File size: 17,391 Bytes
1741386 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 | """
Training Utilities - Helper functions for model training
"""
import logging
import os
import json
import hashlib
from typing import Dict, Any, List, Optional, Tuple
from datetime import datetime
import torch
from transformers import (
AutoTokenizer,
AutoModelForCausalLM,
AutoModelForSeq2SeqLM,
AutoModelForTokenClassification,
AutoModelForQuestionAnswering,
AutoModelForSequenceClassification,
AutoConfig,
TrainingArguments,
Trainer,
DataCollatorForLanguageModeling,
DataCollatorForSeq2Seq,
DataCollatorForTokenClassification,
)
from peft import LoraConfig, get_peft_model, prepare_model_for_kbit_training
from datasets import Dataset
import numpy as np
logger = logging.getLogger(__name__)
def get_model_class_for_task(task_type: str):
"""Get the appropriate model class for a task type."""
model_map = {
"causal-lm": AutoModelForCausalLM,
"seq2seq": AutoModelForSeq2SeqLM,
"token-classification": AutoModelForTokenClassification,
"question-answering": AutoModelForQuestionAnswering,
"text-classification": AutoModelForSequenceClassification,
}
if task_type not in model_map:
raise ValueError(f"Unknown task type: {task_type}")
return model_map[task_type]
def compute_model_hash(model_path: str) -> str:
"""Compute a hash of model configuration for tracking."""
config_path = os.path.join(model_path, "config.json")
if os.path.exists(config_path):
with open(config_path, "rb") as f:
return hashlib.md5(f.read()).hexdigest()[:12]
return "unknown"
def estimate_memory_requirements(
model_name: str,
task_type: str,
batch_size: int = 1,
max_length: int = 512,
use_peft: bool = False
) -> Dict[str, float]:
"""Estimate memory requirements for training."""
try:
config = AutoConfig.from_pretrained(model_name)
# Estimate parameters
if hasattr(config, "hidden_size"):
hidden = config.hidden_size
elif hasattr(config, "n_embd"):
hidden = config.n_embd
else:
hidden = 768
if hasattr(config, "num_hidden_layers"):
layers = config.num_hidden_layers
elif hasattr(config, "n_layer"):
layers = config.n_layer
else:
layers = 12
# Rough parameter estimation
params = hidden ** 2 * layers * 12 # Very rough estimate
params_billion = params / 1e9
# Memory estimation (very approximate)
# FP32: 4 bytes per param, FP16: 2 bytes
model_memory_fp32 = params_billion * 4 # GB
model_memory_fp16 = params_billion * 2 # GB
# Gradients (same as model)
gradients_memory = model_memory_fp16
# Optimizer states (Adam: 2x model size)
optimizer_memory = model_memory_fp16 * 2
# Activations depend on batch size and sequence length
activation_memory = (batch_size * max_length * hidden * 4) / 1e9 # Rough estimate
# Total
if use_peft:
# PEFT reduces memory significantly
total_fp16 = (model_memory_fp16 * 0.1) + gradients_memory + optimizer_memory * 0.1 + activation_memory
else:
total_fp16 = model_memory_fp16 + gradients_memory + optimizer_memory + activation_memory
return {
"estimated_params_billion": round(params_billion, 2),
"model_memory_gb": round(model_memory_fp16, 2),
"optimizer_memory_gb": round(optimizer_memory, 2),
"activation_memory_gb": round(activation_memory, 2),
"total_memory_gb": round(total_fp16, 2),
"recommended_memory_gb": round(total_fp16 * 1.5, 2),
"can_run_on_cpu": total_fp16 < 8,
"recommended_hardware": "gpu" if total_fp16 > 4 else "cpu"
}
except Exception as e:
logger.warning(f"Could not estimate memory: {e}")
return {
"estimated_params_billion": 0.1,
"model_memory_gb": 0.5,
"optimizer_memory_gb": 1.0,
"activation_memory_gb": 0.5,
"total_memory_gb": 2.0,
"recommended_memory_gb": 4.0,
"can_run_on_cpu": True,
"recommended_hardware": "cpu"
}
def get_available_hardware() -> List[Dict[str, Any]]:
"""Get available hardware options."""
hardware = [
{"id": "cpu-basic", "name": "CPU Basic", "memory_gb": 16, "gpu": False, "cost": "Free"},
{"id": "cpu-upgrade", "name": "CPU Upgrade", "memory_gb": 32, "gpu": False, "cost": "Low"},
{"id": "t4-small", "name": "T4 Small", "memory_gb": 16, "gpu": True, "gpu_memory_gb": 16, "cost": "Medium"},
{"id": "t4-medium", "name": "T4 Medium", "memory_gb": 32, "gpu": True, "gpu_memory_gb": 16, "cost": "Medium"},
{"id": "l4x1", "name": "L4 x1", "memory_gb": 32, "gpu": True, "gpu_memory_gb": 24, "cost": "High"},
{"id": "l4x4", "name": "L4 x4", "memory_gb": 96, "gpu": True, "gpu_memory_gb": 96, "cost": "Very High"},
{"id": "a10g-small", "name": "A10G Small", "memory_gb": 24, "gpu": True, "gpu_memory_gb": 24, "cost": "High"},
{"id": "a10g-large", "name": "A10G Large", "memory_gb": 48, "gpu": True, "gpu_memory_gb": 48, "cost": "Very High"},
{"id": "a100-large", "name": "A100 Large", "memory_gb": 80, "gpu": True, "gpu_memory_gb": 80, "cost": "Premium"},
]
# Check what's actually available
if torch.cuda.is_available():
gpu_count = torch.cuda.device_count()
gpu_name = torch.cuda.get_device_name(0) if gpu_count > 0 else "Unknown"
gpu_memory = torch.cuda.get_device_properties(0).total_memory / 1e9 if gpu_count > 0 else 0
return hardware, {
"cuda_available": True,
"gpu_count": gpu_count,
"gpu_name": gpu_name,
"gpu_memory_gb": round(gpu_memory, 1)
}
else:
return hardware, {
"cuda_available": False,
"gpu_count": 0,
"gpu_name": None,
"gpu_memory_gb": 0
}
def get_training_args(
output_dir: str,
config: Dict[str, Any],
task_type: str
) -> TrainingArguments:
"""Create TrainingArguments from config."""
# Base arguments
args = {
"output_dir": output_dir,
"overwrite_output_dir": True,
# Training
"num_train_epochs": config.get("epochs", 3),
"per_device_train_batch_size": config.get("batch_size", 1),
"per_device_eval_batch_size": config.get("batch_size", 1),
"gradient_accumulation_steps": config.get("gradient_accumulation_steps", 1),
# Learning rate
"learning_rate": config.get("learning_rate", 5e-5),
"weight_decay": config.get("weight_decay", 0.01),
"warmup_steps": config.get("warmup_steps", 100),
"lr_scheduler_type": config.get("lr_scheduler_type", "cosine"),
# Logging
"logging_dir": os.path.join(output_dir, "logs"),
"logging_steps": config.get("logging_steps", 10),
"save_steps": config.get("save_steps", 500),
"save_total_limit": config.get("save_total_limit", 3),
# Evaluation
"evaluation_strategy": "steps" if config.get("eval_steps") else "no",
"eval_steps": config.get("eval_steps", 500),
# Optimization
"fp16": config.get("fp16", True) and torch.cuda.is_available(),
"bf16": config.get("bf16", False) and torch.cuda.is_bf16_supported(),
# Misc
"dataloader_num_workers": config.get("dataloader_num_workers", 0),
"dataloader_pin_memory": config.get("pin_memory", True) and torch.cuda.is_available(),
"gradient_checkpointing": config.get("gradient_checkpointing", False),
# Reporting
"report_to": config.get("report_to", ["none"]),
# Seed
"seed": config.get("seed", 42),
}
# Task-specific adjustments
if task_type == "causal-lm":
args["max_steps"] = config.get("max_steps", -1)
if config.get("max_length"):
args["max_length"] = config["max_length"]
elif task_type == "seq2seq":
args["predict_with_generate"] = config.get("predict_with_generate", False)
args["generation_max_length"] = config.get("generation_max_length", 128)
args["generation_num_beams"] = config.get("generation_num_beams", 4)
elif task_type == "token-classification":
args["label_names"] = config.get("label_names", [])
# DeepSpeed config if enabled
if config.get("deepspeed_config"):
args["deepspeed"] = config["deepspeed_config"]
return TrainingArguments(**args)
def get_peft_config(config: Dict[str, Any]) -> Optional[LoraConfig]:
"""Create PEFT/LoRA config if enabled."""
if not config.get("use_peft", False):
return None
peft_config = LoraConfig(
r=config.get("lora_r", 8),
lora_alpha=config.get("lora_alpha", 32),
lora_dropout=config.get("lora_dropout", 0.1),
bias=config.get("lora_bias", "none"),
task_type=config.get("peft_task_type", "CAUSAL_LM"),
target_modules=config.get("lora_target_modules", None),
)
return peft_config
def get_data_collator(
tokenizer: Any,
task_type: str,
config: Dict[str, Any]
) -> Any:
"""Get appropriate data collator for task type."""
if task_type == "causal-lm":
return DataCollatorForLanguageModeling(
tokenizer=tokenizer,
mlm=False,
pad_to_multiple_of=config.get("pad_to_multiple_of", 8)
)
elif task_type == "seq2seq":
return DataCollatorForSeq2Seq(
tokenizer=tokenizer,
model=None,
padding=config.get("padding", "max_length"),
max_length=config.get("max_length", 512),
pad_to_multiple_of=config.get("pad_to_multiple_of", 8)
)
elif task_type == "token-classification":
return DataCollatorForTokenClassification(
tokenizer=tokenizer,
padding=config.get("padding", "max_length"),
max_length=config.get("max_length", 512),
pad_to_multiple_of=config.get("pad_to_multiple_of", 8)
)
elif task_type == "question-answering":
return DataCollatorForSeq2Seq(
tokenizer=tokenizer,
model=None,
padding=config.get("padding", "max_length"),
max_length=config.get("max_length", 384),
)
elif task_type == "text-classification":
from transformers import DataCollatorWithPadding
return DataCollatorWithPadding(
tokenizer=tokenizer,
padding=config.get("padding", "max_length"),
max_length=config.get("max_length", 512),
)
else:
logger.warning(f"Unknown task type {task_type}, using default collator")
from transformers import DataCollatorWithPadding
return DataCollatorWithPadding(tokenizer=tokenizer)
def compute_metrics_factory(task_type: str, tokenizer: Any = None):
"""Factory for creating compute_metrics function."""
if task_type == "causal-lm":
def compute_metrics(eval_preds):
"""Compute perplexity for language modeling."""
logits, labels = eval_preds
# Shift for causal LM
shift_logits = logits[..., :-1, :].contiguous()
shift_labels = labels[..., 1:].contiguous()
loss_fct = torch.nn.CrossEntropyLoss(reduction='mean')
loss = loss_fct(
shift_logits.view(-1, shift_logits.size(-1)),
shift_labels.view(-1)
)
perplexity = torch.exp(loss)
return {
"perplexity": perplexity.item(),
"loss": loss.item()
}
return compute_metrics
elif task_type == "seq2seq":
def compute_metrics(eval_preds):
"""Compute ROUGE scores for summarization."""
from evaluate import load
rouge = load("rouge")
predictions, labels = eval_preds
decoded_preds = tokenizer.batch_decode(predictions, skip_special_tokens=True)
labels = np.where(labels != -100, labels, tokenizer.pad_token_id)
decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)
result = rouge.compute(
predictions=decoded_preds,
references=decoded_labels,
use_stemmer=True
)
return {k: round(v * 100, 4) for k, v in result.items()}
return compute_metrics
elif task_type == "token-classification":
def compute_metrics(eval_preds):
"""Compute precision, recall, F1 for NER."""
from evaluate import load
seqeval = load("seqeval")
predictions, labels = eval_preds
predictions = np.argmax(predictions, axis=2)
# Remove ignored index
true_predictions = [
[p for (p, l) in zip(prediction, label) if l != -100]
for prediction, label in zip(predictions, labels)
]
true_labels = [
[l for (p, l) in zip(prediction, label) if l != -100]
for prediction, label in zip(predictions, labels)
]
results = seqeval.compute(predictions=true_predictions, references=true_labels)
return {
"precision": results["overall_precision"],
"recall": results["overall_recall"],
"f1": results["overall_f1"],
"accuracy": results["overall_accuracy"]
}
return compute_metrics
elif task_type == "text-classification":
def compute_metrics(eval_preds):
"""Compute accuracy and F1 for classification."""
from sklearn.metrics import accuracy_score, f1_score
predictions, labels = eval_preds
predictions = np.argmax(predictions, axis=1)
return {
"accuracy": accuracy_score(labels, predictions),
"f1": f1_score(labels, predictions, average="weighted")
}
return compute_metrics
elif task_type == "question-answering":
def compute_metrics(eval_preds):
"""Compute SQuAD metrics."""
from evaluate import load
squad_metric = load("squad_v2")
predictions, labels = eval_preds
# Process predictions and labels for QA
# This is simplified - real implementation needs proper post-processing
return {
"exact_match": 0.0,
"f1": 0.0
}
return compute_metrics
else:
def compute_metrics(eval_preds):
return {}
return compute_metrics
def save_training_artifacts(
output_dir: str,
model: Any,
tokenizer: Any,
config: Dict[str, Any],
metrics: Dict[str, float]
) -> Dict[str, str]:
"""Save training artifacts."""
os.makedirs(output_dir, exist_ok=True)
saved_files = []
# Save model
model.save_pretrained(output_dir)
saved_files.append("model")
# Save tokenizer
tokenizer.save_pretrained(output_dir)
saved_files.append("tokenizer")
# Save config
with open(os.path.join(output_dir, "training_config.json"), "w") as f:
json.dump(config, f, indent=2)
saved_files.append("training_config.json")
# Save metrics
with open(os.path.join(output_dir, "metrics.json"), "w") as f:
json.dump(metrics, f, indent=2)
saved_files.append("metrics.json")
# Create README
readme_content = f"""# Model Fine-tuned with Universal Model Trainer
## Model Details
- Base Model: {config.get('model_name', 'Unknown')}
- Task: {config.get('task_type', 'Unknown')}
- Training Date: {datetime.utcnow().isoformat()}
## Training Configuration
- Epochs: {config.get('epochs', 'Unknown')}
- Batch Size: {config.get('batch_size', 'Unknown')}
- Learning Rate: {config.get('learning_rate', 'Unknown')}
- PEFT/LoRA: {'Yes' if config.get('use_peft') else 'No'}
## Metrics
```
{json.dumps(metrics, indent=2)}
```
## Usage
```python
from transformers import AutoModel, AutoTokenizer
model = AutoModel.from_pretrained("path/to/model")
tokenizer = AutoTokenizer.from_pretrained("path/to/model")
```
"""
with open(os.path.join(output_dir, "README.md"), "w") as f:
f.write(readme_content)
saved_files.append("README.md")
return {
"output_dir": output_dir,
"saved_files": saved_files,
"total_size": sum(os.path.getsize(os.path.join(output_dir, f)) for f in os.listdir(output_dir) if os.path.isfile(os.path.join(output_dir, f)))
}
def generate_job_id(config: Dict[str, Any]) -> str:
"""Generate unique job ID."""
import uuid
return f"train_{config['task_type']}_{uuid.uuid4().hex[:8]}" |