Spaces:
Sleeping
Sleeping
File size: 16,561 Bytes
33d4721 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 |
import os
import shlex
import torch
from autotrain import logger
from autotrain.trainers.clm.params import LLMTrainingParams
from autotrain.trainers.extractive_question_answering.params import ExtractiveQuestionAnsweringParams
from autotrain.trainers.generic.params import GenericParams
from autotrain.trainers.image_classification.params import ImageClassificationParams
from autotrain.trainers.image_regression.params import ImageRegressionParams
from autotrain.trainers.object_detection.params import ObjectDetectionParams
from autotrain.trainers.sent_transformers.params import SentenceTransformersParams
from autotrain.trainers.seq2seq.params import Seq2SeqParams
from autotrain.trainers.tabular.params import TabularParams
from autotrain.trainers.text_classification.params import TextClassificationParams
from autotrain.trainers.text_regression.params import TextRegressionParams
from autotrain.trainers.token_classification.params import TokenClassificationParams
from autotrain.trainers.vlm.params import VLMTrainingParams
CPU_COMMAND = [
"accelerate",
"launch",
"--cpu",
]
SINGLE_GPU_COMMAND = [
"accelerate",
"launch",
"--num_machines",
"1",
"--num_processes",
"1",
]
def get_accelerate_command(num_gpus, gradient_accumulation_steps=1, distributed_backend=None):
"""
Generates the appropriate command to launch a training job using the `accelerate` library based on the number of GPUs
and the specified distributed backend.
Args:
num_gpus (int): The number of GPUs available for training. If 0, training will be forced on CPU.
gradient_accumulation_steps (int, optional): The number of gradient accumulation steps. Defaults to 1.
distributed_backend (str, optional): The distributed backend to use. Can be "ddp" (Distributed Data Parallel),
"deepspeed", or None. Defaults to None.
Returns:
list or str: The command to be executed as a list of strings. If no GPU is found, returns a CPU command string.
If a single GPU is found, returns a single GPU command string. Otherwise, returns a list of
command arguments for multi-GPU or DeepSpeed training.
Raises:
ValueError: If an unsupported distributed backend is specified.
"""
if num_gpus == 0:
logger.warning("No GPU found. Forcing training on CPU. This will be super slow!")
return CPU_COMMAND
if num_gpus == 1:
return SINGLE_GPU_COMMAND
if distributed_backend in ("ddp", None):
return [
"accelerate",
"launch",
"--multi_gpu",
"--num_machines",
"1",
"--num_processes",
str(num_gpus),
]
elif distributed_backend == "deepspeed":
return [
"accelerate",
"launch",
"--use_deepspeed",
"--zero_stage",
"3",
"--offload_optimizer_device",
"none",
"--offload_param_device",
"none",
"--zero3_save_16bit_model",
"true",
"--zero3_init_flag",
"true",
"--deepspeed_multinode_launcher",
"standard",
"--gradient_accumulation_steps",
str(gradient_accumulation_steps),
]
else:
raise ValueError("Unsupported distributed backend")
def launch_command(params):
"""
Launches the appropriate training command based on the type of training parameters provided.
Args:
params (object): An instance of one of the training parameter classes. This can be one of the following:
- LLMTrainingParams
- GenericParams
- TabularParams
- TextClassificationParams
- TextRegressionParams
- SentenceTransformersParams
- ExtractiveQuestionAnsweringParams
- TokenClassificationParams
- ImageClassificationParams
- ObjectDetectionParams
- ImageRegressionParams
- Seq2SeqParams
- VLMTrainingParams
Returns:
list: A list of command line arguments to be executed for training.
Raises:
ValueError: If the provided params type is unsupported.
"""
params.project_name = shlex.split(params.project_name)[0]
cuda_available = torch.cuda.is_available()
mps_available = torch.backends.mps.is_available()
if cuda_available:
num_gpus = torch.cuda.device_count()
elif mps_available:
num_gpus = 1
else:
num_gpus = 0
if isinstance(params, LLMTrainingParams):
cmd = get_accelerate_command(num_gpus, params.gradient_accumulation, params.distributed_backend)
if num_gpus > 0:
cmd.append("--mixed_precision")
if params.mixed_precision == "fp16":
cmd.append("fp16")
elif params.mixed_precision == "bf16":
cmd.append("bf16")
else:
cmd.append("no")
cmd.extend(
[
"-m",
"autotrain.trainers.clm",
"--training_config",
os.path.join(params.project_name, "training_params.json"),
]
)
elif isinstance(params, GenericParams):
cmd = [
"python",
"-m",
"autotrain.trainers.generic",
"--config",
os.path.join(params.project_name, "training_params.json"),
]
elif isinstance(params, TabularParams):
cmd = [
"python",
"-m",
"autotrain.trainers.tabular",
"--training_config",
os.path.join(params.project_name, "training_params.json"),
]
elif (
isinstance(params, TextClassificationParams)
or isinstance(params, TextRegressionParams)
or isinstance(params, SentenceTransformersParams)
or isinstance(params, ExtractiveQuestionAnsweringParams)
):
if num_gpus == 0:
cmd = [
"accelerate",
"launch",
"--cpu",
]
elif num_gpus == 1:
cmd = [
"accelerate",
"launch",
"--num_machines",
"1",
"--num_processes",
"1",
]
else:
cmd = [
"accelerate",
"launch",
"--multi_gpu",
"--num_machines",
"1",
"--num_processes",
str(num_gpus),
]
if num_gpus > 0:
cmd.append("--mixed_precision")
if params.mixed_precision == "fp16":
cmd.append("fp16")
elif params.mixed_precision == "bf16":
cmd.append("bf16")
else:
cmd.append("no")
if isinstance(params, TextRegressionParams):
cmd.extend(
[
"-m",
"autotrain.trainers.text_regression",
"--training_config",
os.path.join(params.project_name, "training_params.json"),
]
)
elif isinstance(params, SentenceTransformersParams):
cmd.extend(
[
"-m",
"autotrain.trainers.sent_transformers",
"--training_config",
os.path.join(params.project_name, "training_params.json"),
]
)
elif isinstance(params, ExtractiveQuestionAnsweringParams):
cmd.extend(
[
"-m",
"autotrain.trainers.extractive_question_answering",
"--training_config",
os.path.join(params.project_name, "training_params.json"),
]
)
else:
cmd.extend(
[
"-m",
"autotrain.trainers.text_classification",
"--training_config",
os.path.join(params.project_name, "training_params.json"),
]
)
elif isinstance(params, TokenClassificationParams):
if num_gpus == 0:
cmd = [
"accelerate",
"launch",
"--cpu",
]
elif num_gpus == 1:
cmd = [
"accelerate",
"launch",
"--num_machines",
"1",
"--num_processes",
"1",
]
else:
cmd = [
"accelerate",
"launch",
"--multi_gpu",
"--num_machines",
"1",
"--num_processes",
str(num_gpus),
]
if num_gpus > 0:
cmd.append("--mixed_precision")
if params.mixed_precision == "fp16":
cmd.append("fp16")
elif params.mixed_precision == "bf16":
cmd.append("bf16")
else:
cmd.append("no")
cmd.extend(
[
"-m",
"autotrain.trainers.token_classification",
"--training_config",
os.path.join(params.project_name, "training_params.json"),
]
)
elif (
isinstance(params, ImageClassificationParams)
or isinstance(params, ObjectDetectionParams)
or isinstance(params, ImageRegressionParams)
):
if num_gpus == 0:
cmd = [
"accelerate",
"launch",
"--cpu",
]
elif num_gpus == 1:
cmd = [
"accelerate",
"launch",
"--num_machines",
"1",
"--num_processes",
"1",
]
else:
cmd = [
"accelerate",
"launch",
"--multi_gpu",
"--num_machines",
"1",
"--num_processes",
str(num_gpus),
]
if num_gpus > 0:
cmd.append("--mixed_precision")
if params.mixed_precision == "fp16":
cmd.append("fp16")
elif params.mixed_precision == "bf16":
cmd.append("bf16")
else:
cmd.append("no")
if isinstance(params, ObjectDetectionParams):
cmd.extend(
[
"-m",
"autotrain.trainers.object_detection",
"--training_config",
os.path.join(params.project_name, "training_params.json"),
]
)
elif isinstance(params, ImageRegressionParams):
cmd.extend(
[
"-m",
"autotrain.trainers.image_regression",
"--training_config",
os.path.join(params.project_name, "training_params.json"),
]
)
else:
cmd.extend(
[
"-m",
"autotrain.trainers.image_classification",
"--training_config",
os.path.join(params.project_name, "training_params.json"),
]
)
elif isinstance(params, Seq2SeqParams):
if num_gpus == 0:
logger.warning("No GPU found. Forcing training on CPU. This will be super slow!")
cmd = [
"accelerate",
"launch",
"--cpu",
]
elif num_gpus == 1:
cmd = [
"accelerate",
"launch",
"--num_machines",
"1",
"--num_processes",
"1",
]
elif num_gpus == 2:
cmd = [
"accelerate",
"launch",
"--multi_gpu",
"--num_machines",
"1",
"--num_processes",
"2",
]
else:
if params.quantization in ("int8", "int4") and params.peft and params.mixed_precision == "bf16":
cmd = [
"accelerate",
"launch",
"--multi_gpu",
"--num_machines",
"1",
"--num_processes",
str(num_gpus),
]
else:
cmd = [
"accelerate",
"launch",
"--use_deepspeed",
"--zero_stage",
"3",
"--offload_optimizer_device",
"none",
"--offload_param_device",
"none",
"--zero3_save_16bit_model",
"true",
"--zero3_init_flag",
"true",
"--deepspeed_multinode_launcher",
"standard",
"--gradient_accumulation_steps",
str(params.gradient_accumulation),
]
if num_gpus > 0:
cmd.append("--mixed_precision")
if params.mixed_precision == "fp16":
cmd.append("fp16")
elif params.mixed_precision == "bf16":
cmd.append("bf16")
else:
cmd.append("no")
cmd.extend(
[
"-m",
"autotrain.trainers.seq2seq",
"--training_config",
os.path.join(params.project_name, "training_params.json"),
]
)
elif isinstance(params, VLMTrainingParams):
if num_gpus == 0:
logger.warning("No GPU found. Forcing training on CPU. This will be super slow!")
cmd = [
"accelerate",
"launch",
"--cpu",
]
elif num_gpus == 1:
cmd = [
"accelerate",
"launch",
"--num_machines",
"1",
"--num_processes",
"1",
]
elif num_gpus == 2:
cmd = [
"accelerate",
"launch",
"--multi_gpu",
"--num_machines",
"1",
"--num_processes",
"2",
]
else:
if params.quantization in ("int8", "int4") and params.peft and params.mixed_precision == "bf16":
cmd = [
"accelerate",
"launch",
"--multi_gpu",
"--num_machines",
"1",
"--num_processes",
str(num_gpus),
]
else:
cmd = [
"accelerate",
"launch",
"--use_deepspeed",
"--zero_stage",
"3",
"--offload_optimizer_device",
"none",
"--offload_param_device",
"none",
"--zero3_save_16bit_model",
"true",
"--zero3_init_flag",
"true",
"--deepspeed_multinode_launcher",
"standard",
"--gradient_accumulation_steps",
str(params.gradient_accumulation),
]
if num_gpus > 0:
cmd.append("--mixed_precision")
if params.mixed_precision == "fp16":
cmd.append("fp16")
elif params.mixed_precision == "bf16":
cmd.append("bf16")
else:
cmd.append("no")
cmd.extend(
[
"-m",
"autotrain.trainers.vlm",
"--training_config",
os.path.join(params.project_name, "training_params.json"),
]
)
else:
raise ValueError("Unsupported params type")
logger.info(cmd)
logger.info(params)
return cmd
|