasses import dataclass, field
from random import randint
from typing import Optional
import datasets
import evaluate
import numpy as np
from datasets import DatasetDict, load_dataset
import transformers
from transformers import (
AutoConfig,
AutoFeatureExtractor,
AutoModelForAudioClassification,
HfArgumentParser,
Trainer,
TrainingArguments,
set_seed,
)
from transformers.trainer_utils import get_last_checkpoint
from transformers.utils import check_min_version, send_example_telemetry
from transformers.utils.versions import require_version
_snake_case = logging.getLogger(__name__)
# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
check_min_version("4.31.0")
require_version("datasets>=1.14.0", "To fix: pip install -r examples/pytorch/audio-classification/requirements.txt")
def lowerCAmelCase_ ( snake_case_,snake_case_,snake_case_ = 16000 ):
_A : Optional[int] = int(round(sample_rate * max_length ) )
if len(snake_case_ ) <= sample_length:
return wav
_A : Any = randint(0,len(snake_case_ ) - sample_length - 1 )
return wav[random_offset : random_offset + sample_length]
@dataclass
class lowercase :
_a = field(default=lowerCAmelCase_,metadata={"help": "Name of a dataset from the datasets package"} )
_a = field(
default=lowerCAmelCase_,metadata={"help": "The configuration name of the dataset to use (via the datasets library)."} )
_a = field(
default=lowerCAmelCase_,metadata={"help": "A file containing the training audio paths and labels."} )
_a = field(
default=lowerCAmelCase_,metadata={"help": "A file containing the validation audio paths and labels."} )
_a = field(
default="train",metadata={
"help": "The name of the training data set split to use (via the datasets library). Defaults to 'train'"
},)
_a = field(
default="validation",metadata={
"help": (
"The name of the training data set split to use (via the datasets library). Defaults to 'validation'"
)
},)
_a = field(
default="audio",metadata={"help": "The name of the dataset column containing the audio data. Defaults to 'audio'"},)
_a = field(
default="label",metadata={"help": "The name of the dataset column containing the labels. Defaults to 'label'"} )
_a = field(
default=lowerCAmelCase_,metadata={
"help": (
"For debugging purposes or quicker training, truncate the number of training examples to this "
"value if set."
)
},)
_a = field(
default=lowerCAmelCase_,metadata={
"help": (
"For debugging purposes or quicker training, truncate the number of evaluation examples to this "
"value if set."
)
},)
_a = field(
default=2_0,metadata={"help": "Audio clips will be randomly cut to this length during training if the value is set."},)
@dataclass
class lowercase :
_a = field(
default="facebook/wav2vec2-base",metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"},)
_a = field(
default=lowerCAmelCase_,metadata={"help": "Pretrained config name or path if not the same as model_name"} )
_a = field(
default=lowerCAmelCase_,metadata={"help": "Where do you want to store the pretrained models downloaded from the Hub"} )
_a = field(
default="main",metadata={"help": "The specific model version to use (can be a branch name, tag name or commit id)."},)
_a = field(
default=lowerCAmelCase_,metadata={"help": "Name or path of preprocessor config."} )
_a = field(
default=lowerCAmelCase_,metadata={"help": "Whether to freeze the feature encoder layers of the model."} )
_a = field(
default=lowerCAmelCase_,metadata={"help": "Whether to generate an attention mask in the feature extractor."} )
_a = field(
default=lowerCAmelCase_,metadata={
"help": (
"Will use the token generated when running `huggingface-cli login` (necessary to use this script "
"with private models)."
)
},)
_a = field(
default=lowerCAmelCase_,metadata={"help": "Whether to freeze the feature extractor layers of the model."} )
_a = field(
default=lowerCAmelCase_,metadata={"help": "Will enable to load a pretrained model whose head dimensions are different."},)
def a__ ( self ) -> str:
if not self.freeze_feature_extractor and self.freeze_feature_encoder:
warnings.warn(
"""The argument `--freeze_feature_extractor` is deprecated and """
"""will be removed in a future version. Use `--freeze_feature_encoder`"""
"""instead. Setting `freeze_feature_encoder==True`.""" , _snake_case , )
if self.freeze_feature_extractor and not self.freeze_feature_encoder:
raise ValueError(
"""The argument `--freeze_feature_extractor` is deprecated and """
"""should not be used in combination with `--freeze_feature_encoder`."""
"""Only make use of `--freeze_feature_encoder`.""" )
def lowerCAmelCase_ ( ):
# See all possible arguments in src/transformers/training_args.py
# or by passing the --help flag to this script.
# We now keep distinct sets of args, for a cleaner separation of concerns.
_A : Optional[int] = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments) )
if len(sys.argv ) == 2 and sys.argv[1].endswith(""".json""" ):
# If we pass only one argument to the script and it's the path to a json file,
# let's parse it to get our arguments.
_A : Dict = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1] ) )
else:
_A : str = parser.parse_args_into_dataclasses()
# Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
# information sent is the one passed as arguments along with your Python/PyTorch versions.
send_example_telemetry("""run_audio_classification""",snake_case_,snake_case_ )
# Setup logging
logging.basicConfig(
format="""%(asctime)s - %(levelname)s - %(name)s - %(message)s""",datefmt="""%m/%d/%Y %H:%M:%S""",handlers=[logging.StreamHandler(sys.stdout )],)
if training_args.should_log:
# The default of training_args.log_level is passive, so we set log level at info here to have that default.
transformers.utils.logging.set_verbosity_info()
_A : Optional[Any] = training_args.get_process_log_level()
logger.setLevel(snake_case_ )
transformers.utils.logging.set_verbosity(snake_case_ )
transformers.utils.logging.enable_default_handler()
transformers.utils.logging.enable_explicit_format()
# Log on each process the small summary:
logger.warning(
f'''Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu} '''
+ f'''distributed training: {bool(training_args.local_rank != -1 )}, 16-bits training: {training_args.fpaa}''' )
logger.info(f'''Training/evaluation parameters {training_args}''' )
# Set seed before initializing model.
set_seed(training_args.seed )
# Detecting last checkpoint.
_A : Dict = None
if os.path.isdir(training_args.output_dir ) and training_args.do_train and not training_args.overwrite_output_dir:
_A : Optional[int] = get_last_checkpoint(training_args.output_dir )
if last_checkpoint is None and len(os.listdir(training_args.output_dir ) ) > 0:
raise ValueError(
f'''Output directory ({training_args.output_dir}) already exists and is not empty. '''
"""Use --overwrite_output_dir to train from scratch.""" )
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
logger.info(
f'''Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change '''
"""the `--output_dir` or add `--overwrite_output_dir` to train from scratch.""" )
# Initialize our dataset and prepare it for the audio classification task.
_A : Dict = DatasetDict()
_A : Tuple = load_dataset(
data_args.dataset_name,data_args.dataset_config_name,split=data_args.train_split_name,use_auth_token=True if model_args.use_auth_token else None,)
_A : Optional[int] = load_dataset(
data_args.dataset_name,data_args.dataset_config_name,split=data_args.eval_split_name,use_auth_token=True if model_args.use_auth_token else None,)
if data_args.audio_column_name not in raw_datasets["train"].column_names:
raise ValueError(
f'''--audio_column_name {data_args.audio_column_name} not found in dataset \'{data_args.dataset_name}\'. '''
"""Make sure to set `--audio_column_name` to the correct audio column - one of """
f'''{", ".join(raw_datasets["train"].column_names )}.''' )
if data_args.label_column_name not in raw_datasets["train"].column_names:
raise ValueError(
f'''--label_column_name {data_args.label_column_name} not found in dataset \'{data_args.dataset_name}\'. '''
"""Make sure to set `--label_column_name` to the correct text column - one of """
f'''{", ".join(raw_datasets["train"].column_names )}.''' )
# Setting `return_attention_mask=True` is the way to get a correctly masked mean-pooling over
# transformer outputs in the classifier, but it doesn't always lead to better accuracy
_A : Optional[Any] = AutoFeatureExtractor.from_pretrained(
model_args.feature_extractor_name or model_args.model_name_or_path,return_attention_mask=model_args.attention_mask,cache_dir=model_args.cache_dir,revision=model_args.model_revision,use_auth_token=True if model_args.use_auth_token else None,)
# `datasets` takes care of automatically loading and resampling the audio,
# so we just need to set the correct target sampling rate.
_A : List[str] = raw_datasets.cast_column(
data_args.audio_column_name,datasets.features.Audio(sampling_rate=feature_extractor.sampling_rate ) )
_A : Union[str, Any] = feature_extractor.model_input_names[0]
def train_transforms(snake_case_ ):
_A : Optional[Any] = []
for audio in batch[data_args.audio_column_name]:
_A : List[Any] = random_subsample(
audio["""array"""],max_length=data_args.max_length_seconds,sample_rate=feature_extractor.sampling_rate )
subsampled_wavs.append(snake_case_ )
_A : Optional[int] = feature_extractor(snake_case_,sampling_rate=feature_extractor.sampling_rate )
_A : Optional[int] = {model_input_name: inputs.get(snake_case_ )}
_A : Any = list(batch[data_args.label_column_name] )
return output_batch
def val_transforms(snake_case_ ):
_A : Optional[Any] = [audio['array'] for audio in batch[data_args.audio_column_name]]
_A : Union[str, Any] = feature_extractor(snake_case_,sampling_rate=feature_extractor.sampling_rate )
_A : List[str] = {model_input_name: inputs.get(snake_case_ )}
_A : List[str] = list(batch[data_args.label_column_name] )
return output_batch
# Prepare label mappings.
# We'll include these in the model's config to get human readable labels in the Inference API.
_A : Dict = raw_datasets['train'].features[data_args.label_column_name].names
_A : Tuple = {}, {}
for i, label in enumerate(snake_case_ ):
_A : Tuple = str(snake_case_ )
_A : List[str] = label
# Load the accuracy metric from the datasets package
_A : Union[str, Any] = evaluate.load("""accuracy""" )
# Define our compute_metrics function. It takes an `EvalPrediction` object (a namedtuple with
# `predictions` and `label_ids` fields) and has to return a dictionary string to float.
def compute_metrics(snake_case_ ):
_A : Tuple = np.argmax(eval_pred.predictions,axis=1 )
return metric.compute(predictions=snake_case_,references=eval_pred.label_ids )
_A : List[str] = AutoConfig.from_pretrained(
model_args.config_name or model_args.model_name_or_path,num_labels=len(snake_case_ ),labelaid=snake_case_,idalabel=snake_case_,finetuning_task="""audio-classification""",cache_dir=model_args.cache_dir,revision=model_args.model_revision,use_auth_token=True if model_args.use_auth_token else None,)
_A : Any = AutoModelForAudioClassification.from_pretrained(
model_args.model_name_or_path,from_tf=bool(""".ckpt""" in model_args.model_name_or_path ),config=snake_case_,cache_dir=model_args.cache_dir,revision=model_args.model_revision,use_auth_token=True if model_args.use_auth_token else None,ignore_mismatched_sizes=model_args.ignore_mismatched_sizes,)
# freeze the convolutional waveform encoder
if model_args.freeze_feature_encoder:
model.freeze_feature_encoder()
if training_args.do_train:
if data_args.max_train_samples is not None:
_A : Dict = (
raw_datasets['train'].shuffle(seed=training_args.seed ).select(range(data_args.max_train_samples ) )
)
# Set the training transforms
raw_datasets["train"].set_transform(snake_case_,output_all_columns=snake_case_ )
if training_args.do_eval:
if data_args.max_eval_samples is not None:
_A : List[Any] = (
raw_datasets['eval'].shuffle(seed=training_args.seed ).select(range(data_args.max_eval_samples ) )
)
# Set the validation transforms
raw_datasets["eval"].set_transform(snake_case_,output_all_columns=snake_case_ )
# Initialize our trainer
_A : Dict = Trainer(
model=snake_case_,args=snake_case_,train_dataset=raw_datasets["""train"""] if training_args.do_train else None,eval_dataset=raw_datasets["""eval"""] if training_args.do_eval else None,compute_metrics=snake_case_,tokenizer=snake_case_,)
# Training
if training_args.do_train:
_A : Tuple = None
if training_args.resume_from_checkpoint is not None:
_A : Any = training_args.resume_from_checkpoint
elif last_checkpoint is not None:
_A : Optional[Any] = last_checkpoint
_A : int = trainer.train(resume_from_checkpoint=snake_case_ )
trainer.save_model()
trainer.log_metrics("""train""",train_result.metrics )
trainer.save_metrics("""train""",train_result.metrics )
trainer.save_state()
# Evaluation
if training_args.do_eval:
_A : int = trainer.evaluate()
trainer.log_metrics("""eval""",snake_case_ )
trainer.save_metrics("""eval""",snake_case_ )
# Write model card and (optionally) push to hub
_A : Any = {
'finetuned_from': model_args.model_name_or_path,
'tasks': 'audio-classification',
'dataset': data_args.dataset_name,
'tags': ['audio-classification'],
}
if training_args.push_to_hub:
trainer.push_to_hub(**snake_case_ )
else:
trainer.create_model_card(**snake_case_ )
if __name__ == "__main__":
main()
| 26 |
import argparse
import json
import os
import fairseq
import torch
from fairseq.data import Dictionary
from transformers import (
WavaVecaConfig,
WavaVecaCTCTokenizer,
WavaVecaFeatureExtractor,
WavaVecaForCTC,
WavaVecaForPreTraining,
WavaVecaProcessor,
logging,
)
from transformers.models.wavaveca.modeling_wavaveca import WavaVecaForSequenceClassification
logging.set_verbosity_info()
a_ :List[Any] = logging.get_logger(__name__)
a_ :List[Any] = {
"post_extract_proj": "feature_projection.projection",
"encoder.pos_conv.0": "encoder.pos_conv_embed.conv",
"self_attn.k_proj": "encoder.layers.*.attention.k_proj",
"self_attn.v_proj": "encoder.layers.*.attention.v_proj",
"self_attn.q_proj": "encoder.layers.*.attention.q_proj",
"self_attn.out_proj": "encoder.layers.*.attention.out_proj",
"self_attn_layer_norm": "encoder.layers.*.layer_norm",
"fc1": "encoder.layers.*.feed_forward.intermediate_dense",
"fc2": "encoder.layers.*.feed_forward.output_dense",
"final_layer_norm": "encoder.layers.*.final_layer_norm",
"encoder.layer_norm": "encoder.layer_norm",
"adapter_layer": "encoder.layers.*.adapter_layer",
"w2v_model.layer_norm": "feature_projection.layer_norm",
"quantizer.weight_proj": "quantizer.weight_proj",
"quantizer.vars": "quantizer.codevectors",
"project_q": "project_q",
"final_proj": "project_hid",
"w2v_encoder.proj": "lm_head",
"mask_emb": "masked_spec_embed",
"pooling_layer.linear": "projector",
"pooling_layer.projection": "classifier",
}
a_ :List[Any] = [
"lm_head",
"quantizer.weight_proj",
"quantizer.codevectors",
"project_q",
"project_hid",
"projector",
"classifier",
]
def lowercase_ (A : Dict ):
snake_case__ : Optional[Any] = {}
with open(A , 'r' ) as file:
for line_number, line in enumerate(A ):
snake_case__ : Dict = line.strip()
if line:
snake_case__ : int = line.split()
snake_case__ : List[str] = line_number
snake_case__ : Dict = words[0]
snake_case__ : Optional[Any] = value
return result
def lowercase_ (A : int , A : int , A : Optional[int] , A : Optional[Any] , A : Tuple ):
for attribute in key.split('.' ):
snake_case__ : Optional[int] = getattr(A , A )
snake_case__ : Union[str, Any] = None
for param_key in PARAM_MAPPING.keys():
if full_name.endswith(A ):
snake_case__ : List[str] = PARAM_MAPPING[full_name.split('.' )[-1]]
snake_case__ : Dict = 'param'
if weight_type is not None and weight_type != "param":
snake_case__ : Union[str, Any] = getattr(A , A ).shape
elif weight_type is not None and weight_type == "param":
snake_case__ : Optional[int] = hf_pointer
for attribute in hf_param_name.split('.' ):
snake_case__ : Optional[Any] = getattr(A , A )
snake_case__ : Dict = shape_pointer.shape
# let's reduce dimension
snake_case__ : List[Any] = value[0]
else:
snake_case__ : Union[str, Any] = hf_pointer.shape
if hf_shape != value.shape:
raise ValueError(
F'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
F''' {value.shape} for {full_name}''' )
if weight_type == "weight":
snake_case__ : Any = value
elif weight_type == "weight_g":
snake_case__ : List[Any] = value
elif weight_type == "weight_v":
snake_case__ : Any = value
elif weight_type == "bias":
snake_case__ : List[Any] = value
elif weight_type == "param":
for attribute in hf_param_name.split('.' ):
snake_case__ : int = getattr(A , A )
snake_case__ : Optional[int] = value
else:
snake_case__ : Optional[Any] = value
logger.info(F'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def lowercase_ (A : Tuple , A : List[Any] , A : int , A : str , A : Tuple ):
snake_case__ : Optional[int] = None
for param_key in PARAM_MAPPING.keys():
if full_name.endswith(A ):
snake_case__ : List[str] = PARAM_MAPPING[full_name.split('.' )[-1]]
snake_case__ : str = 'param'
if weight_type is not None and weight_type != "param":
snake_case__ : int = '.'.join([key, weight_type] )
elif weight_type is not None and weight_type == "param":
snake_case__ : Any = '.'.join([key, hf_param_name] )
else:
snake_case__ : Dict = key
snake_case__ : List[str] = value if 'lm_head' in full_key else value[0]
a_ :List[str] = {
"W_a": "linear_1.weight",
"W_b": "linear_2.weight",
"b_a": "linear_1.bias",
"b_b": "linear_2.bias",
"ln_W": "norm.weight",
"ln_b": "norm.bias",
}
def lowercase_ (A : str , A : Optional[Any] , A : Optional[Any]=None , A : List[str]=None ):
snake_case__ : Optional[int] = False
for key, mapped_key in MAPPING.items():
snake_case__ : Tuple = 'wav2vec2.' + mapped_key if mapped_key not in TOP_LEVEL_KEYS else mapped_key
if key in name or key.split('w2v_model.' )[-1] == name.split('.' )[0]:
snake_case__ : Optional[int] = True
if "*" in mapped_key:
snake_case__ : List[Any] = name.split(A )[0].split('.' )[-2]
snake_case__ : Union[str, Any] = mapped_key.replace('*' , A )
if "weight_g" in name:
snake_case__ : Tuple = 'weight_g'
elif "weight_v" in name:
snake_case__ : List[str] = 'weight_v'
elif "bias" in name:
snake_case__ : Dict = 'bias'
elif "weight" in name:
# TODO: don't match quantizer.weight_proj
snake_case__ : Optional[int] = 'weight'
else:
snake_case__ : str = None
if hf_dict is not None:
rename_dict(A , A , A , A , A )
else:
set_recursively(A , A , A , A , A )
return is_used
return is_used
def lowercase_ (A : Optional[Any] , A : Dict , A : Optional[int] ):
snake_case__ : Dict = []
snake_case__ : Tuple = fairseq_model.state_dict()
snake_case__ : str = hf_model.wavaveca.feature_extractor
for name, value in fairseq_dict.items():
snake_case__ : str = False
if "conv_layers" in name:
load_conv_layer(
A , A , A , A , hf_model.config.feat_extract_norm == 'group' , )
snake_case__ : Any = True
else:
snake_case__ : Dict = load_wavaveca_layer(A , A , A )
if not is_used:
unused_weights.append(A )
logger.warning(F'''Unused weights: {unused_weights}''' )
def lowercase_ (A : Dict , A : Optional[Any] , A : Tuple , A : str , A : List[str] ):
snake_case__ : List[Any] = full_name.split('conv_layers.' )[-1]
snake_case__ : List[str] = name.split('.' )
snake_case__ : List[Any] = int(items[0] )
snake_case__ : str = int(items[1] )
if type_id == 0:
if "bias" in name:
if value.shape != feature_extractor.conv_layers[layer_id].conv.bias.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.''' )
snake_case__ : Any = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
if value.shape != feature_extractor.conv_layers[layer_id].conv.weight.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.''' )
snake_case__ : str = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
if value.shape != feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape} was found.''' )
snake_case__ : str = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
if value.shape != feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape:
raise ValueError(
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape} was found.''' )
snake_case__ : int = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(A )
@torch.no_grad()
def lowercase_ (A : Union[str, Any] , A : str , A : Tuple=None , A : List[str]=None , A : Any=True , A : Optional[int]=False ):
if config_path is not None:
snake_case__ : List[Any] = WavaVecaConfig.from_pretrained(A )
else:
snake_case__ : List[Any] = WavaVecaConfig()
if is_seq_class:
snake_case__ : Dict = read_txt_into_dict(A )
snake_case__ : Any = idalabel
snake_case__ : Union[str, Any] = WavaVecaForSequenceClassification(A )
snake_case__ : Any = WavaVecaFeatureExtractor(
feature_size=1 , sampling_rate=1_6_0_0_0 , padding_value=0 , do_normalize=A , return_attention_mask=A , )
feature_extractor.save_pretrained(A )
elif is_finetuned:
if dict_path:
snake_case__ : str = Dictionary.load(A )
# important change bos & pad token id since CTC symbol is <pad> and
# not <s> as in fairseq
snake_case__ : List[str] = target_dict.pad_index
snake_case__ : Optional[int] = target_dict.bos_index
snake_case__ : Optional[int] = target_dict.eos_index
snake_case__ : List[Any] = len(target_dict.symbols )
snake_case__ : str = os.path.join(A , 'vocab.json' )
if not os.path.isdir(A ):
logger.error('--pytorch_dump_folder_path ({}) should be a directory'.format(A ) )
return
os.makedirs(A , exist_ok=A )
snake_case__ : Optional[Any] = target_dict.indices
# fairseq has the <pad> and <s> switched
snake_case__ : Optional[Any] = 0
snake_case__ : Union[str, Any] = 1
with open(A , 'w' , encoding='utf-8' ) as vocab_handle:
json.dump(A , A )
snake_case__ : List[Any] = WavaVecaCTCTokenizer(
A , unk_token=target_dict.unk_word , pad_token=target_dict.pad_word , bos_token=target_dict.bos_word , eos_token=target_dict.eos_word , word_delimiter_token='|' , do_lower_case=A , )
snake_case__ : str = True if config.feat_extract_norm == 'layer' else False
snake_case__ : Optional[Any] = WavaVecaFeatureExtractor(
feature_size=1 , sampling_rate=1_6_0_0_0 , padding_value=0 , do_normalize=A , return_attention_mask=A , )
snake_case__ : Union[str, Any] = WavaVecaProcessor(feature_extractor=A , tokenizer=A )
processor.save_pretrained(A )
snake_case__ : str = WavaVecaForCTC(A )
else:
snake_case__ : int = WavaVecaForPreTraining(A )
if is_finetuned or is_seq_class:
snake_case__ , snake_case__ , snake_case__ : str = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] , arg_overrides={'data': '/'.join(dict_path.split('/' )[:-1] )} )
else:
snake_case__ : Tuple = argparse.Namespace(task='audio_pretraining' )
snake_case__ : str = fairseq.tasks.setup_task(A )
snake_case__ , snake_case__ , snake_case__ : Any = fairseq.checkpoint_utils.load_model_ensemble_and_task([checkpoint_path] , task=A )
snake_case__ : List[Any] = model[0].eval()
recursively_load_weights(A , A , not is_finetuned )
hf_wavavec.save_pretrained(A )
if __name__ == "__main__":
a_ :List[Any] = argparse.ArgumentParser()
parser.add_argument("--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model.")
parser.add_argument("--checkpoint_path", default=None, type=str, help="Path to fairseq checkpoint")
parser.add_argument("--dict_path", default=None, type=str, help="Path to dict of fine-tuned model")
parser.add_argument("--config_path", default=None, type=str, help="Path to hf config.json of model to convert")
parser.add_argument(
"--not_finetuned", action="store_true", help="Whether the model to convert is a fine-tuned model or not"
)
parser.add_argument(
"--is_seq_class",
action="store_true",
help="Whether the model to convert is a fine-tuned sequence classification model or not",
)
a_ :str = parser.parse_args()
a_ :Tuple = not args.not_finetuned and not args.is_seq_class
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.config_path,
args.dict_path,
is_finetuned,
args.is_seq_class,
)
| 277 | 0 |
from math import factorial, radians
def A ( _UpperCAmelCase : float , _UpperCAmelCase : int = 18 , _UpperCAmelCase : int = 10 ) -> str:
'''simple docstring'''
_UpperCAmelCase = angle_in_degrees - ((angle_in_degrees // 360.0) * 360.0)
# Converting from degrees to radians
_UpperCAmelCase = radians(_UpperCAmelCase )
_UpperCAmelCase = angle_in_radians
_UpperCAmelCase = 3
_UpperCAmelCase = -1
for _ in range(_UpperCAmelCase ):
result += (b * (angle_in_radians**a)) / factorial(_UpperCAmelCase )
_UpperCAmelCase = -b # One positive term and the next will be negative and so on...
a += 2 # Increased by 2 for every term.
return round(_UpperCAmelCase , _UpperCAmelCase )
if __name__ == "__main__":
__import__("doctest").testmod()
| 339 |
from typing import Dict, List
from nltk.translate import gleu_score
import datasets
from datasets import MetricInfo
a_ :Any = "\\n@misc{wu2016googles,\n title={Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation},\n author={Yonghui Wu and Mike Schuster and Zhifeng Chen and Quoc V. Le and Mohammad Norouzi and Wolfgang Macherey\n and Maxim Krikun and Yuan Cao and Qin Gao and Klaus Macherey and Jeff Klingner and Apurva Shah and Melvin\n Johnson and Xiaobing Liu and Łukasz Kaiser and Stephan Gouws and Yoshikiyo Kato and Taku Kudo and Hideto\n Kazawa and Keith Stevens and George Kurian and Nishant Patil and Wei Wang and Cliff Young and\n Jason Smith and Jason Riesa and Alex Rudnick and Oriol Vinyals and Greg Corrado and Macduff Hughes\n and Jeffrey Dean},\n year={2016},\n eprint={1609.08144},\n archivePrefix={arXiv},\n primaryClass={cs.CL}\n}\n"
a_ :List[str] = "\\nThe BLEU score has some undesirable properties when used for single\nsentences, as it was designed to be a corpus measure. We therefore\nuse a slightly different score for our RL experiments which we call\nthe 'GLEU score'. For the GLEU score, we record all sub-sequences of\n1, 2, 3 or 4 tokens in output and target sequence (n-grams). We then\ncompute a recall, which is the ratio of the number of matching n-grams\nto the number of total n-grams in the target (ground truth) sequence,\nand a precision, which is the ratio of the number of matching n-grams\nto the number of total n-grams in the generated output sequence. Then\nGLEU score is simply the minimum of recall and precision. This GLEU\nscore's range is always between 0 (no matches) and 1 (all match) and\nit is symmetrical when switching output and target. According to\nour experiments, GLEU score correlates quite well with the BLEU\nmetric on a corpus level but does not have its drawbacks for our per\nsentence reward objective.\n"
a_ :List[str] = "\\nComputes corpus-level Google BLEU (GLEU) score of translated segments against one or more references.\nInstead of averaging the sentence level GLEU scores (i.e. macro-average precision), Wu et al. (2016) sum up the matching\ntokens and the max of hypothesis and reference tokens for each sentence, then compute using the aggregate values.\n\nArgs:\n predictions (list of str): list of translations to score.\n Each translation should be tokenized into a list of tokens.\n references (list of list of str): list of lists of references for each translation.\n Each reference should be tokenized into a list of tokens.\n min_len (int): The minimum order of n-gram this function should extract. Defaults to 1.\n max_len (int): The maximum order of n-gram this function should extract. Defaults to 4.\n\nReturns:\n 'google_bleu': google_bleu score\n\nExamples:\n Example 1:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses, references=list_of_references)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.44\n\n Example 2:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n >>> ref1b = ['It', 'is', 'a', 'guide', 'to', 'action', 'that',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'will', 'never',\n ... 'heed', 'the', 'cat', 'commands']\n >>> ref1c = ['It', 'is', 'the', 'practical', 'guide', 'for', 'the',\n ... 'rubber', 'duck', 'army', 'never', 'to', 'heed', 'the', 'directions',\n ... 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a, ref1b, ref1c], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses, references=list_of_references)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.61\n\n Example 3:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n >>> ref1b = ['It', 'is', 'a', 'guide', 'to', 'action', 'that',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'will', 'never',\n ... 'heed', 'the', 'cat', 'commands']\n >>> ref1c = ['It', 'is', 'the', 'practical', 'guide', 'for', 'the',\n ... 'rubber', 'duck', 'army', 'never', 'to', 'heed', 'the', 'directions',\n ... 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a, ref1b, ref1c], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses, references=list_of_references, min_len=2)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.53\n\n Example 4:\n >>> hyp1 = ['It', 'is', 'a', 'guide', 'to', 'action', 'which',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'always',\n ... 'disobeys', 'the', 'commands', 'of', 'the', 'cat']\n >>> ref1a = ['It', 'is', 'the', 'guiding', 'principle', 'which',\n ... 'guarantees', 'the', 'rubber', 'duck', 'forces', 'never',\n ... 'being', 'under', 'the', 'command', 'of', 'the', 'cat']\n >>> ref1b = ['It', 'is', 'a', 'guide', 'to', 'action', 'that',\n ... 'ensures', 'that', 'the', 'rubber', 'duck', 'will', 'never',\n ... 'heed', 'the', 'cat', 'commands']\n >>> ref1c = ['It', 'is', 'the', 'practical', 'guide', 'for', 'the',\n ... 'rubber', 'duck', 'army', 'never', 'to', 'heed', 'the', 'directions',\n ... 'of', 'the', 'cat']\n\n >>> hyp2 = ['he', 'read', 'the', 'book', 'because', 'he', 'was',\n ... 'interested', 'in', 'world', 'history']\n >>> ref2a = ['he', 'was', 'interested', 'in', 'world', 'history',\n ... 'because', 'he', 'read', 'the', 'book']\n\n >>> list_of_references = [[ref1a, ref1b, ref1c], [ref2a]]\n >>> hypotheses = [hyp1, hyp2]\n >>> google_bleu = datasets.load_metric(\"google_bleu\")\n >>> results = google_bleu.compute(predictions=hypotheses,references=list_of_references, min_len=2, max_len=6)\n >>> print(round(results[\"google_bleu\"], 2))\n 0.4\n"
@datasets.utils.file_utils.add_start_docstrings(_DESCRIPTION , _KWARGS_DESCRIPTION )
class snake_case__ ( datasets.Metric ):
"""simple docstring"""
def lowercase_ ( self : str ) ->MetricInfo:
return datasets.MetricInfo(
description=_DESCRIPTION, citation=_CITATION, inputs_description=_KWARGS_DESCRIPTION, features=datasets.Features(
{
'predictions': datasets.Sequence(datasets.Value('string', id='token' ), id='sequence' ),
'references': datasets.Sequence(
datasets.Sequence(datasets.Value('string', id='token' ), id='sequence' ), id='references' ),
} ), )
def lowercase_ ( self : str, _snake_case : List[List[List[str]]], _snake_case : List[List[str]], _snake_case : int = 1, _snake_case : int = 4, ) ->Dict[str, float]:
return {
"google_bleu": gleu_score.corpus_gleu(
list_of_references=_snake_case, hypotheses=_snake_case, min_len=_snake_case, max_len=_snake_case )
}
| 277 | 0 |
from math import factorial
def __UpperCAmelCase ( a_ , a_ , a_):
if successes > trials:
raise ValueError('successes must be lower or equal to trials')
if trials < 0 or successes < 0:
raise ValueError('the function is defined for non-negative integers')
if not isinstance(a_ , a_) or not isinstance(a_ , a_):
raise ValueError('the function is defined for non-negative integers')
if not 0 < prob < 1:
raise ValueError('prob has to be in range of 1 - 0')
snake_case_ = (prob**successes) * ((1 - prob) ** (trials - successes))
# Calculate the binomial coefficient: n! / k!(n-k)!
snake_case_ = float(factorial(a_))
coefficient /= factorial(a_) * factorial(trials - successes)
return probability * coefficient
if __name__ == "__main__":
from doctest import testmod
testmod()
print("Probability of 2 successes out of 4 trails")
print("with probability of 0.75 is:", end=" ")
print(binomial_distribution(2, 4, 0.75))
| 178 |
from math import factorial
def lowercase_ (A : int , A : int , A : float ):
if successes > trials:
raise ValueError('successes must be lower or equal to trials' )
if trials < 0 or successes < 0:
raise ValueError('the function is defined for non-negative integers' )
if not isinstance(A , A ) or not isinstance(A , A ):
raise ValueError('the function is defined for non-negative integers' )
if not 0 < prob < 1:
raise ValueError('prob has to be in range of 1 - 0' )
snake_case__ : List[Any] = (prob**successes) * ((1 - prob) ** (trials - successes))
# Calculate the binomial coefficient: n! / k!(n-k)!
snake_case__ : List[str] = float(factorial(A ) )
coefficient /= factorial(A ) * factorial(trials - successes )
return probability * coefficient
if __name__ == "__main__":
from doctest import testmod
testmod()
print("Probability of 2 successes out of 4 trails")
print("with probability of 0.75 is:", end=" ")
print(binomial_distribution(2, 4, 0.75))
| 277 | 0 |
"""simple docstring"""
import os
import unittest
from transformers import BertTokenizerFast
from transformers.models.bert.tokenization_bert import (
VOCAB_FILES_NAMES,
BasicTokenizer,
BertTokenizer,
WordpieceTokenizer,
_is_control,
_is_punctuation,
_is_whitespace,
)
from transformers.testing_utils import require_tokenizers, slow
from ...test_tokenization_common import TokenizerTesterMixin, filter_non_english
@require_tokenizers
class lowerCAmelCase ( lowerCAmelCase_ , unittest.TestCase ):
'''simple docstring'''
SCREAMING_SNAKE_CASE_ : Any = BertTokenizer
SCREAMING_SNAKE_CASE_ : Optional[int] = BertTokenizerFast
SCREAMING_SNAKE_CASE_ : Optional[Any] = True
SCREAMING_SNAKE_CASE_ : Tuple = True
SCREAMING_SNAKE_CASE_ : int = filter_non_english
def __A ( self ) -> Any:
super().setUp()
SCREAMING_SNAKE_CASE = [
'[UNK]',
'[CLS]',
'[SEP]',
'[PAD]',
'[MASK]',
'want',
'##want',
'##ed',
'wa',
'un',
'runn',
'##ing',
',',
'low',
'lowest',
]
SCREAMING_SNAKE_CASE = os.path.join(self.tmpdirname , VOCAB_FILES_NAMES['vocab_file'] )
with open(self.vocab_file , 'w' , encoding='utf-8' ) as vocab_writer:
vocab_writer.write(''.join([x + '\n' for x in vocab_tokens] ) )
def __A ( self , lowerCAmelCase__ ) -> str:
SCREAMING_SNAKE_CASE = 'UNwant\u00E9d,running'
SCREAMING_SNAKE_CASE = 'unwanted, running'
return input_text, output_text
def __A ( self ) -> Dict:
SCREAMING_SNAKE_CASE = self.tokenizer_class(self.vocab_file )
SCREAMING_SNAKE_CASE = tokenizer.tokenize('UNwant\u00E9d,running' )
self.assertListEqual(_snake_case , ['un', '##want', '##ed', ',', 'runn', '##ing'] )
self.assertListEqual(tokenizer.convert_tokens_to_ids(_snake_case ) , [9, 6, 7, 12, 10, 11] )
def __A ( self ) -> List[str]:
if not self.test_rust_tokenizer:
return
SCREAMING_SNAKE_CASE = self.get_tokenizer()
SCREAMING_SNAKE_CASE = self.get_rust_tokenizer()
SCREAMING_SNAKE_CASE = 'UNwant\u00E9d,running'
SCREAMING_SNAKE_CASE = tokenizer.tokenize(_snake_case )
SCREAMING_SNAKE_CASE = rust_tokenizer.tokenize(_snake_case )
self.assertListEqual(_snake_case , _snake_case )
SCREAMING_SNAKE_CASE = tokenizer.encode(_snake_case , add_special_tokens=_snake_case )
SCREAMING_SNAKE_CASE = rust_tokenizer.encode(_snake_case , add_special_tokens=_snake_case )
self.assertListEqual(_snake_case , _snake_case )
SCREAMING_SNAKE_CASE = self.get_rust_tokenizer()
SCREAMING_SNAKE_CASE = tokenizer.encode(_snake_case )
SCREAMING_SNAKE_CASE = rust_tokenizer.encode(_snake_case )
self.assertListEqual(_snake_case , _snake_case )
# With lower casing
SCREAMING_SNAKE_CASE = self.get_tokenizer(do_lower_case=_snake_case )
SCREAMING_SNAKE_CASE = self.get_rust_tokenizer(do_lower_case=_snake_case )
SCREAMING_SNAKE_CASE = 'UNwant\u00E9d,running'
SCREAMING_SNAKE_CASE = tokenizer.tokenize(_snake_case )
SCREAMING_SNAKE_CASE = rust_tokenizer.tokenize(_snake_case )
self.assertListEqual(_snake_case , _snake_case )
SCREAMING_SNAKE_CASE = tokenizer.encode(_snake_case , add_special_tokens=_snake_case )
SCREAMING_SNAKE_CASE = rust_tokenizer.encode(_snake_case , add_special_tokens=_snake_case )
self.assertListEqual(_snake_case , _snake_case )
SCREAMING_SNAKE_CASE = self.get_rust_tokenizer()
SCREAMING_SNAKE_CASE = tokenizer.encode(_snake_case )
SCREAMING_SNAKE_CASE = rust_tokenizer.encode(_snake_case )
self.assertListEqual(_snake_case , _snake_case )
def __A ( self ) -> Any:
SCREAMING_SNAKE_CASE = BasicTokenizer()
self.assertListEqual(tokenizer.tokenize('ah\u535A\u63A8zz' ) , ['ah', '\u535A', '\u63A8', 'zz'] )
def __A ( self ) -> Any:
SCREAMING_SNAKE_CASE = BasicTokenizer(do_lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHeLLo!how \n Are yoU? ' ) , ['hello', '!', 'how', 'are', 'you', '?'] )
self.assertListEqual(tokenizer.tokenize('H\u00E9llo' ) , ['hello'] )
def __A ( self ) -> Any:
SCREAMING_SNAKE_CASE = BasicTokenizer(do_lower_case=_snake_case , strip_accents=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHäLLo!how \n Are yoU? ' ) , ['hällo', '!', 'how', 'are', 'you', '?'] )
self.assertListEqual(tokenizer.tokenize('H\u00E9llo' ) , ['h\u00E9llo'] )
def __A ( self ) -> str:
SCREAMING_SNAKE_CASE = BasicTokenizer(do_lower_case=_snake_case , strip_accents=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHäLLo!how \n Are yoU? ' ) , ['hallo', '!', 'how', 'are', 'you', '?'] )
self.assertListEqual(tokenizer.tokenize('H\u00E9llo' ) , ['hello'] )
def __A ( self ) -> List[str]:
SCREAMING_SNAKE_CASE = BasicTokenizer(do_lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHäLLo!how \n Are yoU? ' ) , ['hallo', '!', 'how', 'are', 'you', '?'] )
self.assertListEqual(tokenizer.tokenize('H\u00E9llo' ) , ['hello'] )
def __A ( self ) -> Optional[Any]:
SCREAMING_SNAKE_CASE = BasicTokenizer(do_lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHeLLo!how \n Are yoU? ' ) , ['HeLLo', '!', 'how', 'Are', 'yoU', '?'] )
def __A ( self ) -> Optional[int]:
SCREAMING_SNAKE_CASE = BasicTokenizer(do_lower_case=_snake_case , strip_accents=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHäLLo!how \n Are yoU? ' ) , ['HäLLo', '!', 'how', 'Are', 'yoU', '?'] )
def __A ( self ) -> List[Any]:
SCREAMING_SNAKE_CASE = BasicTokenizer(do_lower_case=_snake_case , strip_accents=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHäLLo!how \n Are yoU? ' ) , ['HaLLo', '!', 'how', 'Are', 'yoU', '?'] )
def __A ( self ) -> Optional[Any]:
SCREAMING_SNAKE_CASE = BasicTokenizer(do_lower_case=_snake_case , never_split=['[UNK]'] )
self.assertListEqual(
tokenizer.tokenize(' \tHeLLo!how \n Are yoU? [UNK]' ) , ['HeLLo', '!', 'how', 'Are', 'yoU', '?', '[UNK]'] )
def __A ( self ) -> Union[str, Any]:
SCREAMING_SNAKE_CASE = BasicTokenizer()
SCREAMING_SNAKE_CASE = 'a\n\'ll !!to?\'d of, can\'t.'
SCREAMING_SNAKE_CASE = ['a', '\'', 'll', '!', '!', 'to', '?', '\'', 'd', 'of', ',', 'can', '\'', 't', '.']
self.assertListEqual(tokenizer.tokenize(_snake_case ) , _snake_case )
def __A ( self ) -> List[Any]:
SCREAMING_SNAKE_CASE = ['[UNK]', '[CLS]', '[SEP]', 'want', '##want', '##ed', 'wa', 'un', 'runn', '##ing']
SCREAMING_SNAKE_CASE = {}
for i, token in enumerate(_snake_case ):
SCREAMING_SNAKE_CASE = i
SCREAMING_SNAKE_CASE = WordpieceTokenizer(vocab=_snake_case , unk_token='[UNK]' )
self.assertListEqual(tokenizer.tokenize('' ) , [] )
self.assertListEqual(tokenizer.tokenize('unwanted running' ) , ['un', '##want', '##ed', 'runn', '##ing'] )
self.assertListEqual(tokenizer.tokenize('unwantedX running' ) , ['[UNK]', 'runn', '##ing'] )
def __A ( self ) -> Any:
self.assertTrue(_is_whitespace(' ' ) )
self.assertTrue(_is_whitespace('\t' ) )
self.assertTrue(_is_whitespace('\r' ) )
self.assertTrue(_is_whitespace('\n' ) )
self.assertTrue(_is_whitespace('\u00A0' ) )
self.assertFalse(_is_whitespace('A' ) )
self.assertFalse(_is_whitespace('-' ) )
def __A ( self ) -> Optional[int]:
self.assertTrue(_is_control('\u0005' ) )
self.assertFalse(_is_control('A' ) )
self.assertFalse(_is_control(' ' ) )
self.assertFalse(_is_control('\t' ) )
self.assertFalse(_is_control('\r' ) )
def __A ( self ) -> Tuple:
self.assertTrue(_is_punctuation('-' ) )
self.assertTrue(_is_punctuation('$' ) )
self.assertTrue(_is_punctuation('`' ) )
self.assertTrue(_is_punctuation('.' ) )
self.assertFalse(_is_punctuation('A' ) )
self.assertFalse(_is_punctuation(' ' ) )
def __A ( self ) -> Any:
SCREAMING_SNAKE_CASE = self.get_tokenizer()
SCREAMING_SNAKE_CASE = self.get_rust_tokenizer()
# Example taken from the issue https://github.com/huggingface/tokenizers/issues/340
self.assertListEqual([tokenizer.tokenize(_snake_case ) for t in ['Test', '\xad', 'test']] , [['[UNK]'], [], ['[UNK]']] )
self.assertListEqual(
[rust_tokenizer.tokenize(_snake_case ) for t in ['Test', '\xad', 'test']] , [['[UNK]'], [], ['[UNK]']] )
@slow
def __A ( self ) -> Any:
SCREAMING_SNAKE_CASE = self.tokenizer_class.from_pretrained('bert-base-uncased' )
SCREAMING_SNAKE_CASE = tokenizer.encode('sequence builders' , add_special_tokens=_snake_case )
SCREAMING_SNAKE_CASE = tokenizer.encode('multi-sequence build' , add_special_tokens=_snake_case )
SCREAMING_SNAKE_CASE = tokenizer.build_inputs_with_special_tokens(_snake_case )
SCREAMING_SNAKE_CASE = tokenizer.build_inputs_with_special_tokens(_snake_case , _snake_case )
assert encoded_sentence == [101] + text + [102]
assert encoded_pair == [101] + text + [102] + text_a + [102]
def __A ( self ) -> List[Any]:
for tokenizer, pretrained_name, kwargs in self.tokenizers_list:
with self.subTest(F'{tokenizer.__class__.__name__} ({pretrained_name})' ):
SCREAMING_SNAKE_CASE = self.rust_tokenizer_class.from_pretrained(_snake_case , **_snake_case )
SCREAMING_SNAKE_CASE = F'A, naïve {tokenizer_r.mask_token} AllenNLP sentence.'
SCREAMING_SNAKE_CASE = tokenizer_r.encode_plus(
_snake_case , return_attention_mask=_snake_case , return_token_type_ids=_snake_case , return_offsets_mapping=_snake_case , add_special_tokens=_snake_case , )
SCREAMING_SNAKE_CASE = tokenizer_r.do_lower_case if hasattr(_snake_case , 'do_lower_case' ) else False
SCREAMING_SNAKE_CASE = (
[
((0, 0), tokenizer_r.cls_token),
((0, 1), 'A'),
((1, 2), ','),
((3, 5), 'na'),
((5, 6), '##ï'),
((6, 8), '##ve'),
((9, 15), tokenizer_r.mask_token),
((16, 21), 'Allen'),
((21, 23), '##NL'),
((23, 24), '##P'),
((25, 33), 'sentence'),
((33, 34), '.'),
((0, 0), tokenizer_r.sep_token),
]
if not do_lower_case
else [
((0, 0), tokenizer_r.cls_token),
((0, 1), 'a'),
((1, 2), ','),
((3, 8), 'naive'),
((9, 15), tokenizer_r.mask_token),
((16, 21), 'allen'),
((21, 23), '##nl'),
((23, 24), '##p'),
((25, 33), 'sentence'),
((33, 34), '.'),
((0, 0), tokenizer_r.sep_token),
]
)
self.assertEqual(
[e[1] for e in expected_results] , tokenizer_r.convert_ids_to_tokens(tokens['input_ids'] ) )
self.assertEqual([e[0] for e in expected_results] , tokens['offset_mapping'] )
def __A ( self ) -> Optional[Any]:
SCREAMING_SNAKE_CASE = ['的', '人', '有']
SCREAMING_SNAKE_CASE = ''.join(_snake_case )
for tokenizer, pretrained_name, kwargs in self.tokenizers_list:
with self.subTest(F'{tokenizer.__class__.__name__} ({pretrained_name})' ):
SCREAMING_SNAKE_CASE = True
SCREAMING_SNAKE_CASE = self.tokenizer_class.from_pretrained(_snake_case , **_snake_case )
SCREAMING_SNAKE_CASE = self.rust_tokenizer_class.from_pretrained(_snake_case , **_snake_case )
SCREAMING_SNAKE_CASE = tokenizer_p.encode(_snake_case , add_special_tokens=_snake_case )
SCREAMING_SNAKE_CASE = tokenizer_r.encode(_snake_case , add_special_tokens=_snake_case )
SCREAMING_SNAKE_CASE = tokenizer_r.convert_ids_to_tokens(_snake_case )
SCREAMING_SNAKE_CASE = tokenizer_p.convert_ids_to_tokens(_snake_case )
# it is expected that each Chinese character is not preceded by "##"
self.assertListEqual(_snake_case , _snake_case )
self.assertListEqual(_snake_case , _snake_case )
SCREAMING_SNAKE_CASE = False
SCREAMING_SNAKE_CASE = self.rust_tokenizer_class.from_pretrained(_snake_case , **_snake_case )
SCREAMING_SNAKE_CASE = self.tokenizer_class.from_pretrained(_snake_case , **_snake_case )
SCREAMING_SNAKE_CASE = tokenizer_r.encode(_snake_case , add_special_tokens=_snake_case )
SCREAMING_SNAKE_CASE = tokenizer_p.encode(_snake_case , add_special_tokens=_snake_case )
SCREAMING_SNAKE_CASE = tokenizer_r.convert_ids_to_tokens(_snake_case )
SCREAMING_SNAKE_CASE = tokenizer_p.convert_ids_to_tokens(_snake_case )
# it is expected that only the first Chinese character is not preceded by "##".
SCREAMING_SNAKE_CASE = [
F'##{token}' if idx != 0 else token for idx, token in enumerate(_snake_case )
]
self.assertListEqual(_snake_case , _snake_case )
self.assertListEqual(_snake_case , _snake_case )
| 113 |
from collections import UserDict
from typing import Union
import numpy as np
import requests
from ..utils import (
add_end_docstrings,
logging,
)
from .audio_classification import ffmpeg_read
from .base import PIPELINE_INIT_ARGS, Pipeline
a_ :List[Any] = logging.get_logger(__name__)
@add_end_docstrings(lowerCAmelCase_ )
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
def __init__( self : Optional[Any], **_snake_case : str ) ->Dict:
super().__init__(**_snake_case )
if self.framework != "pt":
raise ValueError(F'''The {self.__class__} is only available in PyTorch.''' )
# No specific FOR_XXX available yet
def __call__( self : Union[str, Any], _snake_case : Union[np.ndarray, bytes, str], **_snake_case : Tuple ) ->Dict:
return super().__call__(_snake_case, **_snake_case )
def lowercase_ ( self : Tuple, **_snake_case : Any ) ->Union[str, Any]:
snake_case__ : str = {}
if "candidate_labels" in kwargs:
snake_case__ : str = kwargs['candidate_labels']
if "hypothesis_template" in kwargs:
snake_case__ : str = kwargs['hypothesis_template']
return preprocess_params, {}, {}
def lowercase_ ( self : Dict, _snake_case : str, _snake_case : Optional[int]=None, _snake_case : List[str]="This is a sound of {}." ) ->int:
if isinstance(_snake_case, _snake_case ):
if audio.startswith('http://' ) or audio.startswith('https://' ):
# We need to actually check for a real protocol, otherwise it's impossible to use a local file
# like http_huggingface_co.png
snake_case__ : List[Any] = requests.get(_snake_case ).content
else:
with open(_snake_case, 'rb' ) as f:
snake_case__ : Union[str, Any] = f.read()
if isinstance(_snake_case, _snake_case ):
snake_case__ : List[Any] = ffmpeg_read(_snake_case, self.feature_extractor.sampling_rate )
if not isinstance(_snake_case, np.ndarray ):
raise ValueError('We expect a numpy ndarray as input' )
if len(audio.shape ) != 1:
raise ValueError('We expect a single channel audio input for ZeroShotAudioClassificationPipeline' )
snake_case__ : Tuple = self.feature_extractor(
[audio], sampling_rate=self.feature_extractor.sampling_rate, return_tensors='pt' )
snake_case__ : int = candidate_labels
snake_case__ : int = [hypothesis_template.format(_snake_case ) for x in candidate_labels]
snake_case__ : Optional[int] = self.tokenizer(_snake_case, return_tensors=self.framework, padding=_snake_case )
snake_case__ : List[Any] = [text_inputs]
return inputs
def lowercase_ ( self : Optional[int], _snake_case : Optional[Any] ) ->int:
snake_case__ : Optional[int] = model_inputs.pop('candidate_labels' )
snake_case__ : str = model_inputs.pop('text_inputs' )
if isinstance(text_inputs[0], _snake_case ):
snake_case__ : Optional[Any] = text_inputs[0]
else:
# Batching case.
snake_case__ : int = text_inputs[0][0]
snake_case__ : Any = self.model(**_snake_case, **_snake_case )
snake_case__ : List[Any] = {
'candidate_labels': candidate_labels,
'logits': outputs.logits_per_audio,
}
return model_outputs
def lowercase_ ( self : Union[str, Any], _snake_case : str ) ->List[str]:
snake_case__ : int = model_outputs.pop('candidate_labels' )
snake_case__ : List[Any] = model_outputs['logits'][0]
if self.framework == "pt":
snake_case__ : Tuple = logits.softmax(dim=0 )
snake_case__ : Union[str, Any] = probs.tolist()
else:
raise ValueError('`tf` framework not supported.' )
snake_case__ : Union[str, Any] = [
{'score': score, 'label': candidate_label}
for score, candidate_label in sorted(zip(_snake_case, _snake_case ), key=lambda _snake_case : -x[0] )
]
return result
| 277 | 0 |
'''simple docstring'''
def UpperCamelCase_ ( ) -> List[Any]:
'''simple docstring'''
return [
a * b * (10_00 - a - b)
for a in range(1 , 9_99 )
for b in range(snake_case_ , 9_99 )
if (a * a + b * b == (10_00 - a - b) ** 2)
][0]
if __name__ == "__main__":
print(f'{solution() = }')
| 229 |
import collections
import inspect
import unittest
from transformers import SwinvaConfig
from transformers.testing_utils import require_torch, require_vision, slow, torch_device
from transformers.utils import cached_property, is_torch_available, is_vision_available
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, _config_zero_init, floats_tensor, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from torch import nn
from transformers import SwinvaForImageClassification, SwinvaForMaskedImageModeling, SwinvaModel
from transformers.models.swinva.modeling_swinva import SWINV2_PRETRAINED_MODEL_ARCHIVE_LIST
if is_vision_available():
from PIL import Image
from transformers import AutoImageProcessor
class snake_case__ :
"""simple docstring"""
def __init__( self : Tuple, _snake_case : Any, _snake_case : int=1_3, _snake_case : Optional[int]=3_2, _snake_case : Tuple=2, _snake_case : Any=3, _snake_case : Tuple=1_6, _snake_case : Tuple=[1, 2, 1], _snake_case : Dict=[2, 2, 4], _snake_case : str=2, _snake_case : Union[str, Any]=2.0, _snake_case : Dict=True, _snake_case : Dict=0.0, _snake_case : str=0.0, _snake_case : str=0.1, _snake_case : List[str]="gelu", _snake_case : int=False, _snake_case : Optional[Any]=True, _snake_case : List[Any]=0.0_2, _snake_case : Union[str, Any]=1e-5, _snake_case : Union[str, Any]=True, _snake_case : List[Any]=None, _snake_case : Any=True, _snake_case : List[Any]=1_0, _snake_case : str=8, ) ->Union[str, Any]:
snake_case__ : Any = parent
snake_case__ : Tuple = batch_size
snake_case__ : Tuple = image_size
snake_case__ : Any = patch_size
snake_case__ : Optional[int] = num_channels
snake_case__ : Tuple = embed_dim
snake_case__ : Any = depths
snake_case__ : Any = num_heads
snake_case__ : List[str] = window_size
snake_case__ : Dict = mlp_ratio
snake_case__ : Optional[int] = qkv_bias
snake_case__ : Optional[Any] = hidden_dropout_prob
snake_case__ : List[str] = attention_probs_dropout_prob
snake_case__ : Union[str, Any] = drop_path_rate
snake_case__ : str = hidden_act
snake_case__ : Union[str, Any] = use_absolute_embeddings
snake_case__ : Union[str, Any] = patch_norm
snake_case__ : Any = layer_norm_eps
snake_case__ : Tuple = initializer_range
snake_case__ : Dict = is_training
snake_case__ : Any = scope
snake_case__ : Optional[Any] = use_labels
snake_case__ : str = type_sequence_label_size
snake_case__ : List[Any] = encoder_stride
def lowercase_ ( self : Tuple ) ->str:
snake_case__ : Tuple = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size] )
snake_case__ : List[Any] = None
if self.use_labels:
snake_case__ : Optional[Any] = ids_tensor([self.batch_size], self.type_sequence_label_size )
snake_case__ : Any = self.get_config()
return config, pixel_values, labels
def lowercase_ ( self : Optional[int] ) ->Optional[int]:
return SwinvaConfig(
image_size=self.image_size, patch_size=self.patch_size, num_channels=self.num_channels, embed_dim=self.embed_dim, depths=self.depths, num_heads=self.num_heads, window_size=self.window_size, mlp_ratio=self.mlp_ratio, qkv_bias=self.qkv_bias, hidden_dropout_prob=self.hidden_dropout_prob, attention_probs_dropout_prob=self.attention_probs_dropout_prob, drop_path_rate=self.drop_path_rate, hidden_act=self.hidden_act, use_absolute_embeddings=self.use_absolute_embeddings, path_norm=self.patch_norm, layer_norm_eps=self.layer_norm_eps, initializer_range=self.initializer_range, encoder_stride=self.encoder_stride, )
def lowercase_ ( self : Optional[int], _snake_case : str, _snake_case : List[str], _snake_case : int ) ->Dict:
snake_case__ : List[Any] = SwinvaModel(config=_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Optional[int] = model(_snake_case )
snake_case__ : List[Any] = ((config.image_size // config.patch_size) ** 2) // (4 ** (len(config.depths ) - 1))
snake_case__ : List[Any] = int(config.embed_dim * 2 ** (len(config.depths ) - 1) )
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, expected_seq_len, expected_dim) )
def lowercase_ ( self : Optional[Any], _snake_case : Any, _snake_case : List[str], _snake_case : Dict ) ->List[Any]:
snake_case__ : List[str] = SwinvaForMaskedImageModeling(config=_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Union[str, Any] = model(_snake_case )
self.parent.assertEqual(
result.logits.shape, (self.batch_size, self.num_channels, self.image_size, self.image_size) )
# test greyscale images
snake_case__ : Optional[Any] = 1
snake_case__ : Optional[int] = SwinvaForMaskedImageModeling(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Tuple = floats_tensor([self.batch_size, 1, self.image_size, self.image_size] )
snake_case__ : Any = model(_snake_case )
self.parent.assertEqual(result.logits.shape, (self.batch_size, 1, self.image_size, self.image_size) )
def lowercase_ ( self : List[str], _snake_case : int, _snake_case : List[Any], _snake_case : Optional[int] ) ->Any:
snake_case__ : Tuple = self.type_sequence_label_size
snake_case__ : int = SwinvaForImageClassification(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Tuple = model(_snake_case, labels=_snake_case )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.type_sequence_label_size) )
def lowercase_ ( self : Any ) ->Dict:
snake_case__ : str = self.prepare_config_and_inputs()
snake_case__ , snake_case__ , snake_case__ : List[str] = config_and_inputs
snake_case__ : Union[str, Any] = {'pixel_values': pixel_values}
return config, inputs_dict
@require_torch
class snake_case__ ( lowerCAmelCase_ , lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = (
(SwinvaModel, SwinvaForImageClassification, SwinvaForMaskedImageModeling) if is_torch_available() else ()
)
_SCREAMING_SNAKE_CASE = (
{"""feature-extraction""": SwinvaModel, """image-classification""": SwinvaForImageClassification}
if is_torch_available()
else {}
)
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
def lowercase_ ( self : Union[str, Any] ) ->Dict:
snake_case__ : Optional[int] = SwinvaModelTester(self )
snake_case__ : int = ConfigTester(self, config_class=_snake_case, embed_dim=3_7 )
def lowercase_ ( self : Tuple ) ->int:
self.config_tester.create_and_test_config_to_json_string()
self.config_tester.create_and_test_config_to_json_file()
self.config_tester.create_and_test_config_from_and_save_pretrained()
self.config_tester.create_and_test_config_with_num_labels()
self.config_tester.check_config_can_be_init_without_params()
self.config_tester.check_config_arguments_init()
def lowercase_ ( self : Any ) ->str:
snake_case__ : int = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*_snake_case )
@unittest.skip(reason='Got `CUDA error: misaligned address` with PyTorch 2.0.0.' )
def lowercase_ ( self : Any ) ->Union[str, Any]:
pass
@unittest.skip(reason='Swinv2 does not use inputs_embeds' )
def lowercase_ ( self : str ) ->Union[str, Any]:
pass
def lowercase_ ( self : Optional[Any] ) ->Union[str, Any]:
snake_case__ , snake_case__ : Optional[int] = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
snake_case__ : Union[str, Any] = model_class(_snake_case )
self.assertIsInstance(model.get_input_embeddings(), (nn.Module) )
snake_case__ : Union[str, Any] = model.get_output_embeddings()
self.assertTrue(x is None or isinstance(_snake_case, nn.Linear ) )
def lowercase_ ( self : List[str] ) ->Optional[int]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
snake_case__ : Any = model_class(_snake_case )
snake_case__ : Dict = inspect.signature(model.forward )
# signature.parameters is an OrderedDict => so arg_names order is deterministic
snake_case__ : Optional[Any] = [*signature.parameters.keys()]
snake_case__ : List[Any] = ['pixel_values']
self.assertListEqual(arg_names[:1], _snake_case )
def lowercase_ ( self : str ) ->Union[str, Any]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : int = True
for model_class in self.all_model_classes:
snake_case__ : str = True
snake_case__ : Union[str, Any] = False
snake_case__ : Tuple = True
snake_case__ : int = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : Optional[int] = model(**self._prepare_for_class(_snake_case, _snake_case ) )
snake_case__ : List[str] = outputs.attentions
snake_case__ : List[Any] = len(self.model_tester.depths )
self.assertEqual(len(_snake_case ), _snake_case )
# check that output_attentions also work using config
del inputs_dict["output_attentions"]
snake_case__ : str = True
snake_case__ : Tuple = config.window_size**2
snake_case__ : Optional[int] = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : str = model(**self._prepare_for_class(_snake_case, _snake_case ) )
snake_case__ : Tuple = outputs.attentions
self.assertEqual(len(_snake_case ), _snake_case )
self.assertListEqual(
list(attentions[0].shape[-3:] ), [self.model_tester.num_heads[0], window_size_squared, window_size_squared], )
snake_case__ : Optional[Any] = len(_snake_case )
# Check attention is always last and order is fine
snake_case__ : Optional[int] = True
snake_case__ : Dict = True
snake_case__ : List[Any] = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : Optional[int] = model(**self._prepare_for_class(_snake_case, _snake_case ) )
if hasattr(self.model_tester, 'num_hidden_states_types' ):
snake_case__ : str = self.model_tester.num_hidden_states_types
else:
# also another +1 for reshaped_hidden_states
snake_case__ : Dict = 2
self.assertEqual(out_len + added_hidden_states, len(_snake_case ) )
snake_case__ : Any = outputs.attentions
self.assertEqual(len(_snake_case ), _snake_case )
self.assertListEqual(
list(self_attentions[0].shape[-3:] ), [self.model_tester.num_heads[0], window_size_squared, window_size_squared], )
def lowercase_ ( self : Dict, _snake_case : Tuple, _snake_case : Any, _snake_case : int, _snake_case : Optional[int] ) ->str:
snake_case__ : Dict = model_class(_snake_case )
model.to(_snake_case )
model.eval()
with torch.no_grad():
snake_case__ : List[Any] = model(**self._prepare_for_class(_snake_case, _snake_case ) )
snake_case__ : Dict = outputs.hidden_states
snake_case__ : int = getattr(
self.model_tester, 'expected_num_hidden_layers', len(self.model_tester.depths ) + 1 )
self.assertEqual(len(_snake_case ), _snake_case )
# Swinv2 has a different seq_length
snake_case__ : int = (
config.patch_size
if isinstance(config.patch_size, collections.abc.Iterable )
else (config.patch_size, config.patch_size)
)
snake_case__ : Optional[Any] = (image_size[1] // patch_size[1]) * (image_size[0] // patch_size[0])
self.assertListEqual(
list(hidden_states[0].shape[-2:] ), [num_patches, self.model_tester.embed_dim], )
snake_case__ : Union[str, Any] = outputs.reshaped_hidden_states
self.assertEqual(len(_snake_case ), _snake_case )
snake_case__ , snake_case__ , snake_case__ , snake_case__ : str = reshaped_hidden_states[0].shape
snake_case__ : Any = (
reshaped_hidden_states[0].view(_snake_case, _snake_case, height * width ).permute(0, 2, 1 )
)
self.assertListEqual(
list(reshaped_hidden_states.shape[-2:] ), [num_patches, self.model_tester.embed_dim], )
def lowercase_ ( self : str ) ->List[Any]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : List[Any] = (
self.model_tester.image_size
if isinstance(self.model_tester.image_size, collections.abc.Iterable )
else (self.model_tester.image_size, self.model_tester.image_size)
)
for model_class in self.all_model_classes:
snake_case__ : Optional[int] = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, _snake_case )
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
snake_case__ : Dict = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, _snake_case )
def lowercase_ ( self : List[str] ) ->str:
snake_case__ , snake_case__ : Optional[Any] = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : List[str] = 3
snake_case__ : Union[str, Any] = (
self.model_tester.image_size
if isinstance(self.model_tester.image_size, collections.abc.Iterable )
else (self.model_tester.image_size, self.model_tester.image_size)
)
snake_case__ : str = (
config.patch_size
if isinstance(config.patch_size, collections.abc.Iterable )
else (config.patch_size, config.patch_size)
)
snake_case__ : Tuple = image_size[0] + patch_size[0] - (image_size[0] % patch_size[0])
snake_case__ : Optional[Any] = image_size[1] + patch_size[1] - (image_size[1] % patch_size[1])
for model_class in self.all_model_classes:
snake_case__ : int = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, (padded_height, padded_width) )
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
snake_case__ : List[str] = True
self.check_hidden_states_output(_snake_case, _snake_case, _snake_case, (padded_height, padded_width) )
def lowercase_ ( self : List[str] ) ->Optional[int]:
snake_case__ : Any = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_masked_image_modeling(*_snake_case )
def lowercase_ ( self : List[Any] ) ->str:
snake_case__ : Any = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_image_classification(*_snake_case )
@slow
def lowercase_ ( self : str ) ->Union[str, Any]:
for model_name in SWINV2_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
snake_case__ : Dict = SwinvaModel.from_pretrained(_snake_case )
self.assertIsNotNone(_snake_case )
def lowercase_ ( self : Optional[int] ) ->List[str]:
snake_case__ , snake_case__ : Any = self.model_tester.prepare_config_and_inputs_for_common()
snake_case__ : List[Any] = _config_zero_init(_snake_case )
for model_class in self.all_model_classes:
snake_case__ : List[str] = model_class(config=_snake_case )
for name, param in model.named_parameters():
if "embeddings" not in name and "logit_scale" not in name and param.requires_grad:
self.assertIn(
((param.data.mean() * 1e9).round() / 1e9).item(), [0.0, 1.0], msg=F'''Parameter {name} of model {model_class} seems not properly initialized''', )
@require_vision
@require_torch
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
@cached_property
def lowercase_ ( self : Union[str, Any] ) ->List[str]:
return (
AutoImageProcessor.from_pretrained('microsoft/swinv2-tiny-patch4-window8-256' )
if is_vision_available()
else None
)
@slow
def lowercase_ ( self : int ) ->List[Any]:
snake_case__ : Any = SwinvaForImageClassification.from_pretrained('microsoft/swinv2-tiny-patch4-window8-256' ).to(
_snake_case )
snake_case__ : int = self.default_image_processor
snake_case__ : Union[str, Any] = Image.open('./tests/fixtures/tests_samples/COCO/000000039769.png' )
snake_case__ : Optional[Any] = image_processor(images=_snake_case, return_tensors='pt' ).to(_snake_case )
# forward pass
with torch.no_grad():
snake_case__ : List[str] = model(**_snake_case )
# verify the logits
snake_case__ : int = torch.Size((1, 1_0_0_0) )
self.assertEqual(outputs.logits.shape, _snake_case )
snake_case__ : Optional[int] = torch.tensor([-0.3_9_4_7, -0.4_3_0_6, 0.0_0_2_6] ).to(_snake_case )
self.assertTrue(torch.allclose(outputs.logits[0, :3], _snake_case, atol=1e-4 ) )
| 277 | 0 |
"""simple docstring"""
import itertools
from dataclasses import dataclass
from typing import Any, Callable, Dict, List, Optional, Union
import pandas as pd
import pyarrow as pa
import datasets
import datasets.config
from datasets.features.features import require_storage_cast
from datasets.table import table_cast
from datasets.utils.py_utils import Literal
UpperCAmelCase = datasets.utils.logging.get_logger(__name__)
UpperCAmelCase = ["names", "prefix"]
UpperCAmelCase = ["warn_bad_lines", "error_bad_lines", "mangle_dupe_cols"]
UpperCAmelCase = ["encoding_errors", "on_bad_lines"]
UpperCAmelCase = ["date_format"]
@dataclass
class UpperCAmelCase_ ( datasets.BuilderConfig):
snake_case__ = ''','''
snake_case__ = None
snake_case__ = '''infer'''
snake_case__ = None
snake_case__ = None
snake_case__ = None
snake_case__ = None
snake_case__ = None
snake_case__ = True
snake_case__ = None
snake_case__ = None
snake_case__ = None
snake_case__ = None
snake_case__ = False
snake_case__ = None
snake_case__ = None
snake_case__ = None
snake_case__ = True
snake_case__ = True
snake_case__ = False
snake_case__ = True
snake_case__ = None
snake_case__ = '''.'''
snake_case__ = None
snake_case__ = '''"'''
snake_case__ = 0
snake_case__ = None
snake_case__ = None
snake_case__ = None
snake_case__ = None
snake_case__ = True
snake_case__ = True
snake_case__ = 0
snake_case__ = True
snake_case__ = False
snake_case__ = None
snake_case__ = 1_00_00
snake_case__ = None
snake_case__ = '''strict'''
snake_case__ = '''error'''
snake_case__ = None
def _UpperCamelCase ( self : Dict ) -> str:
if self.delimiter is not None:
_UpperCamelCase = self.delimiter
if self.column_names is not None:
_UpperCamelCase = self.column_names
@property
def _UpperCamelCase ( self : Optional[Any] ) -> List[Any]:
_UpperCamelCase = {
'sep': self.sep,
'header': self.header,
'names': self.names,
'index_col': self.index_col,
'usecols': self.usecols,
'prefix': self.prefix,
'mangle_dupe_cols': self.mangle_dupe_cols,
'engine': self.engine,
'converters': self.converters,
'true_values': self.true_values,
'false_values': self.false_values,
'skipinitialspace': self.skipinitialspace,
'skiprows': self.skiprows,
'nrows': self.nrows,
'na_values': self.na_values,
'keep_default_na': self.keep_default_na,
'na_filter': self.na_filter,
'verbose': self.verbose,
'skip_blank_lines': self.skip_blank_lines,
'thousands': self.thousands,
'decimal': self.decimal,
'lineterminator': self.lineterminator,
'quotechar': self.quotechar,
'quoting': self.quoting,
'escapechar': self.escapechar,
'comment': self.comment,
'encoding': self.encoding,
'dialect': self.dialect,
'error_bad_lines': self.error_bad_lines,
'warn_bad_lines': self.warn_bad_lines,
'skipfooter': self.skipfooter,
'doublequote': self.doublequote,
'memory_map': self.memory_map,
'float_precision': self.float_precision,
'chunksize': self.chunksize,
'encoding_errors': self.encoding_errors,
'on_bad_lines': self.on_bad_lines,
'date_format': self.date_format,
}
# some kwargs must not be passed if they don't have a default value
# some others are deprecated and we can also not pass them if they are the default value
for pd_read_csv_parameter in _PANDAS_READ_CSV_NO_DEFAULT_PARAMETERS + _PANDAS_READ_CSV_DEPRECATED_PARAMETERS:
if pd_read_csv_kwargs[pd_read_csv_parameter] == getattr(CsvConfig() , _snake_case ):
del pd_read_csv_kwargs[pd_read_csv_parameter]
# Remove 2.0 new arguments
if not (datasets.config.PANDAS_VERSION.major >= 2):
for pd_read_csv_parameter in _PANDAS_READ_CSV_NEW_2_0_0_PARAMETERS:
del pd_read_csv_kwargs[pd_read_csv_parameter]
# Remove 1.3 new arguments
if not (datasets.config.PANDAS_VERSION.major >= 1 and datasets.config.PANDAS_VERSION.minor >= 3):
for pd_read_csv_parameter in _PANDAS_READ_CSV_NEW_1_3_0_PARAMETERS:
del pd_read_csv_kwargs[pd_read_csv_parameter]
return pd_read_csv_kwargs
class UpperCAmelCase_ ( datasets.ArrowBasedBuilder):
snake_case__ = CsvConfig
def _UpperCamelCase ( self : Union[str, Any] ) -> Any:
return datasets.DatasetInfo(features=self.config.features )
def _UpperCamelCase ( self : int , __UpperCamelCase : Optional[Any] ) -> Any:
if not self.config.data_files:
raise ValueError(F'''At least one data file must be specified, but got data_files={self.config.data_files}''' )
_UpperCamelCase = dl_manager.download_and_extract(self.config.data_files )
if isinstance(_snake_case , (str, list, tuple) ):
_UpperCamelCase = data_files
if isinstance(_snake_case , _snake_case ):
_UpperCamelCase = [files]
_UpperCamelCase = [dl_manager.iter_files(_snake_case ) for file in files]
return [datasets.SplitGenerator(name=datasets.Split.TRAIN , gen_kwargs={'''files''': files} )]
_UpperCamelCase = []
for split_name, files in data_files.items():
if isinstance(_snake_case , _snake_case ):
_UpperCamelCase = [files]
_UpperCamelCase = [dl_manager.iter_files(_snake_case ) for file in files]
splits.append(datasets.SplitGenerator(name=_snake_case , gen_kwargs={'''files''': files} ) )
return splits
def _UpperCamelCase ( self : List[str] , __UpperCamelCase : pa.Table ) -> pa.Table:
if self.config.features is not None:
_UpperCamelCase = self.config.features.arrow_schema
if all(not require_storage_cast(_snake_case ) for feature in self.config.features.values() ):
# cheaper cast
_UpperCamelCase = pa.Table.from_arrays([pa_table[field.name] for field in schema] , schema=_snake_case )
else:
# more expensive cast; allows str <-> int/float or str to Audio for example
_UpperCamelCase = table_cast(_snake_case , _snake_case )
return pa_table
def _UpperCamelCase ( self : str , __UpperCamelCase : List[str] ) -> List[Any]:
_UpperCamelCase = self.config.features.arrow_schema if self.config.features else None
# dtype allows reading an int column as str
_UpperCamelCase = (
{
name: dtype.to_pandas_dtype() if not require_storage_cast(_snake_case ) else object
for name, dtype, feature in zip(schema.names , schema.types , self.config.features.values() )
}
if schema is not None
else None
)
for file_idx, file in enumerate(itertools.chain.from_iterable(_snake_case ) ):
_UpperCamelCase = pd.read_csv(_snake_case , iterator=_snake_case , dtype=_snake_case , **self.config.pd_read_csv_kwargs )
try:
for batch_idx, df in enumerate(_snake_case ):
_UpperCamelCase = pa.Table.from_pandas(_snake_case )
# Uncomment for debugging (will print the Arrow table size and elements)
# logger.warning(f"pa_table: {pa_table} num rows: {pa_table.num_rows}")
# logger.warning('\n'.join(str(pa_table.slice(i, 1).to_pydict()) for i in range(pa_table.num_rows)))
yield (file_idx, batch_idx), self._cast_table(_snake_case )
except ValueError as e:
logger.error(F'''Failed to read file \'{file}\' with error {type(_snake_case )}: {e}''' )
raise
| 256 |
import unittest
import numpy as np
from transformers.testing_utils import require_torch, require_vision
from transformers.utils import is_torch_available, is_vision_available
from ...test_image_processing_common import ImageProcessingSavingTestMixin, prepare_image_inputs
if is_torch_available():
import torch
if is_vision_available():
from PIL import Image
from transformers import PoolFormerImageProcessor
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
def __init__( self : Optional[int], _snake_case : List[Any], _snake_case : str=7, _snake_case : Tuple=3, _snake_case : List[str]=3_0, _snake_case : Tuple=4_0_0, _snake_case : Any=True, _snake_case : List[Any]=None, _snake_case : int=0.9, _snake_case : Optional[Any]=None, _snake_case : str=True, _snake_case : Union[str, Any]=[0.5, 0.5, 0.5], _snake_case : Union[str, Any]=[0.5, 0.5, 0.5], ) ->List[Any]:
snake_case__ : int = size if size is not None else {'shortest_edge': 3_0}
snake_case__ : Tuple = crop_size if crop_size is not None else {'height': 3_0, 'width': 3_0}
snake_case__ : Union[str, Any] = parent
snake_case__ : Dict = batch_size
snake_case__ : int = num_channels
snake_case__ : Tuple = min_resolution
snake_case__ : Any = max_resolution
snake_case__ : List[Any] = do_resize_and_center_crop
snake_case__ : str = size
snake_case__ : str = crop_pct
snake_case__ : List[str] = crop_size
snake_case__ : Optional[int] = do_normalize
snake_case__ : Tuple = image_mean
snake_case__ : Tuple = image_std
def lowercase_ ( self : Optional[int] ) ->int:
return {
"size": self.size,
"do_resize_and_center_crop": self.do_resize_and_center_crop,
"crop_pct": self.crop_pct,
"crop_size": self.crop_size,
"do_normalize": self.do_normalize,
"image_mean": self.image_mean,
"image_std": self.image_std,
}
@require_torch
@require_vision
class snake_case__ ( lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = PoolFormerImageProcessor if is_vision_available() else None
def lowercase_ ( self : Union[str, Any] ) ->Dict:
snake_case__ : Union[str, Any] = PoolFormerImageProcessingTester(self )
@property
def lowercase_ ( self : int ) ->Dict:
return self.image_processor_tester.prepare_image_processor_dict()
def lowercase_ ( self : Union[str, Any] ) ->Optional[int]:
snake_case__ : List[str] = self.image_processing_class(**self.image_processor_dict )
self.assertTrue(hasattr(_snake_case, 'do_resize_and_center_crop' ) )
self.assertTrue(hasattr(_snake_case, 'size' ) )
self.assertTrue(hasattr(_snake_case, 'crop_pct' ) )
self.assertTrue(hasattr(_snake_case, 'do_normalize' ) )
self.assertTrue(hasattr(_snake_case, 'image_mean' ) )
self.assertTrue(hasattr(_snake_case, 'image_std' ) )
def lowercase_ ( self : List[str] ) ->List[str]:
snake_case__ : Union[str, Any] = self.image_processing_class.from_dict(self.image_processor_dict )
self.assertEqual(image_processor.size, {'shortest_edge': 3_0} )
self.assertEqual(image_processor.crop_size, {'height': 3_0, 'width': 3_0} )
snake_case__ : int = self.image_processing_class.from_dict(self.image_processor_dict, size=4_2, crop_size=8_4 )
self.assertEqual(image_processor.size, {'shortest_edge': 4_2} )
self.assertEqual(image_processor.crop_size, {'height': 8_4, 'width': 8_4} )
def lowercase_ ( self : List[Any] ) ->List[Any]:
pass
def lowercase_ ( self : List[str] ) ->str:
# Initialize image_processing
snake_case__ : Union[str, Any] = self.image_processing_class(**self.image_processor_dict )
# create random PIL images
snake_case__ : List[str] = prepare_image_inputs(self.image_processor_tester, equal_resolution=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case, Image.Image )
# Test not batched input
snake_case__ : Optional[int] = image_processing(image_inputs[0], return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
# Test batched
snake_case__ : str = image_processing(_snake_case, return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
def lowercase_ ( self : int ) ->List[Any]:
# Initialize image_processing
snake_case__ : Optional[int] = self.image_processing_class(**self.image_processor_dict )
# create random numpy tensors
snake_case__ : Dict = prepare_image_inputs(self.image_processor_tester, equal_resolution=_snake_case, numpify=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case, np.ndarray )
# Test not batched input
snake_case__ : Dict = image_processing(image_inputs[0], return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
# Test batched
snake_case__ : List[Any] = image_processing(_snake_case, return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
def lowercase_ ( self : List[str] ) ->List[str]:
# Initialize image_processing
snake_case__ : Tuple = self.image_processing_class(**self.image_processor_dict )
# create random PyTorch tensors
snake_case__ : List[str] = prepare_image_inputs(self.image_processor_tester, equal_resolution=_snake_case, torchify=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case, torch.Tensor )
# Test not batched input
snake_case__ : Tuple = image_processing(image_inputs[0], return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
# Test batched
snake_case__ : Optional[Any] = image_processing(_snake_case, return_tensors='pt' ).pixel_values
self.assertEqual(
encoded_images.shape, (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size['height'],
self.image_processor_tester.crop_size['width'],
), )
| 277 | 0 |
from __future__ import annotations
from random import random
class A_ :
def __init__( self : Tuple , UpperCAmelCase : int | None = None ) -> Optional[Any]:
__lowerCAmelCase: int = value
__lowerCAmelCase: List[str] = random()
__lowerCAmelCase: Node | None = None
__lowerCAmelCase: Node | None = None
def __repr__( self : Optional[int] ) -> str:
from pprint import pformat
if self.left is None and self.right is None:
return F'''\'{self.value}: {self.prior:.5}\''''
else:
return pformat(
{F'''{self.value}: {self.prior:.5}''': (self.left, self.right)} , indent=1 )
def __str__( self : int ) -> str:
__lowerCAmelCase: str = str(self.value ) + ' '
__lowerCAmelCase: Optional[Any] = str(self.left or '' )
__lowerCAmelCase: List[Any] = str(self.right or '' )
return value + left + right
def _a ( SCREAMING_SNAKE_CASE : Node | None , SCREAMING_SNAKE_CASE : int ) -> List[Any]:
"""simple docstring"""
if root is None: # None tree is split into 2 Nones
return None, None
elif root.value is None:
return None, None
else:
if value < root.value:
__lowerCAmelCase: Optional[Any] = split(root.left , SCREAMING_SNAKE_CASE )
return left, root
else:
__lowerCAmelCase: int = split(root.right , SCREAMING_SNAKE_CASE )
return root, right
def _a ( SCREAMING_SNAKE_CASE : Node | None , SCREAMING_SNAKE_CASE : Node | None ) -> Union[str, Any]:
"""simple docstring"""
if (not left) or (not right): # If one node is None, return the other
return left or right
elif left.prior < right.prior:
__lowerCAmelCase: List[str] = merge(left.right , SCREAMING_SNAKE_CASE )
return left
else:
__lowerCAmelCase: List[str] = merge(SCREAMING_SNAKE_CASE , right.left )
return right
def _a ( SCREAMING_SNAKE_CASE : Node | None , SCREAMING_SNAKE_CASE : int ) -> Tuple:
"""simple docstring"""
__lowerCAmelCase: Any = Node(SCREAMING_SNAKE_CASE )
__lowerCAmelCase: str = split(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE )
return merge(merge(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE ) , SCREAMING_SNAKE_CASE )
def _a ( SCREAMING_SNAKE_CASE : Node | None , SCREAMING_SNAKE_CASE : int ) -> Any:
"""simple docstring"""
__lowerCAmelCase: Dict = split(SCREAMING_SNAKE_CASE , value - 1 )
__lowerCAmelCase: Union[str, Any] = split(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE )
return merge(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE )
def _a ( SCREAMING_SNAKE_CASE : Node | None ) -> Any:
"""simple docstring"""
if not root: # None
return
else:
inorder(root.left )
print(root.value , end=',' )
inorder(root.right )
def _a ( SCREAMING_SNAKE_CASE : Node | None , SCREAMING_SNAKE_CASE : str ) -> str:
"""simple docstring"""
for arg in args.split():
if arg[0] == "+":
__lowerCAmelCase: Optional[Any] = insert(SCREAMING_SNAKE_CASE , int(arg[1:] ) )
elif arg[0] == "-":
__lowerCAmelCase: Dict = erase(SCREAMING_SNAKE_CASE , int(arg[1:] ) )
else:
print('Unknown command' )
return root
def _a ( ) -> List[str]:
"""simple docstring"""
__lowerCAmelCase: Optional[int] = None
print(
'enter numbers to create a tree, + value to add value into treap, '
'- value to erase all nodes with value. \'q\' to quit. ' )
__lowerCAmelCase: str = input()
while args != "q":
__lowerCAmelCase: Dict = interact_treap(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE )
print(SCREAMING_SNAKE_CASE )
__lowerCAmelCase: List[str] = input()
print('good by!' )
if __name__ == "__main__":
import doctest
doctest.testmod()
main()
| 322 |
from collections import deque
from .hash_table import HashTable
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
def __init__( self : Optional[Any], *_snake_case : Optional[Any], **_snake_case : List[Any] ) ->Optional[int]:
super().__init__(*_snake_case, **_snake_case )
def lowercase_ ( self : Optional[Any], _snake_case : Tuple, _snake_case : Dict ) ->Dict:
snake_case__ : int = deque([] ) if self.values[key] is None else self.values[key]
self.values[key].appendleft(_snake_case )
snake_case__ : Dict = self.values[key]
def lowercase_ ( self : Any ) ->Optional[Any]:
return (
sum(self.charge_factor - len(_snake_case ) for slot in self.values )
/ self.size_table
* self.charge_factor
)
def lowercase_ ( self : Union[str, Any], _snake_case : str, _snake_case : Optional[int]=None ) ->Optional[Any]:
if not (
len(self.values[key] ) == self.charge_factor and self.values.count(_snake_case ) == 0
):
return key
return super()._collision_resolution(_snake_case, _snake_case )
| 277 | 0 |
import argparse
from argparse import Namespace
import torch
from torch import nn
from transformers import XGLMConfig, XGLMForCausalLM
def _A ( SCREAMING_SNAKE_CASE : Optional[Any] ):
"""simple docstring"""
a__ : Dict =[
'decoder.version',
'decoder.output_projection.weight',
'_float_tensor',
'decoder.embed_positions._float_tensor',
]
for k in ignore_keys:
state_dict.pop(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE )
def _A ( SCREAMING_SNAKE_CASE : Tuple ):
"""simple docstring"""
a__ : Tuple =emb.weight.shape
a__ : List[str] =nn.Linear(SCREAMING_SNAKE_CASE , SCREAMING_SNAKE_CASE , bias=SCREAMING_SNAKE_CASE )
a__ : Tuple =emb.weight.data
return lin_layer
def _A ( SCREAMING_SNAKE_CASE : Any ):
"""simple docstring"""
a__ : Any =torch.load(SCREAMING_SNAKE_CASE , map_location="cpu" )
a__ : Any =Namespace(**checkpoint["cfg"]["model"] )
a__ : Optional[Any] =checkpoint['model']
remove_ignore_keys_(SCREAMING_SNAKE_CASE )
a__ : List[Any] =state_dict['decoder.embed_tokens.weight'].shape[0]
a__ : str ={key.replace("decoder" , "model" ): val for key, val in state_dict.items()}
a__ : Optional[Any] =XGLMConfig(
vocab_size=SCREAMING_SNAKE_CASE , max_position_embeddings=args.max_target_positions , num_layers=args.decoder_layers , attention_heads=args.decoder_attention_heads , ffn_dim=args.decoder_ffn_embed_dim , d_model=args.decoder_embed_dim , layerdrop=args.decoder_layerdrop , dropout=args.dropout , attention_dropout=args.attention_dropout , activation_dropout=args.activation_dropout , activation_function="gelu" , scale_embedding=not args.no_scale_embedding , tie_word_embeddings=args.share_decoder_input_output_embed , )
a__ : Dict =XGLMForCausalLM(SCREAMING_SNAKE_CASE )
a__ : Tuple =model.load_state_dict(SCREAMING_SNAKE_CASE , strict=SCREAMING_SNAKE_CASE )
print(SCREAMING_SNAKE_CASE )
a__ : int =make_linear_from_emb(model.model.embed_tokens )
return model
if __name__ == "__main__":
UpperCAmelCase : Optional[Any] = argparse.ArgumentParser()
# Required parameters
parser.add_argument("""fairseq_path""", type=str, help="""path to a model.pt on local filesystem.""")
parser.add_argument("""pytorch_dump_folder_path""", default=None, type=str, help="""Path to the output PyTorch model.""")
UpperCAmelCase : List[str] = parser.parse_args()
UpperCAmelCase : int = convert_fairseq_xglm_checkpoint_from_disk(args.fairseq_path)
model.save_pretrained(args.pytorch_dump_folder_path)
| 95 |
def lowercase_ (A : Union[str, Any] , A : List[str] , A : int , A : Optional[int] ):
global f # a global dp table for knapsack
if f[i][j] < 0:
if j < wt[i - 1]:
snake_case__ : Union[str, Any] = mf_knapsack(i - 1 , A , A , A )
else:
snake_case__ : Any = max(
mf_knapsack(i - 1 , A , A , A ) , mf_knapsack(i - 1 , A , A , j - wt[i - 1] ) + val[i - 1] , )
snake_case__ : Optional[int] = val
return f[i][j]
def lowercase_ (A : Optional[int] , A : Union[str, Any] , A : str , A : Dict ):
snake_case__ : int = [[0] * (w + 1) for _ in range(n + 1 )]
for i in range(1 , n + 1 ):
for w_ in range(1 , w + 1 ):
if wt[i - 1] <= w_:
snake_case__ : Union[str, Any] = max(val[i - 1] + dp[i - 1][w_ - wt[i - 1]] , dp[i - 1][w_] )
else:
snake_case__ : str = dp[i - 1][w_]
return dp[n][w_], dp
def lowercase_ (A : int , A : list , A : list ):
if not (isinstance(A , (list, tuple) ) and isinstance(A , (list, tuple) )):
raise ValueError(
'Both the weights and values vectors must be either lists or tuples' )
snake_case__ : Dict = len(A )
if num_items != len(A ):
snake_case__ : str = (
'The number of weights must be the same as the number of values.\n'
F'''But got {num_items} weights and {len(A )} values'''
)
raise ValueError(A )
for i in range(A ):
if not isinstance(wt[i] , A ):
snake_case__ : Optional[int] = (
'All weights must be integers but got weight of '
F'''type {type(wt[i] )} at index {i}'''
)
raise TypeError(A )
snake_case__ , snake_case__ : Optional[int] = knapsack(A , A , A , A )
snake_case__ : set = set()
_construct_solution(A , A , A , A , A )
return optimal_val, example_optional_set
def lowercase_ (A : list , A : list , A : int , A : int , A : set ):
# for the current item i at a maximum weight j to be part of an optimal subset,
# the optimal value at (i, j) must be greater than the optimal value at (i-1, j).
# where i - 1 means considering only the previous items at the given maximum weight
if i > 0 and j > 0:
if dp[i - 1][j] == dp[i][j]:
_construct_solution(A , A , i - 1 , A , A )
else:
optimal_set.add(A )
_construct_solution(A , A , i - 1 , j - wt[i - 1] , A )
if __name__ == "__main__":
a_ :Any = [3, 2, 4, 4]
a_ :List[Any] = [4, 3, 2, 3]
a_ :Union[str, Any] = 4
a_ :List[str] = 6
a_ :Union[str, Any] = [[0] * (w + 1)] + [[0] + [-1] * (w + 1) for _ in range(n + 1)]
a_ , a_ :List[Any] = knapsack(w, wt, val, n)
print(optimal_solution)
print(mf_knapsack(n, wt, val, w)) # switched the n and w
# testing the dynamic programming problem with example
# the optimal subset for the above example are items 3 and 4
a_ , a_ :Any = knapsack_with_example_solution(w, wt, val)
assert optimal_solution == 8
assert optimal_subset == {3, 4}
print("optimal_value = ", optimal_solution)
print("An optimal subset corresponding to the optimal value", optimal_subset)
| 277 | 0 |
from dataclasses import dataclass, field
from typing import ClassVar, Dict
from ..features import Features, Value
from .base import TaskTemplate
@dataclass(frozen=lowerCAmelCase_ )
class lowercase__( lowerCAmelCase_ ):
"""simple docstring"""
a :List[Any] = field(default='language-modeling' , metadata={'include_in_asdict_even_if_is_default': True} )
a :Optional[int] = Features({'text': Value('string' )} )
a :int = Features({} )
a :Tuple = 'text'
@property
def _lowercase ( self : List[Any] ) -> Dict[str, str]:
return {self.text_column: "text"}
| 30 |
from typing import TYPE_CHECKING
from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_flax_available, is_torch_available
a_ :int = {
"configuration_longt5": ["LONGT5_PRETRAINED_CONFIG_ARCHIVE_MAP", "LongT5Config", "LongT5OnnxConfig"],
}
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :List[str] = [
"LONGT5_PRETRAINED_MODEL_ARCHIVE_LIST",
"LongT5EncoderModel",
"LongT5ForConditionalGeneration",
"LongT5Model",
"LongT5PreTrainedModel",
]
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :int = [
"FlaxLongT5ForConditionalGeneration",
"FlaxLongT5Model",
"FlaxLongT5PreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_longta import LONGT5_PRETRAINED_CONFIG_ARCHIVE_MAP, LongTaConfig, LongTaOnnxConfig
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_longta import (
LONGT5_PRETRAINED_MODEL_ARCHIVE_LIST,
LongTaEncoderModel,
LongTaForConditionalGeneration,
LongTaModel,
LongTaPreTrainedModel,
)
try:
if not is_flax_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_flax_longta import (
FlaxLongTaForConditionalGeneration,
FlaxLongTaModel,
FlaxLongTaPreTrainedModel,
)
else:
import sys
a_ :Optional[int] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)
| 277 | 0 |
"""simple docstring"""
import torch
def lowercase ( ) -> int:
if torch.cuda.is_available():
__a = torch.cuda.device_count()
else:
__a = 0
print(f'''Successfully ran on {num_gpus} GPUs''' )
if __name__ == "__main__":
main()
| 45 |
import argparse
import json
import os
import torch
from torch import nn
from transformers import NllbMoeConfig, NllbMoeModel
from transformers.modeling_utils import dtype_byte_size
from transformers.utils import WEIGHTS_INDEX_NAME, WEIGHTS_NAME
def lowercase_ (A : List[str] ):
snake_case__ : Tuple = [
'encoder.version',
'decoder.version',
'model.encoder.version',
'model.decoder.version',
'decoder.output_projection.weight',
'_float_tensor',
'encoder.embed_positions._float_tensor',
'decoder.embed_positions._float_tensor',
]
for k in ignore_keys:
state_dict.pop(A , A )
def lowercase_ (A : str ):
snake_case__ , snake_case__ : Union[str, Any] = emb.weight.shape
snake_case__ : str = nn.Linear(A , A , bias=A )
snake_case__ : str = emb.weight.data
return lin_layer
def lowercase_ (A : Optional[int] , A : Union[str, Any]=None ):
snake_case__ : Any = {}
for old_key in state_dict.keys():
snake_case__ : Tuple = old_key
if "moe_layer.experts." in key:
if expert_idx is not None:
snake_case__ : int = key.replace('moe_layer.experts.0' , F'''ffn.experts.expert_{expert_idx}''' )
else:
snake_case__ : Any = key.replace('moe_layer.experts.' , 'ffn.experts.expert_' )
if "gate" in key:
snake_case__ : Dict = key.replace('.moe_layer.gate.wg' , '.ffn.router.classifier' )
if "fc2" and "experts" not in key:
snake_case__ : str = key.replace('.fc2.' , '.ffn.fc2.' )
if "fc1" and "experts" not in key:
snake_case__ : str = key.replace('.fc1.' , '.ffn.fc1.' )
if ".encoder_attn." in key:
snake_case__ : Tuple = key.replace('.encoder_attn.' , '.cross_attention.' )
if "encoder_attn_layer_norm" in key:
snake_case__ : Tuple = key.replace('encoder_attn_layer_norm' , 'cross_attention_layer_norm' )
if "final_layer_norm" in key:
snake_case__ : Optional[int] = key.replace('final_layer_norm' , 'ff_layer_norm' )
snake_case__ : Dict = state_dict[old_key]
return new_dict
def lowercase_ (A : List[Any] , A : Tuple , A : List[Any] , A : List[str] , A : str = WEIGHTS_NAME ):
snake_case__ : Dict = []
snake_case__ : str = 0
os.makedirs(A , exist_ok=A )
for expert in range(A ):
snake_case__ : Tuple = switch_checkpoint_path + F'''-rank-{expert}.pt'''
if os.path.isfile(A ):
snake_case__ : Optional[Any] = torch.load(A )['model']
remove_ignore_keys_(A )
snake_case__ : Optional[Any] = rename_fairseq_keys(A , A )
snake_case__ : Dict = os.path.join(
A , weights_name.replace('.bin' , F'''-{len(A )+1:05d}-of-???.bin''' ) )
torch.save(A , A )
sharded_state_dicts.append(expert_state.keys() )
total_size += sum([value.numel() for key, value in expert_state.items()] ) * dtype_byte_size(
expert_state[list(A )[0]].dtype )
# Add the last block
snake_case__ : Tuple = os.path.join(A , weights_name.replace('.bin' , F'''-{len(A )+1:05d}-of-???.bin''' ) )
snake_case__ : Union[str, Any] = torch.load(switch_checkpoint_path + '-shared.pt' )['model']
remove_ignore_keys_(A )
snake_case__ : str = rename_fairseq_keys(A , A )
snake_case__ : Any = shared_weights['decoder.embed_tokens.weight']
sharded_state_dicts.append(shared_weights.keys() )
# If we only have the shared weights (dummy model/experts saved on the same file)
if len(A ) == 1:
snake_case__ : Any = os.path.join(A , A )
torch.save(A , A )
return {weights_name: sharded_state_dicts[0]}, None
else:
torch.save(A , A )
# Otherwise, let's build the index
snake_case__ : Tuple = {}
for idx, shard in enumerate(A ):
snake_case__ : Optional[int] = weights_name.replace('.bin' , F'''-{idx+1:05d}-of-{len(A ):05d}.bin''' )
snake_case__ : List[Any] = os.path.join(A , weights_name.replace('.bin' , F'''-{idx+1:05d}-of-???.bin''' ) )
os.rename(A , os.path.join(A , A ) )
for key in shard:
snake_case__ : Any = shard_file
# Add the metadata
snake_case__ : int = {'total_size': total_size}
snake_case__ : Dict = {'metadata': metadata, 'weight_map': weight_map}
with open(os.path.join(A , A ) , 'w' , encoding='utf-8' ) as f:
snake_case__ : Any = json.dumps(A , indent=2 , sort_keys=A ) + '\n'
f.write(A )
return metadata, index
if __name__ == "__main__":
a_ :int = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--nllb_moe_checkpoint_path",
default="/home/arthur_huggingface_co/fairseq/weights/checkpoints/model_moe_54b/checkpoint_2_300000",
type=str,
required=False,
help="Path to a directory containing a folder per layer. Follows the original Google format.",
)
parser.add_argument("--dtype", default="float32", type=str, required=False, help="dtype of the saved model")
parser.add_argument(
"--pytorch_dump_folder_path",
default="/home/arthur_huggingface_co/fairseq/weights/checkpoints/hf-converted-moe-54b",
type=str,
required=False,
help="Path to the output pytorch model.",
)
a_ :Optional[Any] = parser.parse_args()
a_ , a_ :Optional[Any] = shard_on_the_fly(
args.nllb_moe_checkpoint_path,
args.pytorch_dump_folder_path,
128,
args.dtype,
)
a_ :List[str] = NllbMoeConfig.from_pretrained(
"facebook/nllb-200-3.3B", encoder_sparse_step=4, decoder_sparse_step=4, num_experts=128
)
config.save_pretrained(args.pytorch_dump_folder_path)
a_ :int = NllbMoeModel.from_pretrained(args.pytorch_dump_folder_path)
print("Done")
model.save_pretrained(args.pytorch_dump_folder_path)
| 277 | 0 |
'''simple docstring'''
def __lowerCAmelCase ( snake_case__ , snake_case__ , snake_case__ ):
return not any(
neighbour == 1 and colored_vertices[i] == color
for i, neighbour in enumerate(snake_case__ ) )
def __lowerCAmelCase ( snake_case__ , snake_case__ , snake_case__ , snake_case__ ):
# Base Case
if index == len(snake_case__ ):
return True
# Recursive Step
for i in range(snake_case__ ):
if valid_coloring(graph[index] , snake_case__ , snake_case__ ):
# Color current vertex
__UpperCamelCase : Tuple = i
# Validate coloring
if util_color(snake_case__ , snake_case__ , snake_case__ , index + 1 ):
return True
# Backtrack
__UpperCamelCase : int = -1
return False
def __lowerCAmelCase ( snake_case__ , snake_case__ ):
__UpperCamelCase : Union[str, Any] = [-1] * len(snake_case__ )
if util_color(snake_case__ , snake_case__ , snake_case__ , 0 ):
return colored_vertices
return []
| 298 |
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_sentencepiece_available,
is_tokenizers_available,
is_torch_available,
)
a_ :Optional[Any] = {"configuration_reformer": ["REFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP", "ReformerConfig"]}
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :str = ["ReformerTokenizer"]
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :int = ["ReformerTokenizerFast"]
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
a_ :List[str] = [
"REFORMER_PRETRAINED_MODEL_ARCHIVE_LIST",
"ReformerAttention",
"ReformerForMaskedLM",
"ReformerForQuestionAnswering",
"ReformerForSequenceClassification",
"ReformerLayer",
"ReformerModel",
"ReformerModelWithLMHead",
"ReformerPreTrainedModel",
]
if TYPE_CHECKING:
from .configuration_reformer import REFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, ReformerConfig
try:
if not is_sentencepiece_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_reformer import ReformerTokenizer
try:
if not is_tokenizers_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .tokenization_reformer_fast import ReformerTokenizerFast
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_reformer import (
REFORMER_PRETRAINED_MODEL_ARCHIVE_LIST,
ReformerAttention,
ReformerForMaskedLM,
ReformerForQuestionAnswering,
ReformerForSequenceClassification,
ReformerLayer,
ReformerModel,
ReformerModelWithLMHead,
ReformerPreTrainedModel,
)
else:
import sys
a_ :Optional[Any] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)
| 277 | 0 |
import unittest
import numpy as np
from transformers.testing_utils import require_torch, require_vision
from transformers.utils import is_torch_available, is_vision_available
from ...test_image_processing_common import ImageProcessingSavingTestMixin, prepare_image_inputs
if is_torch_available():
import torch
if is_vision_available():
from PIL import Image
from transformers import PoolFormerImageProcessor
class lowercase ( unittest.TestCase ):
def __init__( self , _a , _a=7 , _a=3 , _a=30 , _a=400 , _a=True , _a=None , _a=0.9 , _a=None , _a=True , _a=[0.5, 0.5, 0.5] , _a=[0.5, 0.5, 0.5] , ) -> List[Any]:
_A : int = size if size is not None else {'shortest_edge': 30}
_A : Tuple = crop_size if crop_size is not None else {'height': 30, 'width': 30}
_A : Union[str, Any] = parent
_A : Dict = batch_size
_A : int = num_channels
_A : Tuple = min_resolution
_A : Any = max_resolution
_A : List[Any] = do_resize_and_center_crop
_A : str = size
_A : str = crop_pct
_A : List[str] = crop_size
_A : Optional[int] = do_normalize
_A : Tuple = image_mean
_A : Tuple = image_std
def a__ ( self ) -> int:
return {
"size": self.size,
"do_resize_and_center_crop": self.do_resize_and_center_crop,
"crop_pct": self.crop_pct,
"crop_size": self.crop_size,
"do_normalize": self.do_normalize,
"image_mean": self.image_mean,
"image_std": self.image_std,
}
@require_torch
@require_vision
class lowercase ( lowerCAmelCase_,unittest.TestCase ):
_a = PoolFormerImageProcessor if is_vision_available() else None
def a__ ( self ) -> Dict:
_A : Union[str, Any] = PoolFormerImageProcessingTester(self )
@property
def a__ ( self ) -> Dict:
return self.image_processor_tester.prepare_image_processor_dict()
def a__ ( self ) -> Optional[int]:
_A : List[str] = self.image_processing_class(**self.image_processor_dict )
self.assertTrue(hasattr(_snake_case , """do_resize_and_center_crop""" ) )
self.assertTrue(hasattr(_snake_case , """size""" ) )
self.assertTrue(hasattr(_snake_case , """crop_pct""" ) )
self.assertTrue(hasattr(_snake_case , """do_normalize""" ) )
self.assertTrue(hasattr(_snake_case , """image_mean""" ) )
self.assertTrue(hasattr(_snake_case , """image_std""" ) )
def a__ ( self ) -> List[str]:
_A : Union[str, Any] = self.image_processing_class.from_dict(self.image_processor_dict )
self.assertEqual(image_processor.size , {"""shortest_edge""": 30} )
self.assertEqual(image_processor.crop_size , {"""height""": 30, """width""": 30} )
_A : int = self.image_processing_class.from_dict(self.image_processor_dict , size=42 , crop_size=84 )
self.assertEqual(image_processor.size , {"""shortest_edge""": 42} )
self.assertEqual(image_processor.crop_size , {"""height""": 84, """width""": 84} )
def a__ ( self ) -> List[Any]:
pass
def a__ ( self ) -> str:
# Initialize image_processing
_A : Union[str, Any] = self.image_processing_class(**self.image_processor_dict )
# create random PIL images
_A : List[str] = prepare_image_inputs(self.image_processor_tester , equal_resolution=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case , Image.Image )
# Test not batched input
_A : Optional[int] = image_processing(image_inputs[0] , return_tensors="""pt""" ).pixel_values
self.assertEqual(
encoded_images.shape , (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size["""height"""],
self.image_processor_tester.crop_size["""width"""],
) , )
# Test batched
_A : str = image_processing(_snake_case , return_tensors="""pt""" ).pixel_values
self.assertEqual(
encoded_images.shape , (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size["""height"""],
self.image_processor_tester.crop_size["""width"""],
) , )
def a__ ( self ) -> List[Any]:
# Initialize image_processing
_A : Optional[int] = self.image_processing_class(**self.image_processor_dict )
# create random numpy tensors
_A : Dict = prepare_image_inputs(self.image_processor_tester , equal_resolution=_snake_case , numpify=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case , np.ndarray )
# Test not batched input
_A : Dict = image_processing(image_inputs[0] , return_tensors="""pt""" ).pixel_values
self.assertEqual(
encoded_images.shape , (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size["""height"""],
self.image_processor_tester.crop_size["""width"""],
) , )
# Test batched
_A : List[Any] = image_processing(_snake_case , return_tensors="""pt""" ).pixel_values
self.assertEqual(
encoded_images.shape , (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size["""height"""],
self.image_processor_tester.crop_size["""width"""],
) , )
def a__ ( self ) -> List[str]:
# Initialize image_processing
_A : Tuple = self.image_processing_class(**self.image_processor_dict )
# create random PyTorch tensors
_A : List[str] = prepare_image_inputs(self.image_processor_tester , equal_resolution=_snake_case , torchify=_snake_case )
for image in image_inputs:
self.assertIsInstance(_snake_case , torch.Tensor )
# Test not batched input
_A : Tuple = image_processing(image_inputs[0] , return_tensors="""pt""" ).pixel_values
self.assertEqual(
encoded_images.shape , (
1,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size["""height"""],
self.image_processor_tester.crop_size["""width"""],
) , )
# Test batched
_A : Optional[Any] = image_processing(_snake_case , return_tensors="""pt""" ).pixel_values
self.assertEqual(
encoded_images.shape , (
self.image_processor_tester.batch_size,
self.image_processor_tester.num_channels,
self.image_processor_tester.crop_size["""height"""],
self.image_processor_tester.crop_size["""width"""],
) , )
| 26 |
import itertools
import random
import unittest
import numpy as np
from transformers import BatchFeature, SpeechTaFeatureExtractor
from transformers.testing_utils import require_torch
from transformers.utils.import_utils import is_torch_available
from ...test_sequence_feature_extraction_common import SequenceFeatureExtractionTestMixin
if is_torch_available():
import torch
a_ :Any = random.Random()
def lowercase_ (A : int , A : Union[str, Any]=1.0 , A : List[str]=None , A : Any=None ):
if rng is None:
snake_case__ : List[str] = global_rng
snake_case__ : int = []
for batch_idx in range(shape[0] ):
values.append([] )
for _ in range(shape[1] ):
values[-1].append(rng.random() * scale )
return values
@require_torch
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
def __init__( self : Optional[Any], _snake_case : List[str], _snake_case : Tuple=7, _snake_case : Union[str, Any]=4_0_0, _snake_case : Any=2_0_0_0, _snake_case : Dict=1, _snake_case : Optional[Any]=0.0, _snake_case : List[Any]=1_6_0_0_0, _snake_case : List[Any]=True, _snake_case : List[Any]=8_0, _snake_case : Dict=1_6, _snake_case : str=6_4, _snake_case : Tuple="hann_window", _snake_case : Union[str, Any]=8_0, _snake_case : Optional[Any]=7_6_0_0, _snake_case : str=1e-10, _snake_case : Any=True, ) ->Union[str, Any]:
snake_case__ : Optional[int] = parent
snake_case__ : Optional[Any] = batch_size
snake_case__ : List[Any] = min_seq_length
snake_case__ : List[Any] = max_seq_length
snake_case__ : Any = (self.max_seq_length - self.min_seq_length) // (self.batch_size - 1)
snake_case__ : Tuple = feature_size
snake_case__ : List[Any] = padding_value
snake_case__ : Any = sampling_rate
snake_case__ : Dict = do_normalize
snake_case__ : Union[str, Any] = num_mel_bins
snake_case__ : Any = hop_length
snake_case__ : Any = win_length
snake_case__ : Any = win_function
snake_case__ : Optional[int] = fmin
snake_case__ : int = fmax
snake_case__ : Union[str, Any] = mel_floor
snake_case__ : Union[str, Any] = return_attention_mask
def lowercase_ ( self : Optional[int] ) ->List[str]:
return {
"feature_size": self.feature_size,
"padding_value": self.padding_value,
"sampling_rate": self.sampling_rate,
"do_normalize": self.do_normalize,
"num_mel_bins": self.num_mel_bins,
"hop_length": self.hop_length,
"win_length": self.win_length,
"win_function": self.win_function,
"fmin": self.fmin,
"fmax": self.fmax,
"mel_floor": self.mel_floor,
"return_attention_mask": self.return_attention_mask,
}
def lowercase_ ( self : Any, _snake_case : Optional[Any]=False, _snake_case : List[str]=False ) ->Union[str, Any]:
def _flatten(_snake_case : List[str] ):
return list(itertools.chain(*_snake_case ) )
if equal_length:
snake_case__ : Any = floats_list((self.batch_size, self.max_seq_length) )
else:
# make sure that inputs increase in size
snake_case__ : int = [
_flatten(floats_list((x, self.feature_size) ) )
for x in range(self.min_seq_length, self.max_seq_length, self.seq_length_diff )
]
if numpify:
snake_case__ : Any = [np.asarray(_snake_case ) for x in speech_inputs]
return speech_inputs
def lowercase_ ( self : Union[str, Any], _snake_case : str=False, _snake_case : Dict=False ) ->List[str]:
if equal_length:
snake_case__ : Optional[Any] = [floats_list((self.max_seq_length, self.num_mel_bins) ) for _ in range(self.batch_size )]
else:
# make sure that inputs increase in size
snake_case__ : List[str] = [
floats_list((x, self.num_mel_bins) )
for x in range(self.min_seq_length, self.max_seq_length, self.seq_length_diff )
]
if numpify:
snake_case__ : int = [np.asarray(_snake_case ) for x in speech_inputs]
return speech_inputs
@require_torch
class snake_case__ ( lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = SpeechTaFeatureExtractor
def lowercase_ ( self : int ) ->Union[str, Any]:
snake_case__ : List[str] = SpeechTaFeatureExtractionTester(self )
def lowercase_ ( self : Any, _snake_case : Dict ) ->Any:
self.assertTrue(np.all(np.mean(_snake_case, axis=0 ) < 1e-3 ) )
self.assertTrue(np.all(np.abs(np.var(_snake_case, axis=0 ) - 1 ) < 1e-3 ) )
def lowercase_ ( self : List[Any] ) ->Union[str, Any]:
# Tests that all call wrap to encode_plus and batch_encode_plus
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
# create three inputs of length 800, 1000, and 1200
snake_case__ : int = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Tuple = [np.asarray(_snake_case ) for speech_input in speech_inputs]
# Test not batched input
snake_case__ : str = feat_extract(speech_inputs[0], return_tensors='np' ).input_values
snake_case__ : List[str] = feat_extract(np_speech_inputs[0], return_tensors='np' ).input_values
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test batched
snake_case__ : Any = feat_extract(_snake_case, return_tensors='np' ).input_values
snake_case__ : Union[str, Any] = feat_extract(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
def lowercase_ ( self : int ) ->Optional[int]:
snake_case__ : List[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Tuple = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : int = ['longest', 'max_length', 'do_not_pad']
snake_case__ : List[str] = [None, 1_6_0_0, None]
for max_length, padding in zip(_snake_case, _snake_case ):
snake_case__ : Optional[int] = feat_extract(_snake_case, padding=_snake_case, max_length=_snake_case, return_tensors='np' )
snake_case__ : Optional[int] = processed.input_values
self._check_zero_mean_unit_variance(input_values[0][:8_0_0] )
self.assertTrue(input_values[0][8_0_0:].sum() < 1e-6 )
self._check_zero_mean_unit_variance(input_values[1][:1_0_0_0] )
self.assertTrue(input_values[0][1_0_0_0:].sum() < 1e-6 )
self._check_zero_mean_unit_variance(input_values[2][:1_2_0_0] )
def lowercase_ ( self : Union[str, Any] ) ->Optional[Any]:
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Tuple = range(8_0_0, 1_4_0_0, 2_0_0 )
snake_case__ : Optional[Any] = [floats_list((1, x) )[0] for x in lengths]
snake_case__ : Union[str, Any] = ['longest', 'max_length', 'do_not_pad']
snake_case__ : str = [None, 1_6_0_0, None]
for max_length, padding in zip(_snake_case, _snake_case ):
snake_case__ : List[str] = feat_extract(_snake_case, max_length=_snake_case, padding=_snake_case )
snake_case__ : Tuple = processed.input_values
self._check_zero_mean_unit_variance(input_values[0][:8_0_0] )
self._check_zero_mean_unit_variance(input_values[1][:1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2][:1_2_0_0] )
def lowercase_ ( self : List[Any] ) ->Optional[Any]:
snake_case__ : Any = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : str = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Optional[Any] = feat_extract(
_snake_case, truncation=_snake_case, max_length=1_0_0_0, padding='max_length', return_tensors='np' )
snake_case__ : int = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1] )
self._check_zero_mean_unit_variance(input_values[2] )
def lowercase_ ( self : int ) ->Union[str, Any]:
snake_case__ : str = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : Dict = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : str = feat_extract(
_snake_case, truncation=_snake_case, max_length=1_0_0_0, padding='longest', return_tensors='np' )
snake_case__ : Dict = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1, :1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2] )
# make sure that if max_length < longest -> then pad to max_length
self.assertTrue(input_values.shape == (3, 1_0_0_0) )
snake_case__ : Tuple = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : List[str] = feat_extract(
_snake_case, truncation=_snake_case, max_length=2_0_0_0, padding='longest', return_tensors='np' )
snake_case__ : Optional[Any] = processed.input_values
self._check_zero_mean_unit_variance(input_values[0, :8_0_0] )
self._check_zero_mean_unit_variance(input_values[1, :1_0_0_0] )
self._check_zero_mean_unit_variance(input_values[2] )
# make sure that if max_length > longest -> then pad to longest
self.assertTrue(input_values.shape == (3, 1_2_0_0) )
def lowercase_ ( self : List[str] ) ->Dict:
snake_case__ : Dict = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
snake_case__ : List[Any] = np.random.rand(1_0_0 ).astype(np.floataa )
snake_case__ : int = np_speech_inputs.tolist()
for inputs in [py_speech_inputs, np_speech_inputs]:
snake_case__ : int = feature_extractor.pad([{'input_values': inputs}], return_tensors='np' )
self.assertTrue(np_processed.input_values.dtype == np.floataa )
snake_case__ : Optional[int] = feature_extractor.pad([{'input_values': inputs}], return_tensors='pt' )
self.assertTrue(pt_processed.input_values.dtype == torch.floataa )
def lowercase_ ( self : Optional[int] ) ->Optional[Any]:
# Tests that all call wrap to encode_plus and batch_encode_plus
snake_case__ : str = self.feature_extraction_class(**self.feat_extract_tester.prepare_feat_extract_dict() )
# create three inputs of length 800, 1000, and 1200
snake_case__ : List[Any] = [floats_list((1, x) )[0] for x in range(8_0_0, 1_4_0_0, 2_0_0 )]
snake_case__ : Dict = [np.asarray(_snake_case ) for speech_input in speech_inputs]
# Test feature size
snake_case__ : Optional[int] = feature_extractor(audio_target=_snake_case, padding=_snake_case, return_tensors='np' ).input_values
self.assertTrue(input_values.ndim == 3 )
self.assertTrue(input_values.shape[-1] == feature_extractor.num_mel_bins )
# Test not batched input
snake_case__ : Dict = feature_extractor(speech_inputs[0], return_tensors='np' ).input_values
snake_case__ : Any = feature_extractor(np_speech_inputs[0], return_tensors='np' ).input_values
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test batched
snake_case__ : Dict = feature_extractor(_snake_case, return_tensors='np' ).input_values
snake_case__ : Dict = feature_extractor(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
# Test 2-D numpy arrays are batched.
snake_case__ : Optional[Any] = [floats_list((1, x) )[0] for x in (8_0_0, 8_0_0, 8_0_0)]
snake_case__ : int = np.asarray(_snake_case )
snake_case__ : Union[str, Any] = feature_extractor(_snake_case, return_tensors='np' ).input_values
snake_case__ : Union[str, Any] = feature_extractor(_snake_case, return_tensors='np' ).input_values
for enc_seq_a, enc_seq_a in zip(_snake_case, _snake_case ):
self.assertTrue(np.allclose(_snake_case, _snake_case, atol=1e-3 ) )
def lowercase_ ( self : Union[str, Any] ) ->str:
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : List[str] = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : Tuple = BatchFeature({input_name: speech_inputs} )
self.assertTrue(all(len(_snake_case ) == len(_snake_case ) for x, y in zip(_snake_case, processed_features[input_name] ) ) )
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target(equal_length=_snake_case )
snake_case__ : Union[str, Any] = BatchFeature({input_name: speech_inputs}, tensor_type='np' )
snake_case__ : Dict = processed_features[input_name]
if len(batch_features_input.shape ) < 3:
snake_case__ : List[str] = batch_features_input[:, :, None]
self.assertTrue(
batch_features_input.shape
== (self.feat_extract_tester.batch_size, len(speech_inputs[0] ), self.feat_extract_tester.num_mel_bins) )
@require_torch
def lowercase_ ( self : List[str] ) ->Any:
snake_case__ : int = self.feat_extract_tester.prepare_inputs_for_target(equal_length=_snake_case )
snake_case__ : Optional[Any] = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Tuple = feat_extract.model_input_names[0]
snake_case__ : List[Any] = BatchFeature({input_name: speech_inputs}, tensor_type='pt' )
snake_case__ : Tuple = processed_features[input_name]
if len(batch_features_input.shape ) < 3:
snake_case__ : Any = batch_features_input[:, :, None]
self.assertTrue(
batch_features_input.shape
== (self.feat_extract_tester.batch_size, len(speech_inputs[0] ), self.feat_extract_tester.num_mel_bins) )
@require_torch
def lowercase_ ( self : Optional[int] ) ->Tuple:
snake_case__ : Dict = self.feature_extraction_class(**self.feat_extract_dict )
snake_case__ : Union[str, Any] = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : List[str] = BatchFeature({input_name: speech_inputs} )
snake_case__ : int = feat_extract.num_mel_bins # hack!
snake_case__ : Tuple = feat_extract.pad(_snake_case, padding='longest', return_tensors='np' )[input_name]
snake_case__ : Union[str, Any] = feat_extract.pad(_snake_case, padding='longest', return_tensors='pt' )[input_name]
self.assertTrue(abs(input_np.astype(np.floataa ).sum() - input_pt.numpy().astype(np.floataa ).sum() ) < 1e-2 )
def lowercase_ ( self : int ) ->Any:
snake_case__ : Any = self.feat_extract_dict
snake_case__ : List[Any] = True
snake_case__ : Union[str, Any] = self.feature_extraction_class(**_snake_case )
snake_case__ : Any = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : List[Any] = [len(_snake_case ) for x in speech_inputs]
snake_case__ : Union[str, Any] = feat_extract.model_input_names[0]
snake_case__ : Optional[int] = BatchFeature({input_name: speech_inputs} )
snake_case__ : List[str] = feat_extract.num_mel_bins # hack!
snake_case__ : str = feat_extract.pad(_snake_case, padding='longest', return_tensors='np' )
self.assertIn('attention_mask', _snake_case )
self.assertListEqual(list(processed.attention_mask.shape ), list(processed[input_name].shape[:2] ) )
self.assertListEqual(processed.attention_mask.sum(-1 ).tolist(), _snake_case )
def lowercase_ ( self : Optional[int] ) ->str:
snake_case__ : int = self.feat_extract_dict
snake_case__ : List[str] = True
snake_case__ : Tuple = self.feature_extraction_class(**_snake_case )
snake_case__ : List[str] = self.feat_extract_tester.prepare_inputs_for_target()
snake_case__ : str = [len(_snake_case ) for x in speech_inputs]
snake_case__ : Optional[Any] = feat_extract.model_input_names[0]
snake_case__ : Optional[int] = BatchFeature({input_name: speech_inputs} )
snake_case__ : Optional[Any] = min(_snake_case )
snake_case__ : Union[str, Any] = feat_extract.num_mel_bins # hack!
snake_case__ : Tuple = feat_extract.pad(
_snake_case, padding='max_length', max_length=_snake_case, truncation=_snake_case, return_tensors='np' )
self.assertIn('attention_mask', _snake_case )
self.assertListEqual(
list(processed_pad.attention_mask.shape ), [processed_pad[input_name].shape[0], max_length] )
self.assertListEqual(
processed_pad.attention_mask[:, :max_length].sum(-1 ).tolist(), [max_length for x in speech_inputs] )
def lowercase_ ( self : List[Any], _snake_case : Optional[int] ) ->Optional[Any]:
from datasets import load_dataset
snake_case__ : str = load_dataset('hf-internal-testing/librispeech_asr_dummy', 'clean', split='validation' )
# automatic decoding with librispeech
snake_case__ : Dict = ds.sort('id' ).select(range(_snake_case ) )[:num_samples]['audio']
return [x["array"] for x in speech_samples]
def lowercase_ ( self : str ) ->str:
# fmt: off
snake_case__ : List[Any] = torch.tensor(
[2.3804e-03, 2.0752e-03, 1.9836e-03, 2.1057e-03, 1.6174e-03,
3.0518e-04, 9.1553e-05, 3.3569e-04, 9.7656e-04, 1.8311e-03,
2.0142e-03, 2.1057e-03, 1.7395e-03, 4.5776e-04, -3.9673e-04,
4.5776e-04, 1.0071e-03, 9.1553e-05, 4.8828e-04, 1.1597e-03,
7.3242e-04, 9.4604e-04, 1.8005e-03, 1.8311e-03, 8.8501e-04,
4.2725e-04, 4.8828e-04, 7.3242e-04, 1.0986e-03, 2.1057e-03] )
# fmt: on
snake_case__ : Union[str, Any] = self._load_datasamples(1 )
snake_case__ : Optional[int] = SpeechTaFeatureExtractor()
snake_case__ : List[Any] = feature_extractor(_snake_case, return_tensors='pt' ).input_values
self.assertEquals(input_values.shape, (1, 9_3_6_8_0) )
self.assertTrue(torch.allclose(input_values[0, :3_0], _snake_case, atol=1e-6 ) )
def lowercase_ ( self : Any ) ->str:
# fmt: off
snake_case__ : Optional[Any] = torch.tensor(
[-2.6_8_7_0, -3.0_1_0_4, -3.1_3_5_6, -3.5_3_5_2, -3.0_0_4_4, -3.0_3_5_3, -3.4_7_1_9, -3.6_7_7_7,
-3.1_5_2_0, -2.9_4_3_5, -2.6_5_5_3, -2.8_7_9_5, -2.9_9_4_4, -2.5_9_2_1, -3.0_2_7_9, -3.0_3_8_6,
-3.0_8_6_4, -3.1_2_9_1, -3.2_3_5_3, -2.7_4_4_4, -2.6_8_3_1, -2.7_2_8_7, -3.1_7_6_1, -3.1_5_7_1,
-3.2_7_2_6, -3.0_5_8_2, -3.1_0_0_7, -3.4_5_3_3, -3.4_6_9_5, -3.0_9_9_8] )
# fmt: on
snake_case__ : List[str] = self._load_datasamples(1 )
snake_case__ : str = SpeechTaFeatureExtractor()
snake_case__ : Optional[Any] = feature_extractor(audio_target=_snake_case, return_tensors='pt' ).input_values
self.assertEquals(input_values.shape, (1, 3_6_6, 8_0) )
self.assertTrue(torch.allclose(input_values[0, 0, :3_0], _snake_case, atol=1e-4 ) )
| 277 | 0 |
import unittest
from transformers import AutoTokenizer, is_flax_available
from transformers.testing_utils import require_flax, require_sentencepiece, require_tokenizers, slow
if is_flax_available():
import jax.numpy as jnp
from transformers import FlaxXLMRobertaModel
@require_sentencepiece
@require_tokenizers
@require_flax
class __lowerCAmelCase ( unittest.TestCase ):
@slow
def _lowerCamelCase ( self : Optional[Any]) -> Optional[Any]:
"""simple docstring"""
_UpperCAmelCase = FlaxXLMRobertaModel.from_pretrained('xlm-roberta-base')
_UpperCAmelCase = AutoTokenizer.from_pretrained('xlm-roberta-base')
_UpperCAmelCase = 'The dog is cute and lives in the garden house'
_UpperCAmelCase = jnp.array([tokenizer.encode(_snake_case)])
_UpperCAmelCase = (1, 12, 7_68) # batch_size, sequence_length, embedding_vector_dim
_UpperCAmelCase = jnp.array(
[[-0.0_1_0_1, 0.1_2_1_8, -0.0_8_0_3, 0.0_8_0_1, 0.1_3_2_7, 0.0_7_7_6, -0.1_2_1_5, 0.2_3_8_3, 0.3_3_3_8, 0.3_1_0_6, 0.0_3_0_0, 0.0_2_5_2]])
_UpperCAmelCase = model(_snake_case)['last_hidden_state']
self.assertEqual(output.shape , _snake_case)
# compare the actual values for a slice of last dim
self.assertTrue(jnp.allclose(output[:, :, -1] , _snake_case , atol=1E-3))
| 339 |
# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from ..models.auto import AutoModelForSeqaSeqLM, AutoTokenizer
from .base import PipelineTool
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """philschmid/bart-large-cnn-samsum"""
_SCREAMING_SNAKE_CASE = (
"""This is a tool that summarizes an English text. It takes an input `text` containing the text to summarize, """
"""and returns a summary of the text."""
)
_SCREAMING_SNAKE_CASE = """summarizer"""
_SCREAMING_SNAKE_CASE = AutoTokenizer
_SCREAMING_SNAKE_CASE = AutoModelForSeqaSeqLM
_SCREAMING_SNAKE_CASE = ["""text"""]
_SCREAMING_SNAKE_CASE = ["""text"""]
def lowercase_ ( self : Optional[Any], _snake_case : str ) ->Any:
return self.pre_processor(_snake_case, return_tensors='pt', truncation=_snake_case )
def lowercase_ ( self : int, _snake_case : List[Any] ) ->Any:
return self.model.generate(**_snake_case )[0]
def lowercase_ ( self : int, _snake_case : int ) ->str:
return self.pre_processor.decode(_snake_case, skip_special_tokens=_snake_case, clean_up_tokenization_spaces=_snake_case )
| 277 | 0 |
from __future__ import annotations
def __UpperCAmelCase ( a_ , a_):
snake_case_ = []
snake_case_ = []
snake_case_ = 0
snake_case_ = sum(a_)
create_state_space_tree(a_ , a_ , a_ , a_ , a_ , a_)
return result
def __UpperCAmelCase ( a_ , a_ , a_ , a_ , a_ , a_ , ):
if sum(a_) > max_sum or (remaining_nums_sum + sum(a_)) < max_sum:
return
if sum(a_) == max_sum:
result.append(a_)
return
for index in range(a_ , len(a_)):
create_state_space_tree(
a_ , a_ , index + 1 , [*path, nums[index]] , a_ , remaining_nums_sum - nums[index] , )
lowercase = [3, 34, 4, 12, 5, 2]
lowercase = 9
lowercase = generate_sum_of_subsets_soln(nums, max_sum)
print(*result)
| 178 |
import argparse
import torch
from transformers import LxmertConfig, LxmertForPreTraining, load_tf_weights_in_lxmert
from transformers.utils import logging
logging.set_verbosity_info()
def lowercase_ (A : str , A : List[Any] , A : Any ):
# Initialise PyTorch model
snake_case__ : List[Any] = LxmertConfig.from_json_file(A )
print(F'''Building PyTorch model from configuration: {config}''' )
snake_case__ : List[str] = LxmertForPreTraining(A )
# Load weights from tf checkpoint
load_tf_weights_in_lxmert(A , A , A )
# Save pytorch-model
print(F'''Save PyTorch model to {pytorch_dump_path}''' )
torch.save(model.state_dict() , A )
if __name__ == "__main__":
a_ :Union[str, Any] = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--tf_checkpoint_path", default=None, type=str, required=True, help="Path to the TensorFlow checkpoint path."
)
parser.add_argument(
"--config_file",
default=None,
type=str,
required=True,
help="The config json file corresponding to the pre-trained model. \nThis specifies the model architecture.",
)
parser.add_argument(
"--pytorch_dump_path", default=None, type=str, required=True, help="Path to the output PyTorch model."
)
a_ :Optional[int] = parser.parse_args()
convert_tf_checkpoint_to_pytorch(args.tf_checkpoint_path, args.config_file, args.pytorch_dump_path)
| 277 | 0 |
"""simple docstring"""
import argparse
import torch
from transformers import GPTaConfig, GPTaModel, load_tf_weights_in_gpta
from transformers.utils import CONFIG_NAME, WEIGHTS_NAME, logging
logging.set_verbosity_info()
def lowercase (SCREAMING_SNAKE_CASE_ : List[str] , SCREAMING_SNAKE_CASE_ : Optional[Any] , SCREAMING_SNAKE_CASE_ : Dict ) -> int:
# Construct model
if gpta_config_file == "":
SCREAMING_SNAKE_CASE = GPTaConfig()
else:
SCREAMING_SNAKE_CASE = GPTaConfig.from_json_file(SCREAMING_SNAKE_CASE_ )
SCREAMING_SNAKE_CASE = GPTaModel(SCREAMING_SNAKE_CASE_ )
# Load weights from numpy
load_tf_weights_in_gpta(SCREAMING_SNAKE_CASE_ , SCREAMING_SNAKE_CASE_ , SCREAMING_SNAKE_CASE_ )
# Save pytorch-model
SCREAMING_SNAKE_CASE = pytorch_dump_folder_path + '/' + WEIGHTS_NAME
SCREAMING_SNAKE_CASE = pytorch_dump_folder_path + '/' + CONFIG_NAME
print(F'Save PyTorch model to {pytorch_weights_dump_path}' )
torch.save(model.state_dict() , SCREAMING_SNAKE_CASE_ )
print(F'Save configuration file to {pytorch_config_dump_path}' )
with open(SCREAMING_SNAKE_CASE_ , 'w' , encoding='utf-8' ) as f:
f.write(config.to_json_string() )
if __name__ == "__main__":
__UpperCamelCase = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
'''--gpt2_checkpoint_path''', default=None, type=str, required=True, help='''Path to the TensorFlow checkpoint path.'''
)
parser.add_argument(
'''--pytorch_dump_folder_path''', default=None, type=str, required=True, help='''Path to the output PyTorch model.'''
)
parser.add_argument(
'''--gpt2_config_file''',
default='''''',
type=str,
help=(
'''An optional config json file corresponding to the pre-trained OpenAI model. \n'''
'''This specifies the model architecture.'''
),
)
__UpperCamelCase = parser.parse_args()
convert_gpta_checkpoint_to_pytorch(args.gpta_checkpoint_path, args.gpta_config_file, args.pytorch_dump_folder_path)
| 113 |
import argparse
import fairseq
import torch
from torch import nn
from transformers import (
MBartaaTokenizer,
MBartConfig,
MBartForCausalLM,
SpeechEncoderDecoderConfig,
SpeechEncoderDecoderModel,
WavaVecaConfig,
WavaVecaFeatureExtractor,
WavaVecaModel,
logging,
)
logging.set_verbosity_info()
a_ :Tuple = logging.get_logger(__name__)
a_ :List[Any] = {
"post_extract_proj": "feature_projection.projection",
"encoder.pos_conv.0": "encoder.pos_conv_embed.conv",
"self_attn.k_proj": "encoder.layers.*.attention.k_proj",
"self_attn.v_proj": "encoder.layers.*.attention.v_proj",
"self_attn.q_proj": "encoder.layers.*.attention.q_proj",
"self_attn.out_proj": "encoder.layers.*.attention.out_proj",
"self_attn_layer_norm": "encoder.layers.*.layer_norm",
"fc1": "encoder.layers.*.feed_forward.intermediate_dense",
"fc2": "encoder.layers.*.feed_forward.output_dense",
"final_layer_norm": "encoder.layers.*.final_layer_norm",
"encoder.layer_norm": "encoder.layer_norm",
"w2v_model.layer_norm": "feature_projection.layer_norm",
"quantizer.weight_proj": "quantizer.weight_proj",
"quantizer.vars": "quantizer.codevectors",
"project_q": "project_q",
"final_proj": "project_hid",
"w2v_encoder.proj": "lm_head",
"mask_emb": "masked_spec_embed",
}
a_ :Optional[int] = [
"lm_head",
"quantizer.weight_proj",
"quantizer.codevectors",
"project_q",
"project_hid",
]
def lowercase_ (A : Union[str, Any] , A : str , A : Dict , A : Optional[Any] , A : Optional[Any] ):
for attribute in key.split('.' ):
snake_case__ : Any = getattr(A , A )
if weight_type is not None:
snake_case__ : Optional[Any] = getattr(A , A ).shape
else:
snake_case__ : Optional[int] = hf_pointer.shape
assert hf_shape == value.shape, (
F'''Shape of hf {key + "." + weight_type if weight_type is not None else ""} is {hf_shape}, but should be'''
F''' {value.shape} for {full_name}'''
)
if weight_type == "weight":
snake_case__ : Tuple = value
elif weight_type == "weight_g":
snake_case__ : Tuple = value
elif weight_type == "weight_v":
snake_case__ : List[Any] = value
elif weight_type == "bias":
snake_case__ : List[Any] = value
else:
snake_case__ : Optional[Any] = value
logger.info(F'''{key + "." + weight_type if weight_type is not None else ""} was initialized from {full_name}.''' )
def lowercase_ (A : str , A : Any ):
snake_case__ : Union[str, Any] = []
snake_case__ : Union[str, Any] = fairseq_model.state_dict()
snake_case__ : Union[str, Any] = hf_model.feature_extractor
snake_case__ : Any = hf_model.adapter
for name, value in fairseq_dict.items():
snake_case__ : Any = False
if "conv_layers" in name:
load_conv_layer(
A , A , A , A , hf_model.config.feat_extract_norm == 'group' , )
snake_case__ : List[Any] = True
elif any(x in name for x in ['adaptor', 'w2v_encoder.proj.', 'w2v_proj_ln.'] ):
load_adapter(A , A , A , A )
snake_case__ : Optional[Any] = True
else:
for key, mapped_key in MAPPING.items():
if key in name or key.split('w2v_model.' )[-1] == name.split('.' )[0]:
snake_case__ : Tuple = True
if "*" in mapped_key:
snake_case__ : List[Any] = name.split(A )[0].split('.' )[-2]
snake_case__ : Optional[int] = mapped_key.replace('*' , A )
if "weight_g" in name:
snake_case__ : Optional[int] = 'weight_g'
elif "weight_v" in name:
snake_case__ : Optional[Any] = 'weight_v'
elif "bias" in name:
snake_case__ : Union[str, Any] = 'bias'
elif "weight" in name:
snake_case__ : Optional[int] = 'weight'
else:
snake_case__ : Tuple = None
set_recursively(A , A , A , A , A )
continue
if not is_used:
unused_weights.append(A )
logger.warning(F'''Unused weights: {unused_weights}''' )
def lowercase_ (A : Union[str, Any] , A : Any , A : str , A : str , A : int ):
snake_case__ : str = full_name.split('conv_layers.' )[-1]
snake_case__ : Optional[int] = name.split('.' )
snake_case__ : Tuple = int(items[0] )
snake_case__ : Any = int(items[1] )
if type_id == 0:
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.bias.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.bias.data.shape} was found.'''
)
snake_case__ : Union[str, Any] = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].conv.weight.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor.conv_layers[layer_id].conv.weight.data.shape} was found.'''
)
snake_case__ : Union[str, Any] = value
logger.info(F'''Feat extract conv layer {layer_id} was initialized from {full_name}.''' )
elif (type_id == 2 and not use_group_norm) or (type_id == 2 and layer_id == 0 and use_group_norm):
if "bias" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.bias.data.shape, (
F'''{full_name} has size {value.shape}, but {feature_extractor[layer_id].layer_norm.bias.data.shape} was'''
" found."
)
snake_case__ : Optional[int] = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
elif "weight" in name:
assert value.shape == feature_extractor.conv_layers[layer_id].layer_norm.weight.data.shape, (
F'''{full_name} has size {value.shape}, but'''
F''' {feature_extractor[layer_id].layer_norm.weight.data.shape} was found.'''
)
snake_case__ : Optional[Any] = value
logger.info(F'''Feat extract layer norm weight of layer {layer_id} was initialized from {full_name}.''' )
else:
unused_weights.append(A )
def lowercase_ (A : Optional[Any] , A : Any , A : Tuple , A : Any ):
snake_case__ : List[str] = full_name.split('adaptor.' )[-1]
snake_case__ : Tuple = name.split('.' )
if items[1].isdigit():
snake_case__ : Optional[int] = int(items[1] )
else:
snake_case__ : Any = None
if "adaptor" not in full_name:
if "proj_ln" in full_name:
# has to be layer norm
if "bias" in name:
assert (
value.shape == adapter.proj_layer_norm.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj_layer_norm.bias.data.shape} was found.'''
snake_case__ : List[Any] = value
logger.info(F'''Adapter proj layer norm bias was initialized from {full_name}.''' )
if "weight" in name:
assert (
value.shape == adapter.proj_layer_norm.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj_layer_norm.weight.data.shape} was found.'''
snake_case__ : int = value
else:
# has to be projection layer
if "bias" in name:
assert (
value.shape == adapter.proj.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj.bias.data.shape} was found.'''
snake_case__ : str = value
logger.info(F'''Adapter proj layer bias was initialized from {full_name}.''' )
if "weight" in name:
assert (
value.shape == adapter.proj.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.proj.weight.data.shape} was found.'''
snake_case__ : Dict = value
logger.info(F'''Adapter proj layer weight was initialized from {full_name}.''' )
elif isinstance(A , A ):
if "bias" in name:
assert (
value.shape == adapter.layers[layer_id].conv.bias.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.layers[layer_id].conv.bias.data.shape} was found.'''
snake_case__ : List[str] = value
logger.info(F'''Adapter layer {layer_id} bias was initialized from {full_name}.''' )
elif "weight" in name:
assert (
value.shape == adapter.layers[layer_id].conv.weight.data.shape
), F'''{full_name} has size {value.shape}, but {adapter.layers[layer_id].conv.weight.data.shape} was found.'''
snake_case__ : List[str] = value
logger.info(F'''Adapter layer {layer_id} bias was initialized from {full_name}.''' )
else:
unused_weights.append(A )
def lowercase_ (A : int ):
snake_case__ , snake_case__ : Union[str, Any] = emb.weight.shape
snake_case__ : int = nn.Linear(A , A , bias=A )
snake_case__ : Optional[Any] = emb.weight.data
return lin_layer
@torch.no_grad()
def lowercase_ (A : Tuple , A : Tuple , A : Any , A : Optional[Any] , A : int , A : Optional[Any] , A : Union[str, Any] , A : Union[str, Any] , A : Optional[Any] , A : List[Any] , A : Union[str, Any] , ):
snake_case__ : Optional[Any] = WavaVecaConfig.from_pretrained(
A , add_adapter=A , adapter_stride=A , adapter_kernel_size=A , use_auth_token=A , output_hidden_size=A , )
snake_case__ : Dict = MBartConfig.from_pretrained(A )
# load model
snake_case__ , snake_case__ , snake_case__ : Any = fairseq.checkpoint_utils.load_model_ensemble_and_task(
[checkpoint_path] , arg_overrides={
'config_yaml': config_yaml_path,
'data': '/'.join(dict_path.split('/' )[:-1] ),
'w2v_path': checkpoint_path,
'load_pretrained_decoder_from': None,
} , )
snake_case__ : List[Any] = model[0].eval()
# load feature extractor
snake_case__ : str = WavaVecaFeatureExtractor.from_pretrained(A , use_auth_token=A )
# set weights for wav2vec2 encoder
snake_case__ : List[str] = WavaVecaModel(A )
recursively_load_weights_wavaveca(model.encoder , A )
# load decoder weights
snake_case__ : Any = MBartForCausalLM(A )
snake_case__ , snake_case__ : int = hf_decoder.model.decoder.load_state_dict(model.decoder.state_dict() , strict=A )
logger.warning(F'''The following keys are missing when loading the decoder weights: {missing_keys}''' )
logger.warning(F'''The following keys are unexpected when loading the decoder weights: {unexpected_keys}''' )
snake_case__ : Union[str, Any] = SpeechEncoderDecoderModel(encoder=A , decoder=A )
snake_case__ : str = False
snake_case__ : int = MBartaaTokenizer(A )
tokenizer.save_pretrained(A )
snake_case__ : Any = hf_wavavec.config.to_dict()
snake_case__ : Tuple = tokenizer.pad_token_id
snake_case__ : Union[str, Any] = tokenizer.bos_token_id
snake_case__ : Dict = tokenizer.eos_token_id
snake_case__ : Optional[int] = 'mbart50'
snake_case__ : Union[str, Any] = 'wav2vec2'
snake_case__ : List[str] = tokenizer.eos_token_id
snake_case__ : Union[str, Any] = 2_5_0_0_0_4
snake_case__ : int = tokenizer.eos_token_id
snake_case__ : Union[str, Any] = SpeechEncoderDecoderConfig.from_dict(A )
hf_wavavec.save_pretrained(A )
feature_extractor.save_pretrained(A )
if __name__ == "__main__":
a_ :str = argparse.ArgumentParser()
parser.add_argument("--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model.")
parser.add_argument("--checkpoint_path", default=None, type=str, help="Path to fairseq checkpoint")
parser.add_argument("--dict_path", default=None, type=str, help="Path to dict of fine-tuned model")
parser.add_argument("--config_yaml_path", default=None, type=str, help="Path to yaml file of fine-tuned model")
parser.add_argument(
"--encoder_config_path",
default="facebook/wav2vec2-xls-r-1b",
type=str,
help="Path to hf encoder wav2vec2 checkpoint config",
)
parser.add_argument(
"--decoder_config_path",
default="facebook/mbart-large-50-one-to-many-mmt",
type=str,
help="Path to hf decoder checkpoint config",
)
parser.add_argument("--add_adapter", default=True, type=bool, help="whethere to add model adapter layers")
parser.add_argument("--adapter_stride", default=2, type=int, help="stride of adapter layers")
parser.add_argument("--adapter_kernel_size", default=3, type=int, help="kernel size of adapter layers")
parser.add_argument("--encoder_output_dim", default=1_024, type=int, help="encoder output dim")
parser.add_argument("--start_token_id", default=250_004, type=int, help="`decoder_start_token_id` of model config")
a_ :Union[str, Any] = parser.parse_args()
convert_wavaveca_checkpoint(
args.checkpoint_path,
args.pytorch_dump_folder_path,
args.dict_path,
args.config_yaml_path,
encoder_config_path=args.encoder_config_path,
decoder_config_path=args.decoder_config_path,
add_adapter=args.add_adapter,
adapter_kernel_size=args.adapter_kernel_size,
adapter_stride=args.adapter_stride,
decoder_start_token_id=args.start_token_id,
encoder_output_dim=args.encoder_output_dim,
)
| 277 | 0 |
'''simple docstring'''
import json
import os
import torch
from diffusers import UNetaDModel
os.makedirs('''hub/hopper-medium-v2/unet/hor32''', exist_ok=True)
os.makedirs('''hub/hopper-medium-v2/unet/hor128''', exist_ok=True)
os.makedirs('''hub/hopper-medium-v2/value_function''', exist_ok=True)
def UpperCamelCase_ ( snake_case_ : int ) -> Optional[Any]:
'''simple docstring'''
if hor == 1_28:
__lowerCAmelCase = ('DownResnetBlock1D', 'DownResnetBlock1D', 'DownResnetBlock1D')
__lowerCAmelCase = (32, 1_28, 2_56)
__lowerCAmelCase = ('UpResnetBlock1D', 'UpResnetBlock1D')
elif hor == 32:
__lowerCAmelCase = ('DownResnetBlock1D', 'DownResnetBlock1D', 'DownResnetBlock1D', 'DownResnetBlock1D')
__lowerCAmelCase = (32, 64, 1_28, 2_56)
__lowerCAmelCase = ('UpResnetBlock1D', 'UpResnetBlock1D', 'UpResnetBlock1D')
__lowerCAmelCase = torch.load(f"""/Users/bglickenhaus/Documents/diffuser/temporal_unet-hopper-mediumv2-hor{hor}.torch""" )
__lowerCAmelCase = model.state_dict()
__lowerCAmelCase = {
'down_block_types': down_block_types,
'block_out_channels': block_out_channels,
'up_block_types': up_block_types,
'layers_per_block': 1,
'use_timestep_embedding': True,
'out_block_type': 'OutConv1DBlock',
'norm_num_groups': 8,
'downsample_each_block': False,
'in_channels': 14,
'out_channels': 14,
'extra_in_channels': 0,
'time_embedding_type': 'positional',
'flip_sin_to_cos': False,
'freq_shift': 1,
'sample_size': 6_55_36,
'mid_block_type': 'MidResTemporalBlock1D',
'act_fn': 'mish',
}
__lowerCAmelCase = UNetaDModel(**snake_case_ )
print(f"""length of state dict: {len(state_dict.keys() )}""" )
print(f"""length of value function dict: {len(hf_value_function.state_dict().keys() )}""" )
__lowerCAmelCase = dict(zip(model.state_dict().keys() , hf_value_function.state_dict().keys() ) )
for k, v in mapping.items():
__lowerCAmelCase = state_dict.pop(snake_case_ )
hf_value_function.load_state_dict(snake_case_ )
torch.save(hf_value_function.state_dict() , f"""hub/hopper-medium-v2/unet/hor{hor}/diffusion_pytorch_model.bin""" )
with open(f"""hub/hopper-medium-v2/unet/hor{hor}/config.json""" , """w""" ) as f:
json.dump(snake_case_ , snake_case_ )
def UpperCamelCase_ ( ) -> Tuple:
'''simple docstring'''
__lowerCAmelCase = {
'in_channels': 14,
'down_block_types': ('DownResnetBlock1D', 'DownResnetBlock1D', 'DownResnetBlock1D', 'DownResnetBlock1D'),
'up_block_types': (),
'out_block_type': 'ValueFunction',
'mid_block_type': 'ValueFunctionMidBlock1D',
'block_out_channels': (32, 64, 1_28, 2_56),
'layers_per_block': 1,
'downsample_each_block': True,
'sample_size': 6_55_36,
'out_channels': 14,
'extra_in_channels': 0,
'time_embedding_type': 'positional',
'use_timestep_embedding': True,
'flip_sin_to_cos': False,
'freq_shift': 1,
'norm_num_groups': 8,
'act_fn': 'mish',
}
__lowerCAmelCase = torch.load("""/Users/bglickenhaus/Documents/diffuser/value_function-hopper-mediumv2-hor32.torch""" )
__lowerCAmelCase = model
__lowerCAmelCase = UNetaDModel(**snake_case_ )
print(f"""length of state dict: {len(state_dict.keys() )}""" )
print(f"""length of value function dict: {len(hf_value_function.state_dict().keys() )}""" )
__lowerCAmelCase = dict(zip(state_dict.keys() , hf_value_function.state_dict().keys() ) )
for k, v in mapping.items():
__lowerCAmelCase = state_dict.pop(snake_case_ )
hf_value_function.load_state_dict(snake_case_ )
torch.save(hf_value_function.state_dict() , """hub/hopper-medium-v2/value_function/diffusion_pytorch_model.bin""" )
with open("""hub/hopper-medium-v2/value_function/config.json""" , """w""" ) as f:
json.dump(snake_case_ , snake_case_ )
if __name__ == "__main__":
unet(32)
# unet(128)
value_function()
| 229 |
from collections import OrderedDict
from typing import TYPE_CHECKING, Any, Mapping, Optional, Union
from ...configuration_utils import PretrainedConfig
from ...onnx import OnnxConfig
from ...utils import logging
if TYPE_CHECKING:
from ... import FeatureExtractionMixin, PreTrainedTokenizerBase, TensorType
a_ :Tuple = logging.get_logger(__name__)
a_ :Union[str, Any] = {
"microsoft/deberta-v2-xlarge": "https://huggingface.co/microsoft/deberta-v2-xlarge/resolve/main/config.json",
"microsoft/deberta-v2-xxlarge": "https://huggingface.co/microsoft/deberta-v2-xxlarge/resolve/main/config.json",
"microsoft/deberta-v2-xlarge-mnli": (
"https://huggingface.co/microsoft/deberta-v2-xlarge-mnli/resolve/main/config.json"
),
"microsoft/deberta-v2-xxlarge-mnli": (
"https://huggingface.co/microsoft/deberta-v2-xxlarge-mnli/resolve/main/config.json"
),
}
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """deberta-v2"""
def __init__( self : Union[str, Any], _snake_case : Dict=1_2_8_1_0_0, _snake_case : Any=1_5_3_6, _snake_case : Tuple=2_4, _snake_case : int=2_4, _snake_case : Optional[int]=6_1_4_4, _snake_case : Optional[int]="gelu", _snake_case : Optional[int]=0.1, _snake_case : List[str]=0.1, _snake_case : str=5_1_2, _snake_case : Optional[int]=0, _snake_case : Optional[int]=0.0_2, _snake_case : Dict=1e-7, _snake_case : int=False, _snake_case : Any=-1, _snake_case : List[str]=0, _snake_case : Tuple=True, _snake_case : Any=None, _snake_case : Union[str, Any]=0, _snake_case : Tuple="gelu", **_snake_case : Union[str, Any], ) ->Optional[int]:
super().__init__(**_snake_case )
snake_case__ : Dict = hidden_size
snake_case__ : Optional[int] = num_hidden_layers
snake_case__ : Any = num_attention_heads
snake_case__ : List[Any] = intermediate_size
snake_case__ : List[Any] = hidden_act
snake_case__ : Union[str, Any] = hidden_dropout_prob
snake_case__ : Dict = attention_probs_dropout_prob
snake_case__ : List[str] = max_position_embeddings
snake_case__ : List[str] = type_vocab_size
snake_case__ : Optional[Any] = initializer_range
snake_case__ : Optional[int] = relative_attention
snake_case__ : Tuple = max_relative_positions
snake_case__ : Union[str, Any] = pad_token_id
snake_case__ : Optional[int] = position_biased_input
# Backwards compatibility
if type(_snake_case ) == str:
snake_case__ : int = [x.strip() for x in pos_att_type.lower().split('|' )]
snake_case__ : List[str] = pos_att_type
snake_case__ : Union[str, Any] = vocab_size
snake_case__ : Optional[int] = layer_norm_eps
snake_case__ : Optional[int] = kwargs.get('pooler_hidden_size', _snake_case )
snake_case__ : int = pooler_dropout
snake_case__ : str = pooler_hidden_act
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
@property
def lowercase_ ( self : Optional[int] ) ->Mapping[str, Mapping[int, str]]:
if self.task == "multiple-choice":
snake_case__ : List[Any] = {0: 'batch', 1: 'choice', 2: 'sequence'}
else:
snake_case__ : int = {0: 'batch', 1: 'sequence'}
if self._config.type_vocab_size > 0:
return OrderedDict(
[('input_ids', dynamic_axis), ('attention_mask', dynamic_axis), ('token_type_ids', dynamic_axis)] )
else:
return OrderedDict([('input_ids', dynamic_axis), ('attention_mask', dynamic_axis)] )
@property
def lowercase_ ( self : Dict ) ->int:
return 1_2
def lowercase_ ( self : Tuple, _snake_case : Union["PreTrainedTokenizerBase", "FeatureExtractionMixin"], _snake_case : int = -1, _snake_case : int = -1, _snake_case : int = -1, _snake_case : bool = False, _snake_case : Optional["TensorType"] = None, _snake_case : int = 3, _snake_case : int = 4_0, _snake_case : int = 4_0, _snake_case : "PreTrainedTokenizerBase" = None, ) ->Mapping[str, Any]:
snake_case__ : Union[str, Any] = super().generate_dummy_inputs(preprocessor=_snake_case, framework=_snake_case )
if self._config.type_vocab_size == 0 and "token_type_ids" in dummy_inputs:
del dummy_inputs["token_type_ids"]
return dummy_inputs
| 277 | 0 |
"""simple docstring"""
import copy
from dataclasses import dataclass, field
from typing import ClassVar, Dict
from ..features import ClassLabel, Features, Value
from .base import TaskTemplate
@dataclass(frozen=lowerCAmelCase_)
class UpperCAmelCase_ ( lowerCAmelCase_):
snake_case__ = field(default='''text-classification''' , metadata={'''include_in_asdict_even_if_is_default''': True})
snake_case__ = Features({'''text''': Value('''string''')})
snake_case__ = Features({'''labels''': ClassLabel})
snake_case__ = '''text'''
snake_case__ = '''labels'''
def _UpperCamelCase ( self : List[str] , __UpperCamelCase : Optional[Any] ) -> Any:
if self.label_column not in features:
raise ValueError(F'''Column {self.label_column} is not present in features.''' )
if not isinstance(features[self.label_column] , _snake_case ):
raise ValueError(F'''Column {self.label_column} is not a ClassLabel.''' )
_UpperCamelCase = copy.deepcopy(self )
_UpperCamelCase = self.label_schema.copy()
_UpperCamelCase = features[self.label_column]
_UpperCamelCase = label_schema
return task_template
@property
def _UpperCamelCase ( self : Optional[int] ) -> Dict[str, str]:
return {
self.text_column: "text",
self.label_column: "labels",
}
| 256 |
import argparse
import json
import pickle
from pathlib import Path
import requests
import torch
from huggingface_hub import hf_hub_download
from PIL import Image
from transformers import MaskFormerConfig, MaskFormerForInstanceSegmentation, MaskFormerImageProcessor, SwinConfig
from transformers.utils import logging
logging.set_verbosity_info()
a_ :str = logging.get_logger(__name__)
def lowercase_ (A : str ):
snake_case__ : Tuple = SwinConfig.from_pretrained(
'microsoft/swin-tiny-patch4-window7-224' , out_features=['stage1', 'stage2', 'stage3', 'stage4'] )
snake_case__ : List[Any] = MaskFormerConfig(backbone_config=A )
snake_case__ : Union[str, Any] = 'huggingface/label-files'
if "ade20k-full" in model_name:
# this should be ok
snake_case__ : Dict = 8_4_7
snake_case__ : List[str] = 'maskformer-ade20k-full-id2label.json'
elif "ade" in model_name:
# this should be ok
snake_case__ : Union[str, Any] = 1_5_0
snake_case__ : Any = 'ade20k-id2label.json'
elif "coco-stuff" in model_name:
# this should be ok
snake_case__ : List[str] = 1_7_1
snake_case__ : Union[str, Any] = 'maskformer-coco-stuff-id2label.json'
elif "coco" in model_name:
# TODO
snake_case__ : Dict = 1_3_3
snake_case__ : str = 'coco-panoptic-id2label.json'
elif "cityscapes" in model_name:
# this should be ok
snake_case__ : List[str] = 1_9
snake_case__ : Union[str, Any] = 'cityscapes-id2label.json'
elif "vistas" in model_name:
# this should be ok
snake_case__ : Tuple = 6_5
snake_case__ : List[str] = 'mapillary-vistas-id2label.json'
snake_case__ : Dict = json.load(open(hf_hub_download(A , A , repo_type='dataset' ) , 'r' ) )
snake_case__ : List[str] = {int(A ): v for k, v in idalabel.items()}
return config
def lowercase_ (A : Any ):
snake_case__ : Optional[int] = []
# stem
# fmt: off
rename_keys.append(('backbone.patch_embed.proj.weight', 'model.pixel_level_module.encoder.model.embeddings.patch_embeddings.projection.weight') )
rename_keys.append(('backbone.patch_embed.proj.bias', 'model.pixel_level_module.encoder.model.embeddings.patch_embeddings.projection.bias') )
rename_keys.append(('backbone.patch_embed.norm.weight', 'model.pixel_level_module.encoder.model.embeddings.norm.weight') )
rename_keys.append(('backbone.patch_embed.norm.bias', 'model.pixel_level_module.encoder.model.embeddings.norm.bias') )
# stages
for i in range(len(config.backbone_config.depths ) ):
for j in range(config.backbone_config.depths[i] ):
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm1.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_before.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm1.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_before.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.relative_position_bias_table''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.self.relative_position_bias_table''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.relative_position_index''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.self.relative_position_index''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.proj.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.output.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.attn.proj.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.attention.output.dense.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm2.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_after.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.norm2.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.layernorm_after.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc1.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.intermediate.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc1.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.intermediate.dense.bias''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc2.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.output.dense.weight''') )
rename_keys.append((F'''backbone.layers.{i}.blocks.{j}.mlp.fc2.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.blocks.{j}.output.dense.bias''') )
if i < 3:
rename_keys.append((F'''backbone.layers.{i}.downsample.reduction.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.reduction.weight''') )
rename_keys.append((F'''backbone.layers.{i}.downsample.norm.weight''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.norm.weight''') )
rename_keys.append((F'''backbone.layers.{i}.downsample.norm.bias''', F'''model.pixel_level_module.encoder.model.encoder.layers.{i}.downsample.norm.bias''') )
rename_keys.append((F'''backbone.norm{i}.weight''', F'''model.pixel_level_module.encoder.hidden_states_norms.{i}.weight''') )
rename_keys.append((F'''backbone.norm{i}.bias''', F'''model.pixel_level_module.encoder.hidden_states_norms.{i}.bias''') )
# FPN
rename_keys.append(('sem_seg_head.layer_4.weight', 'model.pixel_level_module.decoder.fpn.stem.0.weight') )
rename_keys.append(('sem_seg_head.layer_4.norm.weight', 'model.pixel_level_module.decoder.fpn.stem.1.weight') )
rename_keys.append(('sem_seg_head.layer_4.norm.bias', 'model.pixel_level_module.decoder.fpn.stem.1.bias') )
for source_index, target_index in zip(range(3 , 0 , -1 ) , range(0 , 3 ) ):
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.0.weight''') )
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.norm.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.1.weight''') )
rename_keys.append((F'''sem_seg_head.adapter_{source_index}.norm.bias''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.proj.1.bias''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.0.weight''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.norm.weight''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.1.weight''') )
rename_keys.append((F'''sem_seg_head.layer_{source_index}.norm.bias''', F'''model.pixel_level_module.decoder.fpn.layers.{target_index}.block.1.bias''') )
rename_keys.append(('sem_seg_head.mask_features.weight', 'model.pixel_level_module.decoder.mask_projection.weight') )
rename_keys.append(('sem_seg_head.mask_features.bias', 'model.pixel_level_module.decoder.mask_projection.bias') )
# Transformer decoder
for idx in range(config.decoder_config.decoder_layers ):
# self-attention out projection
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.out_proj.weight''', F'''model.transformer_module.decoder.layers.{idx}.self_attn.out_proj.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.out_proj.bias''', F'''model.transformer_module.decoder.layers.{idx}.self_attn.out_proj.bias''') )
# cross-attention out projection
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.out_proj.weight''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn.out_proj.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.out_proj.bias''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn.out_proj.bias''') )
# MLP 1
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear1.weight''', F'''model.transformer_module.decoder.layers.{idx}.fc1.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear1.bias''', F'''model.transformer_module.decoder.layers.{idx}.fc1.bias''') )
# MLP 2
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear2.weight''', F'''model.transformer_module.decoder.layers.{idx}.fc2.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.linear2.bias''', F'''model.transformer_module.decoder.layers.{idx}.fc2.bias''') )
# layernorm 1 (self-attention layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm1.weight''', F'''model.transformer_module.decoder.layers.{idx}.self_attn_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm1.bias''', F'''model.transformer_module.decoder.layers.{idx}.self_attn_layer_norm.bias''') )
# layernorm 2 (cross-attention layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm2.weight''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm2.bias''', F'''model.transformer_module.decoder.layers.{idx}.encoder_attn_layer_norm.bias''') )
# layernorm 3 (final layernorm)
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm3.weight''', F'''model.transformer_module.decoder.layers.{idx}.final_layer_norm.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.norm3.bias''', F'''model.transformer_module.decoder.layers.{idx}.final_layer_norm.bias''') )
rename_keys.append(('sem_seg_head.predictor.transformer.decoder.norm.weight', 'model.transformer_module.decoder.layernorm.weight') )
rename_keys.append(('sem_seg_head.predictor.transformer.decoder.norm.bias', 'model.transformer_module.decoder.layernorm.bias') )
# heads on top
rename_keys.append(('sem_seg_head.predictor.query_embed.weight', 'model.transformer_module.queries_embedder.weight') )
rename_keys.append(('sem_seg_head.predictor.input_proj.weight', 'model.transformer_module.input_projection.weight') )
rename_keys.append(('sem_seg_head.predictor.input_proj.bias', 'model.transformer_module.input_projection.bias') )
rename_keys.append(('sem_seg_head.predictor.class_embed.weight', 'class_predictor.weight') )
rename_keys.append(('sem_seg_head.predictor.class_embed.bias', 'class_predictor.bias') )
for i in range(3 ):
rename_keys.append((F'''sem_seg_head.predictor.mask_embed.layers.{i}.weight''', F'''mask_embedder.{i}.0.weight''') )
rename_keys.append((F'''sem_seg_head.predictor.mask_embed.layers.{i}.bias''', F'''mask_embedder.{i}.0.bias''') )
# fmt: on
return rename_keys
def lowercase_ (A : Tuple , A : Tuple , A : Optional[Any] ):
snake_case__ : Optional[int] = dct.pop(A )
snake_case__ : Union[str, Any] = val
def lowercase_ (A : Optional[Any] , A : Tuple ):
snake_case__ : Optional[int] = [int(backbone_config.embed_dim * 2**i ) for i in range(len(backbone_config.depths ) )]
for i in range(len(backbone_config.depths ) ):
snake_case__ : Optional[int] = num_features[i]
for j in range(backbone_config.depths[i] ):
# fmt: off
# read in weights + bias of input projection layer (in original implementation, this is a single matrix + bias)
snake_case__ : int = state_dict.pop(F'''backbone.layers.{i}.blocks.{j}.attn.qkv.weight''' )
snake_case__ : Tuple = state_dict.pop(F'''backbone.layers.{i}.blocks.{j}.attn.qkv.bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : str = in_proj_weight[:dim, :]
snake_case__ : int = in_proj_bias[: dim]
snake_case__ : List[Any] = in_proj_weight[
dim : dim * 2, :
]
snake_case__ : List[str] = in_proj_bias[
dim : dim * 2
]
snake_case__ : List[Any] = in_proj_weight[
-dim :, :
]
snake_case__ : Dict = in_proj_bias[-dim :]
# fmt: on
def lowercase_ (A : List[str] , A : List[Any] ):
# fmt: off
snake_case__ : str = config.decoder_config.hidden_size
for idx in range(config.decoder_config.decoder_layers ):
# read in weights + bias of self-attention input projection layer (in the original implementation, this is a single matrix + bias)
snake_case__ : List[Any] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.in_proj_weight''' )
snake_case__ : int = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.self_attn.in_proj_bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : Any = in_proj_weight[: hidden_size, :]
snake_case__ : Tuple = in_proj_bias[:config.hidden_size]
snake_case__ : List[str] = in_proj_weight[hidden_size : hidden_size * 2, :]
snake_case__ : Dict = in_proj_bias[hidden_size : hidden_size * 2]
snake_case__ : Any = in_proj_weight[-hidden_size :, :]
snake_case__ : int = in_proj_bias[-hidden_size :]
# read in weights + bias of cross-attention input projection layer (in the original implementation, this is a single matrix + bias)
snake_case__ : List[Any] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.in_proj_weight''' )
snake_case__ : List[str] = state_dict.pop(F'''sem_seg_head.predictor.transformer.decoder.layers.{idx}.multihead_attn.in_proj_bias''' )
# next, add query, keys and values (in that order) to the state dict
snake_case__ : Optional[int] = in_proj_weight[: hidden_size, :]
snake_case__ : Optional[Any] = in_proj_bias[:config.hidden_size]
snake_case__ : int = in_proj_weight[hidden_size : hidden_size * 2, :]
snake_case__ : List[str] = in_proj_bias[hidden_size : hidden_size * 2]
snake_case__ : List[str] = in_proj_weight[-hidden_size :, :]
snake_case__ : str = in_proj_bias[-hidden_size :]
# fmt: on
def lowercase_ ():
snake_case__ : Any = 'http://images.cocodataset.org/val2017/000000039769.jpg'
snake_case__ : int = Image.open(requests.get(A , stream=A ).raw )
return im
@torch.no_grad()
def lowercase_ (A : str , A : str , A : str , A : bool = False ):
snake_case__ : Optional[int] = get_maskformer_config(A )
# load original state_dict
with open(A , 'rb' ) as f:
snake_case__ : List[Any] = pickle.load(A )
snake_case__ : Optional[int] = data['model']
# for name, param in state_dict.items():
# print(name, param.shape)
# rename keys
snake_case__ : List[str] = create_rename_keys(A )
for src, dest in rename_keys:
rename_key(A , A , A )
read_in_swin_q_k_v(A , config.backbone_config )
read_in_decoder_q_k_v(A , A )
# update to torch tensors
for key, value in state_dict.items():
snake_case__ : int = torch.from_numpy(A )
# load 🤗 model
snake_case__ : str = MaskFormerForInstanceSegmentation(A )
model.eval()
for name, param in model.named_parameters():
print(A , param.shape )
snake_case__ , snake_case__ : Union[str, Any] = model.load_state_dict(A , strict=A )
assert missing_keys == [
"model.pixel_level_module.encoder.model.layernorm.weight",
"model.pixel_level_module.encoder.model.layernorm.bias",
]
assert len(A ) == 0, F'''Unexpected keys: {unexpected_keys}'''
# verify results
snake_case__ : Optional[Any] = prepare_img()
if "vistas" in model_name:
snake_case__ : int = 6_5
elif "cityscapes" in model_name:
snake_case__ : Dict = 6_5_5_3_5
else:
snake_case__ : Tuple = 2_5_5
snake_case__ : Optional[int] = True if 'ade' in model_name else False
snake_case__ : Dict = MaskFormerImageProcessor(ignore_index=A , reduce_labels=A )
snake_case__ : Any = image_processor(A , return_tensors='pt' )
snake_case__ : Any = model(**A )
print('Logits:' , outputs.class_queries_logits[0, :3, :3] )
if model_name == "maskformer-swin-tiny-ade":
snake_case__ : Tuple = torch.tensor(
[[3.6353, -4.4770, -2.6065], [0.5081, -4.2394, -3.5343], [2.1909, -5.0353, -1.9323]] )
assert torch.allclose(outputs.class_queries_logits[0, :3, :3] , A , atol=1e-4 )
print('Looks ok!' )
if pytorch_dump_folder_path is not None:
print(F'''Saving model and image processor to {pytorch_dump_folder_path}''' )
Path(A ).mkdir(exist_ok=A )
model.save_pretrained(A )
image_processor.save_pretrained(A )
if push_to_hub:
print('Pushing model and image processor to the hub...' )
model.push_to_hub(F'''nielsr/{model_name}''' )
image_processor.push_to_hub(F'''nielsr/{model_name}''' )
if __name__ == "__main__":
a_ :Optional[int] = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
"--model_name",
default="maskformer-swin-tiny-ade",
type=str,
help=("Name of the MaskFormer model you'd like to convert",),
)
parser.add_argument(
"--checkpoint_path",
default="/Users/nielsrogge/Documents/MaskFormer_checkpoints/MaskFormer-Swin-tiny-ADE20k/model.pkl",
type=str,
help="Path to the original state dict (.pth file).",
)
parser.add_argument(
"--pytorch_dump_folder_path", default=None, type=str, help="Path to the output PyTorch model directory."
)
parser.add_argument(
"--push_to_hub", action="store_true", help="Whether or not to push the converted model to the 🤗 hub."
)
a_ :Dict = parser.parse_args()
convert_maskformer_checkpoint(
args.model_name, args.checkpoint_path, args.pytorch_dump_folder_path, args.push_to_hub
)
| 277 | 0 |
from datetime import datetime
import matplotlib.pyplot as plt
import torch
def _a ( SCREAMING_SNAKE_CASE : int ) -> List[str]:
"""simple docstring"""
for param in module.parameters():
__lowerCAmelCase: str = False
def _a ( ) -> Any:
"""simple docstring"""
__lowerCAmelCase: Optional[Any] = 'cuda' if torch.cuda.is_available() else 'cpu'
if torch.backends.mps.is_available() and torch.backends.mps.is_built():
__lowerCAmelCase: Optional[Any] = 'mps'
if device == "mps":
print(
'WARNING: MPS currently doesn\'t seem to work, and messes up backpropagation without any visible torch'
' errors. I recommend using CUDA on a colab notebook or CPU instead if you\'re facing inexplicable issues'
' with generations.' )
return device
def _a ( SCREAMING_SNAKE_CASE : Optional[int] ) -> int:
"""simple docstring"""
__lowerCAmelCase: List[Any] = plt.imshow(SCREAMING_SNAKE_CASE )
fig.axes.get_xaxis().set_visible(SCREAMING_SNAKE_CASE )
fig.axes.get_yaxis().set_visible(SCREAMING_SNAKE_CASE )
plt.show()
def _a ( ) -> Tuple:
"""simple docstring"""
__lowerCAmelCase: Union[str, Any] = datetime.now()
__lowerCAmelCase: Union[str, Any] = current_time.strftime('%H:%M:%S' )
return timestamp
| 322 |
import unittest
from transformers import is_torch_available
from transformers.testing_utils import require_torch, slow, torch_device
from ...generation.test_utils import GenerationTesterMixin
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers import (
OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST,
OpenAIGPTConfig,
OpenAIGPTDoubleHeadsModel,
OpenAIGPTForSequenceClassification,
OpenAIGPTLMHeadModel,
OpenAIGPTModel,
)
class snake_case__ :
"""simple docstring"""
def __init__( self : List[str], _snake_case : Any, _snake_case : int=1_3, _snake_case : Optional[int]=7, _snake_case : int=True, _snake_case : Optional[Any]=True, _snake_case : Optional[Any]=True, _snake_case : Union[str, Any]=9_9, _snake_case : Optional[Any]=3_2, _snake_case : Tuple=5, _snake_case : str=4, _snake_case : Any=3_7, _snake_case : int="gelu", _snake_case : Optional[Any]=0.1, _snake_case : str=0.1, _snake_case : str=5_1_2, _snake_case : Dict=1_6, _snake_case : str=2, _snake_case : Union[str, Any]=0.0_2, _snake_case : Optional[int]=3, _snake_case : Union[str, Any]=4, _snake_case : Tuple=None, ) ->Optional[Any]:
snake_case__ : Optional[int] = parent
snake_case__ : List[Any] = batch_size
snake_case__ : Tuple = seq_length
snake_case__ : str = is_training
snake_case__ : Optional[int] = use_token_type_ids
snake_case__ : Any = use_labels
snake_case__ : Dict = vocab_size
snake_case__ : str = hidden_size
snake_case__ : Union[str, Any] = num_hidden_layers
snake_case__ : List[str] = num_attention_heads
snake_case__ : Union[str, Any] = intermediate_size
snake_case__ : List[Any] = hidden_act
snake_case__ : int = hidden_dropout_prob
snake_case__ : str = attention_probs_dropout_prob
snake_case__ : Any = max_position_embeddings
snake_case__ : Union[str, Any] = type_vocab_size
snake_case__ : Optional[Any] = type_sequence_label_size
snake_case__ : Optional[int] = initializer_range
snake_case__ : Optional[int] = num_labels
snake_case__ : str = num_choices
snake_case__ : int = scope
snake_case__ : List[str] = self.vocab_size - 1
def lowercase_ ( self : Union[str, Any] ) ->Tuple:
snake_case__ : List[str] = ids_tensor([self.batch_size, self.seq_length], self.vocab_size )
snake_case__ : List[str] = None
if self.use_token_type_ids:
snake_case__ : Optional[Any] = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size )
snake_case__ : Tuple = None
snake_case__ : str = None
snake_case__ : List[Any] = None
if self.use_labels:
snake_case__ : Dict = ids_tensor([self.batch_size], self.type_sequence_label_size )
snake_case__ : int = ids_tensor([self.batch_size, self.seq_length], self.num_labels )
snake_case__ : List[str] = ids_tensor([self.batch_size], self.num_choices )
snake_case__ : Union[str, Any] = OpenAIGPTConfig(
vocab_size=self.vocab_size, n_embd=self.hidden_size, n_layer=self.num_hidden_layers, n_head=self.num_attention_heads, n_positions=self.max_position_embeddings, pad_token_id=self.pad_token_id, )
snake_case__ : List[str] = ids_tensor([self.num_hidden_layers, self.num_attention_heads], 2 )
return (
config,
input_ids,
head_mask,
token_type_ids,
sequence_labels,
token_labels,
choice_labels,
)
def lowercase_ ( self : Any, _snake_case : List[str], _snake_case : Any, _snake_case : List[Any], _snake_case : Tuple, *_snake_case : Optional[Any] ) ->Tuple:
snake_case__ : Union[str, Any] = OpenAIGPTModel(config=_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Optional[Any] = model(_snake_case, token_type_ids=_snake_case, head_mask=_snake_case )
snake_case__ : Union[str, Any] = model(_snake_case, token_type_ids=_snake_case )
snake_case__ : Optional[Any] = model(_snake_case )
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size) )
def lowercase_ ( self : Optional[int], _snake_case : Optional[Any], _snake_case : Union[str, Any], _snake_case : Optional[int], _snake_case : List[Any], *_snake_case : Dict ) ->Optional[int]:
snake_case__ : Optional[Any] = OpenAIGPTLMHeadModel(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Tuple = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.loss.shape, () )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size) )
def lowercase_ ( self : int, _snake_case : Tuple, _snake_case : List[str], _snake_case : List[Any], _snake_case : List[Any], *_snake_case : List[Any] ) ->Optional[int]:
snake_case__ : List[str] = OpenAIGPTDoubleHeadsModel(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : Optional[Any] = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.loss.shape, () )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size) )
def lowercase_ ( self : Optional[int], _snake_case : Tuple, _snake_case : Dict, _snake_case : List[str], _snake_case : Optional[Any], *_snake_case : Union[str, Any] ) ->str:
snake_case__ : List[str] = self.num_labels
snake_case__ : Dict = OpenAIGPTForSequenceClassification(_snake_case )
model.to(_snake_case )
model.eval()
snake_case__ : List[str] = ids_tensor([self.batch_size], self.type_sequence_label_size )
snake_case__ : List[str] = model(_snake_case, token_type_ids=_snake_case, labels=_snake_case )
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels) )
def lowercase_ ( self : Dict ) ->int:
snake_case__ : List[Any] = self.prepare_config_and_inputs()
(
(
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) , (
snake_case__
) ,
) : Optional[Any] = config_and_inputs
snake_case__ : str = {
'input_ids': input_ids,
'token_type_ids': token_type_ids,
'head_mask': head_mask,
}
return config, inputs_dict
@require_torch
class snake_case__ ( lowerCAmelCase_ , lowerCAmelCase_ , lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = (
(OpenAIGPTModel, OpenAIGPTLMHeadModel, OpenAIGPTDoubleHeadsModel, OpenAIGPTForSequenceClassification)
if is_torch_available()
else ()
)
_SCREAMING_SNAKE_CASE = (
(OpenAIGPTLMHeadModel,) if is_torch_available() else ()
) # TODO (PVP): Add Double HeadsModel when generate() function is changed accordingly
_SCREAMING_SNAKE_CASE = (
{
"""feature-extraction""": OpenAIGPTModel,
"""text-classification""": OpenAIGPTForSequenceClassification,
"""text-generation""": OpenAIGPTLMHeadModel,
"""zero-shot""": OpenAIGPTForSequenceClassification,
}
if is_torch_available()
else {}
)
def lowercase_ ( self : Optional[int], _snake_case : Union[str, Any], _snake_case : int, _snake_case : Tuple, _snake_case : Tuple, _snake_case : List[str] ) ->Optional[Any]:
if pipeline_test_casse_name == "ZeroShotClassificationPipelineTests":
# Get `tokenizer does not have a padding token` error for both fast/slow tokenizers.
# `OpenAIGPTConfig` was never used in pipeline tests, either because of a missing checkpoint or because a
# tiny config could not be created.
return True
return False
def lowercase_ ( self : Optional[Any], _snake_case : Union[str, Any], _snake_case : List[str], _snake_case : Any=False ) ->Tuple:
snake_case__ : Optional[int] = super()._prepare_for_class(_snake_case, _snake_case, return_labels=_snake_case )
if return_labels:
if model_class.__name__ == "OpenAIGPTDoubleHeadsModel":
snake_case__ : Union[str, Any] = torch.zeros(
(self.model_tester.batch_size, self.model_tester.num_choices, self.model_tester.seq_length), dtype=torch.long, device=_snake_case, )
snake_case__ : List[Any] = inputs_dict['labels']
snake_case__ : List[Any] = inputs_dict['labels']
snake_case__ : Any = torch.zeros(
(self.model_tester.batch_size, self.model_tester.num_choices), dtype=torch.long, device=_snake_case, )
snake_case__ : Tuple = torch.zeros(
self.model_tester.batch_size, dtype=torch.long, device=_snake_case )
return inputs_dict
def lowercase_ ( self : Union[str, Any] ) ->List[str]:
snake_case__ : List[str] = OpenAIGPTModelTester(self )
snake_case__ : Any = ConfigTester(self, config_class=_snake_case, n_embd=3_7 )
def lowercase_ ( self : Optional[int] ) ->str:
self.config_tester.run_common_tests()
def lowercase_ ( self : int ) ->Tuple:
snake_case__ : str = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_openai_gpt_model(*_snake_case )
def lowercase_ ( self : Tuple ) ->List[str]:
snake_case__ : Optional[int] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_lm_head_model(*_snake_case )
def lowercase_ ( self : Dict ) ->int:
snake_case__ : Optional[Any] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_double_lm_head_model(*_snake_case )
def lowercase_ ( self : int ) ->str:
snake_case__ : List[Any] = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_openai_gpt_for_sequence_classification(*_snake_case )
@slow
def lowercase_ ( self : Optional[Any] ) ->str:
for model_name in OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
snake_case__ : Optional[int] = OpenAIGPTModel.from_pretrained(_snake_case )
self.assertIsNotNone(_snake_case )
@require_torch
class snake_case__ ( unittest.TestCase ):
"""simple docstring"""
@slow
def lowercase_ ( self : Tuple ) ->Optional[int]:
snake_case__ : Union[str, Any] = OpenAIGPTLMHeadModel.from_pretrained('openai-gpt' )
model.to(_snake_case )
snake_case__ : Tuple = torch.tensor([[4_8_1, 4_7_3_5, 5_4_4]], dtype=torch.long, device=_snake_case ) # the president is
snake_case__ : int = [
4_8_1,
4_7_3_5,
5_4_4,
2_4_6,
9_6_3,
8_7_0,
7_6_2,
2_3_9,
2_4_4,
4_0_4_7_7,
2_4_4,
2_4_9,
7_1_9,
8_8_1,
4_8_7,
5_4_4,
2_4_0,
2_4_4,
6_0_3,
4_8_1,
] # the president is a very good man. " \n " i\'m sure he is, " said the
snake_case__ : Optional[int] = model.generate(_snake_case, do_sample=_snake_case )
self.assertListEqual(output_ids[0].tolist(), _snake_case )
| 277 | 0 |
import numpy as np
import torch
from torch.utils.data import Dataset
from utils import logger
class __lowerCAmelCase ( lowerCAmelCase_):
def __init__( self , lowerCAmelCase__ , lowerCAmelCase__ ) -> Tuple:
'''simple docstring'''
a__ : Union[str, Any] =params
a__ : Optional[Any] =np.array(_snake_case )
a__ : Tuple =np.array([len(_snake_case ) for t in data] )
self.check()
self.remove_long_sequences()
self.remove_empty_sequences()
self.remove_unknown_sequences()
self.check()
self.print_statistics()
def __getitem__( self , lowerCAmelCase__ ) -> str:
'''simple docstring'''
return (self.token_ids[index], self.lengths[index])
def __len__( self ) -> str:
'''simple docstring'''
return len(self.lengths )
def _lowercase ( self ) -> Optional[int]:
'''simple docstring'''
assert len(self.token_ids ) == len(self.lengths )
assert all(self.lengths[i] == len(self.token_ids[i] ) for i in range(len(self.lengths ) ) )
def _lowercase ( self ) -> Optional[int]:
'''simple docstring'''
a__ : Dict =self.params.max_model_input_size
a__ : int =self.lengths > max_len
logger.info(F'''Splitting {sum(_snake_case )} too long sequences.''' )
def divide_chunks(lowerCAmelCase__ , lowerCAmelCase__ ):
return [l[i : i + n] for i in range(0 , len(_snake_case ) , _snake_case )]
a__ : Dict =[]
a__ : Any =[]
if self.params.mlm:
a__ : str =self.params.special_tok_ids['cls_token'], self.params.special_tok_ids['sep_token']
else:
a__ : Dict =self.params.special_tok_ids['bos_token'], self.params.special_tok_ids['eos_token']
for seq_, len_ in zip(self.token_ids , self.lengths ):
assert (seq_[0] == cls_id) and (seq_[-1] == sep_id), seq_
if len_ <= max_len:
new_tok_ids.append(seq_ )
new_lengths.append(len_ )
else:
a__ : int =[]
for sub_s in divide_chunks(seq_ , max_len - 2 ):
if sub_s[0] != cls_id:
a__ : List[Any] =np.insert(_snake_case , 0 , _snake_case )
if sub_s[-1] != sep_id:
a__ : Optional[Any] =np.insert(_snake_case , len(_snake_case ) , _snake_case )
assert len(_snake_case ) <= max_len
assert (sub_s[0] == cls_id) and (sub_s[-1] == sep_id), sub_s
sub_seqs.append(_snake_case )
new_tok_ids.extend(_snake_case )
new_lengths.extend([len(_snake_case ) for l in sub_seqs] )
a__ : int =np.array(_snake_case )
a__ : Union[str, Any] =np.array(_snake_case )
def _lowercase ( self ) -> Optional[int]:
'''simple docstring'''
a__ : int =len(self )
a__ : Dict =self.lengths > 1_1
a__ : List[Any] =self.token_ids[indices]
a__ : Any =self.lengths[indices]
a__ : Dict =len(self )
logger.info(F'''Remove {init_size - new_size} too short (<=11 tokens) sequences.''' )
def _lowercase ( self ) -> Tuple:
'''simple docstring'''
if "unk_token" not in self.params.special_tok_ids:
return
else:
a__ : str =self.params.special_tok_ids['unk_token']
a__ : str =len(self )
a__ : Any =np.array([np.count_nonzero(a == unk_token_id ) for a in self.token_ids] )
a__ : Any =(unk_occs / self.lengths) < 0.5
a__ : Optional[int] =self.token_ids[indices]
a__ : List[str] =self.lengths[indices]
a__ : List[Any] =len(self )
logger.info(F'''Remove {init_size - new_size} sequences with a high level of unknown tokens (50%).''' )
def _lowercase ( self ) -> Optional[int]:
'''simple docstring'''
if not self.params.is_master:
return
logger.info(F'''{len(self )} sequences''' )
# data_len = sum(self.lengths)
# nb_unique_tokens = len(Counter(list(chain(*self.token_ids))))
# logger.info(f'{data_len} tokens ({nb_unique_tokens} unique)')
# unk_idx = self.params.special_tok_ids['unk_token']
# nb_unknown = sum([(t==unk_idx).sum() for t in self.token_ids])
# logger.info(f'{nb_unknown} unknown tokens (covering {100*nb_unknown/data_len:.2f}% of the data)')
def _lowercase ( self , lowerCAmelCase__ ) -> List[Any]:
'''simple docstring'''
a__ : Any =[t[0] for t in batch]
a__ : Optional[Any] =[t[1] for t in batch]
assert len(_snake_case ) == len(_snake_case )
# Max for paddings
a__ : Tuple =max(_snake_case )
# Pad token ids
if self.params.mlm:
a__ : Dict =self.params.special_tok_ids['pad_token']
else:
a__ : List[Any] =self.params.special_tok_ids['unk_token']
a__ : Union[str, Any] =[list(t.astype(_snake_case ) ) + [pad_idx] * (max_seq_len_ - len(_snake_case )) for t in token_ids]
assert len(tk_ ) == len(_snake_case )
assert all(len(_snake_case ) == max_seq_len_ for t in tk_ )
a__ : int =torch.tensor(tk_ ) # (bs, max_seq_len_)
a__ : Optional[int] =torch.tensor(_snake_case ) # (bs)
return tk_t, lg_t
| 95 |
import os
import unittest
from transformers.models.transfo_xl.tokenization_transfo_xl import VOCAB_FILES_NAMES, TransfoXLTokenizer
from ...test_tokenization_common import TokenizerTesterMixin
class snake_case__ ( lowerCAmelCase_ , unittest.TestCase ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = TransfoXLTokenizer
_SCREAMING_SNAKE_CASE = False
_SCREAMING_SNAKE_CASE = False
def lowercase_ ( self : Optional[int] ) ->Any:
super().setUp()
snake_case__ : Tuple = [
'<unk>',
'[CLS]',
'[SEP]',
'want',
'unwanted',
'wa',
'un',
'running',
',',
'low',
'l',
]
snake_case__ : Any = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES['vocab_file'] )
with open(self.vocab_file, 'w', encoding='utf-8' ) as vocab_writer:
vocab_writer.write(''.join([x + '\n' for x in vocab_tokens] ) )
def lowercase_ ( self : Union[str, Any], **_snake_case : List[Any] ) ->Dict:
snake_case__ : str = True
return TransfoXLTokenizer.from_pretrained(self.tmpdirname, **_snake_case )
def lowercase_ ( self : Optional[Any], _snake_case : str ) ->Dict:
snake_case__ : List[Any] = '<unk> UNwanted , running'
snake_case__ : List[Any] = '<unk> unwanted, running'
return input_text, output_text
def lowercase_ ( self : List[Any] ) ->Tuple:
snake_case__ : Dict = TransfoXLTokenizer(vocab_file=self.vocab_file, lower_case=_snake_case )
snake_case__ : str = tokenizer.tokenize('<unk> UNwanted , running' )
self.assertListEqual(_snake_case, ['<unk>', 'unwanted', ',', 'running'] )
self.assertListEqual(tokenizer.convert_tokens_to_ids(_snake_case ), [0, 4, 8, 7] )
def lowercase_ ( self : List[str] ) ->List[Any]:
snake_case__ : str = TransfoXLTokenizer(lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHeLLo ! how \n Are yoU ? ' ), ['hello', '!', 'how', 'are', 'you', '?'] )
def lowercase_ ( self : Optional[int] ) ->Optional[Any]:
snake_case__ : Optional[int] = TransfoXLTokenizer(lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(' \tHeLLo ! how \n Are yoU ? ' ), ['HeLLo', '!', 'how', 'Are', 'yoU', '?'] )
def lowercase_ ( self : Optional[int] ) ->Union[str, Any]:
snake_case__ : List[Any] = TransfoXLTokenizer(lower_case=_snake_case )
snake_case__ : Dict = 'Hello (bracket) and side-scrolled [and] Henry\'s $5,000 with 3.34 m. What\'s up!?'
snake_case__ : List[Any] = [
'Hello',
'(',
'bracket',
')',
'and',
'side',
'@-@',
'scrolled',
'[',
'and',
']',
'Henry',
'\'s',
'$',
'5',
'@,@',
'000',
'with',
'3',
'@.@',
'34',
'm',
'.',
'What',
'\'s',
'up',
'!',
'?',
]
self.assertListEqual(tokenizer.tokenize(_snake_case ), _snake_case )
self.assertEqual(tokenizer.convert_tokens_to_string(_snake_case ), _snake_case )
def lowercase_ ( self : Dict ) ->Any:
snake_case__ : Dict = self.get_tokenizer()
snake_case__ : Optional[Any] = len(_snake_case )
tokenizer.add_tokens(['new1', 'new2'] )
tokenizer.move_added_token('new1', 1 )
# Check that moved token is not copied (duplicate)
self.assertEqual(len(_snake_case ), original_len + 2 )
# Check that token is moved to specified id
self.assertEqual(tokenizer.encode('new1' ), [1] )
self.assertEqual(tokenizer.decode([1] ), 'new1' )
| 277 | 0 |
from typing import TYPE_CHECKING
from ...utils import (
OptionalDependencyNotAvailable,
_LazyModule,
is_flax_available,
is_tf_available,
is_tokenizers_available,
is_torch_available,
is_vision_available,
)
__a = {
"configuration_owlvit": [
"OWLVIT_PRETRAINED_CONFIG_ARCHIVE_MAP",
"OwlViTConfig",
"OwlViTOnnxConfig",
"OwlViTTextConfig",
"OwlViTVisionConfig",
],
"processing_owlvit": ["OwlViTProcessor"],
}
try:
if not is_vision_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a = ["OwlViTFeatureExtractor"]
__a = ["OwlViTImageProcessor"]
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
__a = [
"OWLVIT_PRETRAINED_MODEL_ARCHIVE_LIST",
"OwlViTModel",
"OwlViTPreTrainedModel",
"OwlViTTextModel",
"OwlViTVisionModel",
"OwlViTForObjectDetection",
]
if TYPE_CHECKING:
from .configuration_owlvit import (
OWLVIT_PRETRAINED_CONFIG_ARCHIVE_MAP,
OwlViTConfig,
OwlViTOnnxConfig,
OwlViTTextConfig,
OwlViTVisionConfig,
)
from .processing_owlvit import OwlViTProcessor
try:
if not is_vision_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .feature_extraction_owlvit import OwlViTFeatureExtractor
from .image_processing_owlvit import OwlViTImageProcessor
try:
if not is_torch_available():
raise OptionalDependencyNotAvailable()
except OptionalDependencyNotAvailable:
pass
else:
from .modeling_owlvit import (
OWLVIT_PRETRAINED_MODEL_ARCHIVE_LIST,
OwlViTForObjectDetection,
OwlViTModel,
OwlViTPreTrainedModel,
OwlViTTextModel,
OwlViTVisionModel,
)
else:
import sys
__a = _LazyModule(__name__, globals()['__file__'], _import_structure, module_spec=__spec__)
| 30 |
from ...configuration_utils import PretrainedConfig
from ...utils import logging
a_ :Optional[int] = logging.get_logger(__name__)
a_ :Dict = {"openai-gpt": "https://huggingface.co/openai-gpt/resolve/main/config.json"}
class snake_case__ ( lowerCAmelCase_ ):
"""simple docstring"""
_SCREAMING_SNAKE_CASE = """openai-gpt"""
_SCREAMING_SNAKE_CASE = {
"""max_position_embeddings""": """n_positions""",
"""hidden_size""": """n_embd""",
"""num_attention_heads""": """n_head""",
"""num_hidden_layers""": """n_layer""",
}
def __init__( self : Optional[int], _snake_case : Dict=4_0_4_7_8, _snake_case : str=5_1_2, _snake_case : int=7_6_8, _snake_case : Tuple=1_2, _snake_case : Any=1_2, _snake_case : str="gelu", _snake_case : List[str]=0.1, _snake_case : Any=0.1, _snake_case : Dict=0.1, _snake_case : int=1e-5, _snake_case : Optional[Any]=0.0_2, _snake_case : List[Any]="cls_index", _snake_case : Any=True, _snake_case : Any=None, _snake_case : int=True, _snake_case : Optional[Any]=0.1, **_snake_case : List[Any], ) ->Optional[int]:
snake_case__ : int = vocab_size
snake_case__ : Dict = n_positions
snake_case__ : str = n_embd
snake_case__ : str = n_layer
snake_case__ : List[Any] = n_head
snake_case__ : List[Any] = afn
snake_case__ : Optional[Any] = resid_pdrop
snake_case__ : List[str] = embd_pdrop
snake_case__ : List[Any] = attn_pdrop
snake_case__ : Optional[int] = layer_norm_epsilon
snake_case__ : str = initializer_range
snake_case__ : List[str] = summary_type
snake_case__ : Optional[int] = summary_use_proj
snake_case__ : List[str] = summary_activation
snake_case__ : Optional[Any] = summary_first_dropout
snake_case__ : int = summary_proj_to_labels
super().__init__(**_snake_case )
| 277 | 0 |
"""simple docstring"""
import json
import os
import unittest
from transformers.models.roc_bert.tokenization_roc_bert import (
VOCAB_FILES_NAMES,
RoCBertBasicTokenizer,
RoCBertTokenizer,
RoCBertWordpieceTokenizer,
_is_control,
_is_punctuation,
_is_whitespace,
)
from transformers.testing_utils import require_tokenizers, slow
from ...test_tokenization_common import TokenizerTesterMixin, filter_non_english
@require_tokenizers
class __lowerCAmelCase ( lowerCAmelCase_ , unittest.TestCase ):
'''simple docstring'''
__UpperCAmelCase : List[str] = RoCBertTokenizer
__UpperCAmelCase : Tuple = None
__UpperCAmelCase : Union[str, Any] = False
__UpperCAmelCase : Tuple = True
__UpperCAmelCase : str = filter_non_english
def __UpperCAmelCase ( self ):
super().setUp()
__a = ['[UNK]', '[CLS]', '[SEP]', '[PAD]', '[MASK]', '你', '好', '是', '谁', 'a', 'b', 'c', 'd']
__a = {}
__a = {}
for i, value in enumerate(_snake_case ):
__a = i
__a = i
__a = os.path.join(self.tmpdirname , VOCAB_FILES_NAMES['''vocab_file'''] )
__a = os.path.join(self.tmpdirname , VOCAB_FILES_NAMES['''word_shape_file'''] )
__a = os.path.join(self.tmpdirname , VOCAB_FILES_NAMES['''word_pronunciation_file'''] )
with open(self.vocab_file , '''w''' , encoding='''utf-8''' ) as vocab_writer:
vocab_writer.write(''''''.join([x + '''\n''' for x in vocab_tokens] ) )
with open(self.word_shape_file , '''w''' , encoding='''utf-8''' ) as word_shape_writer:
json.dump(_snake_case , _snake_case , ensure_ascii=_snake_case )
with open(self.word_pronunciation_file , '''w''' , encoding='''utf-8''' ) as word_pronunciation_writer:
json.dump(_snake_case , _snake_case , ensure_ascii=_snake_case )
def __UpperCAmelCase ( self ):
__a = self.tokenizer_class(self.vocab_file , self.word_shape_file , self.word_pronunciation_file )
__a = tokenizer.tokenize('''你好[SEP]你是谁''' )
self.assertListEqual(_snake_case , ['''你''', '''好''', '''[SEP]''', '''你''', '''是''', '''谁'''] )
self.assertListEqual(tokenizer.convert_tokens_to_ids(_snake_case ) , [5, 6, 2, 5, 7, 8] )
self.assertListEqual(tokenizer.convert_tokens_to_shape_ids(_snake_case ) , [5, 6, 2, 5, 7, 8] )
self.assertListEqual(tokenizer.convert_tokens_to_pronunciation_ids(_snake_case ) , [5, 6, 2, 5, 7, 8] )
def __UpperCAmelCase ( self ):
__a = RoCBertBasicTokenizer()
self.assertListEqual(tokenizer.tokenize('''ah\u535A\u63A8zz''' ) , ['''ah''', '''\u535A''', '''\u63A8''', '''zz'''] )
def __UpperCAmelCase ( self ):
__a = RoCBertBasicTokenizer(do_lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(''' \tHeLLo!how \n Are yoU? ''' ) , ['''hello''', '''!''', '''how''', '''are''', '''you''', '''?'''] )
self.assertListEqual(tokenizer.tokenize('''H\u00E9llo''' ) , ['''hello'''] )
def __UpperCAmelCase ( self ):
__a = RoCBertBasicTokenizer(do_lower_case=_snake_case , strip_accents=_snake_case )
self.assertListEqual(
tokenizer.tokenize(''' \tHäLLo!how \n Are yoU? ''' ) , ['''hällo''', '''!''', '''how''', '''are''', '''you''', '''?'''] )
self.assertListEqual(tokenizer.tokenize('''H\u00E9llo''' ) , ['''h\u00E9llo'''] )
def __UpperCAmelCase ( self ):
__a = RoCBertBasicTokenizer(do_lower_case=_snake_case , strip_accents=_snake_case )
self.assertListEqual(
tokenizer.tokenize(''' \tHäLLo!how \n Are yoU? ''' ) , ['''hallo''', '''!''', '''how''', '''are''', '''you''', '''?'''] )
self.assertListEqual(tokenizer.tokenize('''H\u00E9llo''' ) , ['''hello'''] )
def __UpperCAmelCase ( self ):
__a = RoCBertBasicTokenizer(do_lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(''' \tHäLLo!how \n Are yoU? ''' ) , ['''hallo''', '''!''', '''how''', '''are''', '''you''', '''?'''] )
self.assertListEqual(tokenizer.tokenize('''H\u00E9llo''' ) , ['''hello'''] )
def __UpperCAmelCase ( self ):
__a = RoCBertBasicTokenizer(do_lower_case=_snake_case )
self.assertListEqual(
tokenizer.tokenize(''' \tHeLLo!how \n Are yoU? ''' ) , ['''HeLLo''', '''!''', '''how''', '''Are''', '''yoU''', '''?'''] )
def __UpperCAmelCase ( self ):
__a = RoCBertBasicTokenizer(do_lower_case=_snake_case , strip_accents=_snake_case )
self.assertListEqual(
tokenizer.tokenize(''' \tHäLLo!how \n Are yoU? ''' ) , ['''HäLLo''', '''!''', '''how''', '''Are''', '''yoU''', '''?'''] )
def __UpperCAmelCase ( self ):
__a = RoCBertBasicTokenizer(do_lower_case=_snake_case , strip_accents=_snake_case )
self.assertListEqual(
tokenizer.tokenize(''' \tHäLLo!how \n Are yoU? ''' ) , ['''HaLLo''', '''!''', '''how''', '''Are''', '''yoU''', '''?'''] )
def __UpperCAmelCase ( self ):
__a = RoCBertBasicTokenizer(do_lower_case=_snake_case , never_split=['''[UNK]'''] )
self.assertListEqual(
tokenizer.tokenize(''' \tHeLLo!how \n Are yoU? [UNK]''' ) , ['''HeLLo''', '''!''', '''how''', '''Are''', '''yoU''', '''?''', '''[UNK]'''] )
def __UpperCAmelCase ( self ):
__a = ['[UNK]', '[CLS]', '[SEP]', 'want', '##want', '##ed', 'wa', 'un', 'runn', '##ing']
__a = {}
for i, token in enumerate(_snake_case ):
__a = i
__a = RoCBertWordpieceTokenizer(vocab=_snake_case , unk_token='''[UNK]''' )
self.assertListEqual(tokenizer.tokenize('''''' ) , [] )
self.assertListEqual(tokenizer.tokenize('''unwanted running''' ) , ['''un''', '''##want''', '''##ed''', '''runn''', '''##ing'''] )
self.assertListEqual(tokenizer.tokenize('''unwantedX running''' ) , ['''[UNK]''', '''runn''', '''##ing'''] )
def __UpperCAmelCase ( self ):
self.assertTrue(_is_whitespace(''' ''' ) )
self.assertTrue(_is_whitespace('''\t''' ) )
self.assertTrue(_is_whitespace('''\r''' ) )
self.assertTrue(_is_whitespace('''\n''' ) )
self.assertTrue(_is_whitespace('''\u00A0''' ) )
self.assertFalse(_is_whitespace('''A''' ) )
self.assertFalse(_is_whitespace('''-''' ) )
def __UpperCAmelCase ( self ):
self.assertTrue(_is_control('''\u0005''' ) )
self.assertFalse(_is_control('''A''' ) )
self.assertFalse(_is_control(''' ''' ) )
self.assertFalse(_is_control('''\t''' ) )
self.assertFalse(_is_control('''\r''' ) )
def __UpperCAmelCase ( self ):
self.assertTrue(_is_punctuation('''-''' ) )
self.assertTrue(_is_punctuation('''$''' ) )
self.assertTrue(_is_punctuation('''`''' ) )
self.assertTrue(_is_punctuation('''.''' ) )
self.assertFalse(_is_punctuation('''A''' ) )
self.assertFalse(_is_punctuation(''' ''' ) )
def __UpperCAmelCase ( self ):
__a = self.get_tokenizer()
# Example taken from the issue https://github.com/huggingface/tokenizers/issues/340
self.assertListEqual([tokenizer.tokenize(_snake_case ) for t in ['''Test''', '''\xad''', '''test''']] , [['''[UNK]'''], [], ['''[UNK]''']] )
if self.test_rust_tokenizer:
__a = self.get_rust_tokenizer()
self.assertListEqual(
[rust_tokenizer.tokenize(_snake_case ) for t in ['''Test''', '''\xad''', '''test''']] , [['''[UNK]'''], [], ['''[UNK]''']] )
def __UpperCAmelCase ( self ):
for tokenizer, pretrained_name, kwargs in self.tokenizers_list:
with self.subTest(f'''{tokenizer.__class__.__name__} ({pretrained_name})''' ):
__a = self.rust_tokenizer_class.from_pretrained(_snake_case , **_snake_case )
__a = f'''A, naïve {tokenizer_r.mask_token} AllenNLP sentence.'''
__a = tokenizer_r.encode_plus(
_snake_case , return_attention_mask=_snake_case , return_token_type_ids=_snake_case , return_offsets_mapping=_snake_case , add_special_tokens=_snake_case , )
__a = tokenizer_r.do_lower_case if hasattr(_snake_case , '''do_lower_case''' ) else False
__a = (
[
((0, 0), tokenizer_r.cls_token),
((0, 1), 'A'),
((1, 2), ','),
((3, 5), 'na'),
((5, 6), '##ï'),
((6, 8), '##ve'),
((9, 15), tokenizer_r.mask_token),
((16, 21), 'Allen'),
((21, 23), '##NL'),
((23, 24), '##P'),
((25, 33), 'sentence'),
((33, 34), '.'),
((0, 0), tokenizer_r.sep_token),
]
if not do_lower_case
else [
((0, 0), tokenizer_r.cls_token),
((0, 1), 'a'),
((1, 2), ','),
((3, 8), 'naive'),
((9, 15), tokenizer_r.mask_token),
((16, 21), 'allen'),
((21, 23), '##nl'),
((23, 24), '##p'),
((25, 33), 'sentence'),
((33, 34), '.'),
((0, 0), tokenizer_r.sep_token),
]
)
self.assertEqual(
[e[1] for e in expected_results] , tokenizer_r.convert_ids_to_tokens(tokens['''input_ids'''] ) )
self.assertEqual([e[0] for e in expected_results] , tokens['''offset_mapping'''] )
def __UpperCAmelCase ( self ):
__a = ['的', '人', '有']
__a = ''.join(_snake_case )
for tokenizer, pretrained_name, kwargs in self.tokenizers_list:
with self.subTest(f'''{tokenizer.__class__.__name__} ({pretrained_name})''' ):
__a = True
__a = self.tokenizer_class.from_pretrained(_snake_case , **_snake_case )
__a = self.rust_tokenizer_class.from_pretrained(_snake_case , **_snake_case )
__a = tokenizer_p.encode(_snake_case , add_special_tokens=_snake_case )
__a = tokenizer_r.encode(_snake_case , add_special_tokens=_snake_case )
__a = tokenizer_r.convert_ids_to_tokens(_snake_case )
__a = tokenizer_p.convert_ids_to_tokens(_snake_case )
# it is expected that each Chinese character is not preceded by "##"
self.assertListEqual(_snake_case , _snake_case )
self.assertListEqual(_snake_case , _snake_case )
__a = False
__a = self.rust_tokenizer_class.from_pretrained(_snake_case , **_snake_case )
__a = self.tokenizer_class.from_pretrained(_snake_case , **_snake_case )
__a = tokenizer_r.encode(_snake_case , add_special_tokens=_snake_case )
__a = tokenizer_p.encode(_snake_case , add_special_tokens=_snake_case )
__a = tokenizer_r.convert_ids_to_tokens(_snake_case )
__a = tokenizer_p.convert_ids_to_tokens(_snake_case )
# it is expected that only the first Chinese character is not preceded by "##".
__a = [
f'''##{token}''' if idx != 0 else token for idx, token in enumerate(_snake_case )
]
self.assertListEqual(_snake_case , _snake_case )
self.assertListEqual(_snake_case , _snake_case )
@slow
def __UpperCAmelCase ( self ):
__a = self.tokenizer_class(self.vocab_file , self.word_shape_file , self.word_pronunciation_file )
__a = tokenizer.encode('''你好''' , add_special_tokens=_snake_case )
__a = tokenizer.encode('''你是谁''' , add_special_tokens=_snake_case )
__a = tokenizer.build_inputs_with_special_tokens(_snake_case )
__a = tokenizer.build_inputs_with_special_tokens(_snake_case , _snake_case )
assert encoded_sentence == [1] + text + [2]
assert encoded_pair == [1] + text + [2] + text_a + [2]
def __UpperCAmelCase ( self ):
__a = self.get_tokenizers(do_lower_case=_snake_case )
for tokenizer in tokenizers:
with self.subTest(f'''{tokenizer.__class__.__name__}''' ):
__a = '你好，你是谁'
__a = tokenizer.tokenize(_snake_case )
__a = tokenizer.convert_tokens_to_ids(_snake_case )
__a = tokenizer.convert_tokens_to_shape_ids(_snake_case )
__a = tokenizer.convert_tokens_to_pronunciation_ids(_snake_case )
__a = tokenizer.prepare_for_model(
_snake_case , _snake_case , _snake_case , add_special_tokens=_snake_case )
__a = tokenizer.encode_plus(_snake_case , add_special_tokens=_snake_case )
self.assertEqual(_snake_case , _snake_case )
| 45 |
import argparse
import logging
import os
from datetime import datetime
import numpy as np
import torch
from torch import nn
from torch.utils.data import DataLoader, RandomSampler, TensorDataset
from tqdm import tqdm
from transformers import GPTaLMHeadModel
a_ :Optional[Any] = logging.getLogger(__name__)
def lowercase_ (A : List[Any] , A : List[Any] ):
# save results
if os.path.exists(A ):
if os.path.exists(os.path.join(A , 'config.json' ) ) and os.path.isfile(
os.path.join(A , 'config.json' ) ):
os.remove(os.path.join(A , 'config.json' ) )
if os.path.exists(os.path.join(A , 'pytorch_model.bin' ) ) and os.path.isfile(
os.path.join(A , 'pytorch_model.bin' ) ):
os.remove(os.path.join(A , 'pytorch_model.bin' ) )
else:
os.makedirs(A )
model.save_pretrained(A )
def lowercase_ (A : Any , A : Optional[Any]=False ):
snake_case__ : str = 2
if unlogit:
snake_case__ : Dict = torch.pow(A , A )
snake_case__ : Any = p * torch.log(A )
snake_case__ : Tuple = 0
return -plogp.sum(dim=-1 )
def lowercase_ (A : List[str] ):
logger.info('lv, h >\t' + '\t'.join(F'''{x + 1}''' for x in range(len(A ) ) ) )
for row in range(len(A ) ):
if tensor.dtype != torch.long:
logger.info(F'''layer {row + 1}:\t''' + '\t'.join(F'''{x:.5f}''' for x in tensor[row].cpu().data ) )
else:
logger.info(F'''layer {row + 1}:\t''' + '\t'.join(F'''{x:d}''' for x in tensor[row].cpu().data ) )
def lowercase_ (A : Tuple , A : Optional[Any] , A : str , A : int=True , A : Optional[int]=True , A : Any=None , A : int=False ):
snake_case__ , snake_case__ : Optional[Any] = model.config.num_hidden_layers, model.config.num_attention_heads
snake_case__ : int = torch.zeros(A , A ).to(args.device )
snake_case__ : Any = torch.zeros(A , A ).to(args.device )
if head_mask is None:
snake_case__ : Dict = torch.ones(A , A ).to(args.device )
head_mask.requires_grad_(requires_grad=A )
# If actually pruned attention multi-head, set head mask to None to avoid shape mismatch
if actually_pruned:
snake_case__ : Optional[int] = None
snake_case__ : List[Any] = 0.0
snake_case__ : str = 0.0
for step, inputs in enumerate(tqdm(A , desc='Iteration' , disable=args.local_rank not in [-1, 0] ) ):
snake_case__ : Union[str, Any] = tuple(t.to(args.device ) for t in inputs )
((snake_case__) , ) : Optional[Any] = inputs
# Do a forward pass (not with torch.no_grad() since we need gradients for importance score - see below)
snake_case__ : Union[str, Any] = model(A , labels=A , head_mask=A )
# (loss), lm_logits, presents, (all hidden_states), (attentions)
snake_case__ , snake_case__ , snake_case__ : Dict = (
outputs[0],
outputs[1],
outputs[-1],
) # Loss and logits are the first, attention the last
loss.backward() # Backpropagate to populate the gradients in the head mask
total_loss += loss.detach().cpu().numpy()
if compute_entropy:
for layer, attn in enumerate(A ):
snake_case__ : Optional[Any] = entropy(attn.detach() , A )
attn_entropy[layer] += masked_entropy.sum(-1 ).sum(0 ).sum(0 ).detach()
if compute_importance:
head_importance += head_mask.grad.abs().detach()
tot_tokens += torch.ones_like(A ).float().detach().sum().data
# Normalize
attn_entropy /= tot_tokens
head_importance /= tot_tokens
# Layerwise importance normalization
if not args.dont_normalize_importance_by_layer:
snake_case__ : Union[str, Any] = 2
snake_case__ : List[Any] = torch.pow(torch.pow(A , A ).sum(-1 ) , 1 / exponent )
head_importance /= norm_by_layer.unsqueeze(-1 ) + 1e-20
if not args.dont_normalize_global_importance:
snake_case__ : Tuple = (head_importance - head_importance.min()) / (head_importance.max() - head_importance.min())
# Print matrices
if compute_entropy:
logger.info('Attention entropies' )
print_ad_tensor(A )
if compute_importance:
logger.info('Head importance scores' )
print_ad_tensor(A )
logger.info('Head ranked by importance scores' )
snake_case__ : Tuple = torch.zeros(head_importance.numel() , dtype=torch.long , device=args.device )
snake_case__ : Union[str, Any] = torch.arange(
head_importance.numel() , device=args.device )
snake_case__ : str = head_ranks.view_as(A )
print_ad_tensor(A )
return attn_entropy, head_importance, total_loss
def lowercase_ (A : Optional[int] , A : Dict , A : Optional[int] ):
snake_case__ , snake_case__ , snake_case__ : Any = compute_heads_importance(A , A , A , compute_entropy=A )
snake_case__ : Tuple = 1 / loss # instead of downsteam score use the LM loss
logger.info('Pruning: original score: %f, threshold: %f' , A , original_score * args.masking_threshold )
snake_case__ : Optional[Any] = torch.ones_like(A )
snake_case__ : Union[str, Any] = max(1 , int(new_head_mask.numel() * args.masking_amount ) )
snake_case__ : Dict = original_score
while current_score >= original_score * args.masking_threshold:
snake_case__ : int = new_head_mask.clone().detach() # save current head mask
# heads from least important to most - keep only not-masked heads
snake_case__ : List[Any] = float('Inf' )
snake_case__ : Union[str, Any] = head_importance.view(-1 ).sort()[1]
if len(A ) <= num_to_mask:
print('BREAK BY num_to_mask' )
break
# mask heads
snake_case__ : int = current_heads_to_mask[:num_to_mask]
logger.info('Heads to mask: %s' , str(current_heads_to_mask.tolist() ) )
snake_case__ : int = new_head_mask.view(-1 )
snake_case__ : int = 0.0
snake_case__ : Union[str, Any] = new_head_mask.view_as(A )
snake_case__ : List[str] = new_head_mask.clone().detach()
print_ad_tensor(A )
# Compute metric and head importance again
snake_case__ , snake_case__ , snake_case__ : Any = compute_heads_importance(
A , A , A , compute_entropy=A , head_mask=A )
snake_case__ : Dict = 1 / loss
logger.info(
'Masking: current score: %f, remaining heads %d (%.1f percents)' , A , new_head_mask.sum() , new_head_mask.sum() / new_head_mask.numel() * 1_0_0 , )
logger.info('Final head mask' )
print_ad_tensor(A )
np.save(os.path.join(args.output_dir , 'head_mask.npy' ) , head_mask.detach().cpu().numpy() )
return head_mask
def lowercase_ (A : List[str] , A : Tuple , A : Optional[Any] , A : int ):
snake_case__ : Any = datetime.now()
snake_case__ , snake_case__ , snake_case__ : str = compute_heads_importance(
A , A , A , compute_entropy=A , compute_importance=A , head_mask=A )
snake_case__ : Tuple = 1 / loss
snake_case__ : Dict = datetime.now() - before_time
snake_case__ : Union[str, Any] = sum(p.numel() for p in model.parameters() )
snake_case__ : Optional[Any] = {
layer: (1 - head_mask[layer].long()).nonzero().squeeze().tolist() for layer in range(len(A ) )
}
for k, v in heads_to_prune.items():
if isinstance(A , A ):
snake_case__ : Any = [
v,
]
assert sum(len(A ) for h in heads_to_prune.values() ) == (1 - head_mask.long()).sum().item()
model.prune_heads(A )
snake_case__ : Dict = sum(p.numel() for p in model.parameters() )
snake_case__ : Tuple = datetime.now()
snake_case__ , snake_case__ , snake_case__ : Dict = compute_heads_importance(
A , A , A , compute_entropy=A , compute_importance=A , head_mask=A , actually_pruned=A , )
snake_case__ : Any = 1 / loss
snake_case__ : int = datetime.now() - before_time
logger.info(
'Pruning: original num of params: %.2e, after pruning %.2e (%.1f percents)' , A , A , pruned_num_params / original_num_params * 1_0_0 , )
logger.info('Pruning: score with masking: %f score with pruning: %f' , A , A )
logger.info('Pruning: speed ratio (original timing / new timing): %f percents' , original_time / new_time * 1_0_0 )
save_model(A , args.output_dir )
def lowercase_ ():
snake_case__ : str = argparse.ArgumentParser()
# Required parameters
parser.add_argument(
'--data_dir' , default=A , type=A , required=A , help='The input data dir. Should contain the .tsv files (or other data files) for the task.' , )
parser.add_argument(
'--model_name_or_path' , default=A , type=A , required=A , help='Path to pretrained model or model identifier from huggingface.co/models' , )
parser.add_argument(
'--output_dir' , default=A , type=A , required=A , help='The output directory where the model predictions and checkpoints will be written.' , )
# Other parameters
parser.add_argument(
'--config_name' , default='' , type=A , help='Pretrained config name or path if not the same as model_name_or_path' , )
parser.add_argument(
'--tokenizer_name' , default='' , type=A , help='Pretrained tokenizer name or path if not the same as model_name_or_path' , )
parser.add_argument(
'--cache_dir' , default=A , type=A , help='Where do you want to store the pre-trained models downloaded from s3' , )
parser.add_argument(
'--data_subset' , type=A , default=-1 , help='If > 0: limit the data to a subset of data_subset instances.' )
parser.add_argument(
'--overwrite_output_dir' , action='store_true' , help='Whether to overwrite data in output directory' )
parser.add_argument(
'--overwrite_cache' , action='store_true' , help='Overwrite the cached training and evaluation sets' )
parser.add_argument(
'--dont_normalize_importance_by_layer' , action='store_true' , help='Don\'t normalize importance score by layers' )
parser.add_argument(
'--dont_normalize_global_importance' , action='store_true' , help='Don\'t normalize all importance scores between 0 and 1' , )
parser.add_argument(
'--try_masking' , action='store_true' , help='Whether to try to mask head until a threshold of accuracy.' )
parser.add_argument(
'--masking_threshold' , default=0.9 , type=A , help='masking threshold in term of metrics (stop masking when metric < threshold * original metric value).' , )
parser.add_argument(
'--masking_amount' , default=0.1 , type=A , help='Amount to heads to masking at each masking step.' )
parser.add_argument('--metric_name' , default='acc' , type=A , help='Metric to use for head masking.' )
parser.add_argument(
'--max_seq_length' , default=1_2_8 , type=A , help=(
'The maximum total input sequence length after WordPiece tokenization. \n'
'Sequences longer than this will be truncated, sequences shorter padded.'
) , )
parser.add_argument('--batch_size' , default=1 , type=A , help='Batch size.' )
parser.add_argument('--seed' , type=A , default=4_2 )
parser.add_argument('--local_rank' , type=A , default=-1 , help='local_rank for distributed training on gpus' )
parser.add_argument('--no_cuda' , action='store_true' , help='Whether not to use CUDA when available' )
parser.add_argument('--server_ip' , type=A , default='' , help='Can be used for distant debugging.' )
parser.add_argument('--server_port' , type=A , default='' , help='Can be used for distant debugging.' )
snake_case__ : Optional[int] = parser.parse_args()
if args.server_ip and args.server_port:
# Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script
import ptvsd
print('Waiting for debugger attach' )
ptvsd.enable_attach(address=(args.server_ip, args.server_port) , redirect_output=A )
ptvsd.wait_for_attach()
# Setup devices and distributed training
if args.local_rank == -1 or args.no_cuda:
snake_case__ : List[Any] = torch.device('cuda' if torch.cuda.is_available() and not args.no_cuda else 'cpu' )
snake_case__ : Optional[Any] = 0 if args.no_cuda else torch.cuda.device_count()
else:
torch.cuda.set_device(args.local_rank )
snake_case__ : int = torch.device('cuda' , args.local_rank )
snake_case__ : List[str] = 1
torch.distributed.init_process_group(backend='nccl' ) # Initializes the distributed backend
# Setup logging
logging.basicConfig(level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN )
logger.info('device: {} n_gpu: {}, distributed: {}'.format(args.device , args.n_gpu , bool(args.local_rank != -1 ) ) )
snake_case__ : Any = GPTaLMHeadModel.from_pretrained(args.model_name_or_path )
# Distributed and parallel training
model.to(args.device )
if args.local_rank != -1:
snake_case__ : List[str] = nn.parallel.DistributedDataParallel(
A , device_ids=[args.local_rank] , output_device=args.local_rank , find_unused_parameters=A )
elif args.n_gpu > 1:
snake_case__ : Optional[int] = nn.DataParallel(A )
# Print/save training arguments
os.makedirs(args.output_dir , exist_ok=A )
torch.save(A , os.path.join(args.output_dir , 'run_args.bin' ) )
logger.info('Training/evaluation parameters %s' , A )
# Prepare dataset
snake_case__ : Optional[Any] = np.concatenate(
[
np.loadtxt(args.data_dir , dtype=np.intaa ),
] )
snake_case__ : List[str] = (torch.from_numpy(A ),)
snake_case__ : int = TensorDataset(*A )
snake_case__ : Union[str, Any] = RandomSampler(A )
snake_case__ : Any = DataLoader(A , sampler=A , batch_size=args.batch_size )
# Compute head entropy and importance score
compute_heads_importance(A , A , A )
# Try head masking (set heads to zero until the score goes under a threshole)
# and head pruning (remove masked heads and see the effect on the network)
if args.try_masking and args.masking_threshold > 0.0 and args.masking_threshold < 1.0:
snake_case__ : Dict = mask_heads(A , A , A )
prune_heads(A , A , A , A )
if __name__ == "__main__":
main()
| 277 | 0 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.