File size: 16,911 Bytes
810b6bc | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 | """
Model Utilities - Helper functions for model operations
"""
import logging
from typing import Dict, Any, List, Optional, Tuple
import torch
from transformers import (
AutoModel,
AutoModelForCausalLM,
AutoModelForSeq2SeqLM,
AutoModelForTokenClassification,
AutoModelForQuestionAnswering,
AutoModelForSequenceClassification,
AutoConfig,
AutoTokenizer,
PreTrainedModel,
PreTrainedTokenizer,
)
from peft import LoraConfig, get_peft_model, TaskType, prepare_model_for_kbit_training
import os
import json
import hashlib
logger = logging.getLogger(__name__)
# Model architectures and their supported tasks
MODEL_TASK_MAPPING = {
"gpt": ["causal-lm"],
"llama": ["causal-lm"],
"mistral": ["causal-lm"],
"falcon": ["causal-lm"],
"qwen": ["causal-lm"],
"phi": ["causal-lm"],
"opt": ["causal-lm"],
"bloom": ["causal-lm"],
"t5": ["seq2seq"],
"bart": ["seq2seq"],
"pegasus": ["seq2seq"],
"mt5": ["seq2seq"],
"bert": ["token-classification", "text-classification", "question-answering"],
"roberta": ["token-classification", "text-classification", "question-answering"],
"deberta": ["token-classification", "text-classification", "question-answering"],
"xlnet": ["token-classification", "text-classification", "question-answering"],
"albert": ["token-classification", "text-classification", "question-answering"],
"electra": ["token-classification", "text-classification"],
"distilbert": ["token-classification", "text-classification", "question-answering"],
}
# PEFT task type mapping
PEFT_TASK_TYPES = {
"causal-lm": TaskType.CAUSAL_LM,
"seq2seq": TaskType.SEQ_2_SEQ_LM,
"token-classification": TaskType.TOKEN_CLS,
"text-classification": TaskType.SEQ_CLS,
"question-answering": TaskType.QUESTION_ANS,
}
def get_model_for_task(model_name: str, task_type: str, **kwargs) -> Tuple[PreTrainedModel, Optional[str]]:
"""Load appropriate model for a task type."""
try:
config = AutoConfig.from_pretrained(model_name)
# Determine model class
if task_type == "causal-lm":
model_class = AutoModelForCausalLM
elif task_type == "seq2seq":
model_class = AutoModelForSeq2SeqLM
elif task_type == "token-classification":
model_class = AutoModelForTokenClassification
elif task_type == "text-classification":
model_class = AutoModelForSequenceClassification
elif task_type == "question-answering":
model_class = AutoModelForQuestionAnswering
else:
model_class = AutoModel
# Load model
model = model_class.from_pretrained(
model_name,
config=config,
**kwargs
)
return model, None
except Exception as e:
logger.error(f"Error loading model {model_name} for task {task_type}: {e}")
return None, str(e)
def load_tokenizer(model_name: str, **kwargs) -> Tuple[PreTrainedTokenizer, Optional[str]]:
"""Load tokenizer for a model."""
try:
tokenizer = AutoTokenizer.from_pretrained(model_name, **kwargs)
# Ensure pad token is set
if tokenizer.pad_token is None:
tokenizer.pad_token = tokenizer.eos_token or "<pad>"
tokenizer.pad_token_id = tokenizer.eos_token_id or tokenizer.convert_tokens_to_ids("<pad>")
return tokenizer, None
except Exception as e:
logger.error(f"Error loading tokenizer for {model_name}: {e}")
return None, str(e)
def get_model_info(model_name: str) -> Dict[str, Any]:
"""Get detailed model information."""
try:
from huggingface_hub import HfApi, model_info
api = HfApi()
info = api.model_info(model_name)
# Try to load config for more details
try:
config = AutoConfig.from_pretrained(model_name)
config_dict = config.to_dict()
except:
config_dict = {}
return {
"model_id": info.id,
"author": info.author,
"sha": info.sha,
"pipeline_tag": info.pipeline_tag,
"library_name": info.library_name,
"downloads": getattr(info, "downloads", 0),
"likes": getattr(info, "likes", 0),
"tags": info.tags or [],
"siblings": [s.rfilename for s in info.siblings] if info.siblings else [],
"config": config_dict,
"hidden_size": config_dict.get("hidden_size"),
"num_hidden_layers": config_dict.get("num_hidden_layers"),
"num_attention_heads": config_dict.get("num_attention_heads"),
"intermediate_size": config_dict.get("intermediate_size"),
"vocab_size": config_dict.get("vocab_size"),
"model_type": config_dict.get("model_type"),
"architectures": config_dict.get("architectures", []),
}
except Exception as e:
logger.error(f"Error getting model info for {model_name}: {e}")
return {"error": str(e)}
def check_model_compatibility(model_name: str, task_type: str) -> Tuple[bool, List[str]]:
"""Check if model is compatible with a task type."""
issues = []
try:
config = AutoConfig.from_pretrained(model_name)
architectures = config.architectures or []
model_type = config.model_type or ""
# Check if architecture supports task
compatible = True
if task_type == "causal-lm":
causal_archs = ["GPT", "LLaMA", "Mistral", "Falcon", "Qwen", "Phi", "OPT", "Bloom", "CausalLM"]
if not any(arch in arch for arch in architectures for arch in causal_archs):
if model_type not in ["gpt2", "llama", "mistral", "falcon", "qwen", "phi"]:
issues.append("Model may not support causal language modeling")
elif task_type == "seq2seq":
seq2seq_archs = ["T5", "BART", "Pegasus", "MT5", "EncoderDecoderModel"]
if not any(arch in arch for arch in architectures for arch in seq2seq_archs):
issues.append("Model may not support seq2seq tasks")
elif task_type == "token-classification":
if not any("TokenClassification" in arch for arch in architectures):
issues.append("Model may not support token classification")
elif task_type == "text-classification":
if not any("Classification" in arch for arch in architectures):
issues.append("Model may not support text classification")
elif task_type == "question-answering":
qa_archs = ["QuestionAnswering", "BertForQA"]
if not any(arch in arch for arch in architectures for arch in qa_archs):
issues.append("Model may not support question answering")
return len(issues) == 0, issues
except Exception as e:
return False, [f"Error checking compatibility: {str(e)}"]
def apply_peft(
model: PreTrainedModel,
task_type: str,
lora_r: int = 8,
lora_alpha: int = 32,
lora_dropout: float = 0.1,
target_modules: Optional[List[str]] = None,
) -> Tuple[PreTrainedModel, Dict[str, Any]]:
"""Apply PEFT/LoRA to a model."""
try:
# Prepare model for training
model = prepare_model_for_kbit_training(model)
# Get PEFT task type
peft_task_type = PEFT_TASK_TYPES.get(task_type, TaskType.CAUSAL_LM)
# Auto-detect target modules if not specified
if target_modules is None:
model_type = getattr(model.config, "model_type", "").lower()
if "llama" in model_type or "mistral" in model_type:
target_modules = ["q_proj", "v_proj", "k_proj", "o_proj", "gate_proj", "up_proj", "down_proj"]
elif "gpt" in model_type:
target_modules = ["c_attn", "c_proj"]
elif "bert" in model_type or "roberta" in model_type:
target_modules = ["query", "value", "key", "dense"]
else:
target_modules = ["q_proj", "v_proj"]
# Create LoRA config
lora_config = LoraConfig(
r=lora_r,
lora_alpha=lora_alpha,
lora_dropout=lora_dropout,
bias="none",
task_type=peft_task_type,
target_modules=target_modules,
)
# Apply LoRA
model = get_peft_model(model, lora_config)
# Get trainable params info
trainable_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
all_params = sum(p.numel() for p in model.parameters())
info = {
"trainable_params": trainable_params,
"all_params": all_params,
"trainable_percentage": 100 * trainable_params / all_params,
"lora_r": lora_r,
"lora_alpha": lora_alpha,
"target_modules": target_modules,
}
return model, info
except Exception as e:
logger.error(f"Error applying PEFT: {e}")
return model, {"error": str(e)}
def estimate_parameters(model_name: str) -> Dict[str, Any]:
"""Estimate model parameters without loading."""
try:
config = AutoConfig.from_pretrained(model_name)
hidden_size = getattr(config, "hidden_size", 768)
num_layers = getattr(config, "num_hidden_layers", 12)
num_heads = getattr(config, "num_attention_heads", 12)
vocab_size = getattr(config, "vocab_size", 30522)
intermediate_size = getattr(config, "intermediate_size", hidden_size * 4)
# Rough estimation formulas
# Embedding params
embedding_params = vocab_size * hidden_size
# Attention params per layer (Q, K, V, O projections)
attention_params = 4 * hidden_size * hidden_size * num_layers
# FFN params per layer
ffn_params = (hidden_size * intermediate_size + intermediate_size * hidden_size) * num_layers
# Layer norm params
layernorm_params = 2 * hidden_size * num_layers
total_params = embedding_params + attention_params + ffn_params + layernorm_params
return {
"estimated_params": total_params,
"estimated_params_billions": round(total_params / 1e9, 2),
"hidden_size": hidden_size,
"num_layers": num_layers,
"num_heads": num_heads,
"vocab_size": vocab_size,
"model_size_mb": round(total_params * 4 / (1024 * 1024), 2), # FP32
"model_size_mb_fp16": round(total_params * 2 / (1024 * 1024), 2), # FP16
}
except Exception as e:
logger.warning(f"Could not estimate parameters: {e}")
return {
"estimated_params": 0,
"estimated_params_billions": 0,
"error": str(e),
}
def get_recommended_settings(model_name: str, task_type: str) -> Dict[str, Any]:
"""Get recommended training settings for a model."""
info = estimate_parameters(model_name)
params_b = info.get("estimated_params_billions", 0.1)
# Base recommendations
settings = {
"batch_size": 1,
"gradient_accumulation_steps": 1,
"learning_rate": "5e-5",
"epochs": 3,
"max_length": 512,
"use_peft": False,
"lora_r": 8,
"warmup_ratio": 0.1,
}
# Adjust based on model size
if params_b > 7: # > 7B parameters
settings["batch_size"] = 1
settings["gradient_accumulation_steps"] = 8
settings["learning_rate"] = "1e-5"
settings["use_peft"] = True
settings["lora_r"] = 8
settings["max_length"] = 256
elif params_b > 3: # > 3B parameters
settings["batch_size"] = 1
settings["gradient_accumulation_steps"] = 4
settings["learning_rate"] = "2e-5"
settings["use_peft"] = True
settings["max_length"] = 512
elif params_b > 1: # > 1B parameters
settings["batch_size"] = 2
settings["gradient_accumulation_steps"] = 2
settings["use_peft"] = True
else: # < 1B parameters
settings["batch_size"] = 4
settings["gradient_accumulation_steps"] = 1
settings["use_peft"] = False
# Task-specific adjustments
if task_type == "seq2seq":
settings["max_length"] = 1024
settings["epochs"] = 5
elif task_type == "token-classification":
settings["max_length"] = 128
settings["learning_rate"] = "2e-5"
elif task_type == "text-classification":
settings["epochs"] = 3
settings["learning_rate"] = "3e-5"
elif task_type == "question-answering":
settings["max_length"] = 384
settings["batch_size"] = 8
return settings
def count_parameters(model: PreTrainedModel) -> Dict[str, int]:
"""Count model parameters."""
trainable = sum(p.numel() for p in model.parameters() if p.requires_grad)
total = sum(p.numel() for p in model.parameters())
frozen = total - trainable
return {
"trainable": trainable,
"frozen": frozen,
"total": total,
"trainable_percentage": 100 * trainable / total if total > 0 else 0,
}
def get_model_memory_footprint(model: PreTrainedModel) -> Dict[str, float]:
"""Get model memory footprint in MB."""
param_size = sum(p.numel() * p.element_size() for p in model.parameters())
buffer_size = sum(b.numel() * b.element_size() for b in model.buffers())
return {
"parameters_mb": param_size / (1024 * 1024),
"buffers_mb": buffer_size / (1024 * 1024),
"total_mb": (param_size + buffer_size) / (1024 * 1024),
}
def save_model_with_metadata(
model: PreTrainedModel,
tokenizer: PreTrainedTokenizer,
output_dir: str,
training_config: Dict[str, Any],
metrics: Dict[str, float],
) -> Dict[str, str]:
"""Save model with comprehensive metadata."""
import json
from datetime import datetime
os.makedirs(output_dir, exist_ok=True)
# Save model and tokenizer
model.save_pretrained(output_dir)
tokenizer.save_pretrained(output_dir)
# Get model info
param_info = count_parameters(model)
memory_info = get_model_memory_footprint(model)
# Create comprehensive metadata
metadata = {
"model_name": training_config.get("model_name", "unknown"),
"task_type": training_config.get("task_type", "unknown"),
"training_config": training_config,
"metrics": metrics,
"parameter_info": param_info,
"memory_info": memory_info,
"created_at": datetime.utcnow().isoformat(),
"transformers_version": __import__("transformers").__version__,
"torch_version": __import__("torch").__version__,
"python_version": __import__("sys").version,
}
# Save metadata
metadata_path = os.path.join(output_dir, "training_metadata.json")
with open(metadata_path, "w") as f:
json.dump(metadata, f, indent=2)
# Create model card
model_card = create_model_card(training_config, metrics, param_info)
model_card_path = os.path.join(output_dir, "README.md")
with open(model_card_path, "w") as f:
f.write(model_card)
return {
"output_dir": output_dir,
"model_path": output_dir,
"metadata_path": metadata_path,
"model_card_path": model_card_path,
}
def create_model_card(
config: Dict[str, Any],
metrics: Dict[str, float],
param_info: Dict[str, int],
) -> str:
"""Create a model card README."""
model_name = config.get("model_name", "unknown")
task_type = config.get("task_type", "unknown")
metrics_str = "\n".join([f"- {k}: {v:.4f}" if isinstance(v, float) else f"- {k}: {v}" for k, v in metrics.items()]) if metrics else "- No metrics available"
return f"""# {model_name} - Fine-tuned
## Model Details
- **Base Model:** {model_name}
- **Task:** {task_type}
- **Total Parameters:** {param_info.get('total', 0):,}
- **Trainable Parameters:** {param_info.get('trainable', 0):,}
## Training Configuration
```json
{json.dumps(config, indent=2)}
```
## Training Metrics
{metrics_str}
## Usage
```python
from transformers import AutoModel, AutoTokenizer
model = AutoModel.from_pretrained("path/to/model")
tokenizer = AutoTokenizer.from_pretrained("path/to/model")
```
## License
Please refer to the original model's license.
## Training Framework
This model was trained using the Universal Model Trainer.
""" |