File size: 36,206 Bytes
9c6594c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 |
# 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.
import copy
import functools
import os
import shutil
import warnings
from collections import defaultdict
from contextlib import nullcontext
from pathlib import Path
from typing import Callable
import torch
from ..logging import get_logger
from .constants import FSDP_MODEL_NAME, OPTIMIZER_NAME, SAFE_WEIGHTS_NAME, WEIGHTS_NAME
from .dataclasses import get_module_class_from_name
from .modeling import get_non_persistent_buffers, is_peft_model
from .other import get_module_children_bottom_up, is_compiled_module, save
from .versions import is_torch_version
logger = get_logger(__name__)
def enable_fsdp_ram_efficient_loading():
"""
Enables RAM efficient loading of Hugging Face models for FSDP in the environment.
"""
# Sets values for `transformers.modeling_utils.is_fsdp_enabled`
if "ACCELERATE_USE_FSDP" not in os.environ:
os.environ["ACCELERATE_USE_FSDP"] = "True"
os.environ["FSDP_CPU_RAM_EFFICIENT_LOADING"] = "True"
def disable_fsdp_ram_efficient_loading():
"""
Disables RAM efficient loading of Hugging Face models for FSDP in the environment.
"""
os.environ["FSDP_CPU_RAM_EFFICIENT_LOADING"] = "False"
def _get_model_state_dict(model, adapter_only=False, sd_options=None):
if adapter_only and is_peft_model(model):
from peft import get_peft_model_state_dict
return get_peft_model_state_dict(model, adapter_name=model.active_adapter)
# Invariant: `sd_options` is not None only for FSDP2
if sd_options is not None:
from torch.distributed.checkpoint.state_dict import get_model_state_dict
return get_model_state_dict(model, options=sd_options)
else:
return model.state_dict()
def _set_model_state_dict(model, state_dict, adapter_only=False, sd_options=None):
if adapter_only and is_peft_model(model):
from peft import set_peft_model_state_dict
return set_peft_model_state_dict(model, state_dict, adapter_name=model.active_adapter)
# Invariant: `sd_options` is not None only for FSDP2
if sd_options is not None:
from torch.distributed.checkpoint.state_dict import set_model_state_dict
return set_model_state_dict(model, state_dict, options=sd_options)
else:
return model.load_state_dict(state_dict)
def _prepare_sd_options(fsdp_plugin):
sd_options = None
# we use this only for FSDP2, as it requires torch >= 2.6.0 and this api requires torch >= 2.2.0
if fsdp_plugin.fsdp_version == 2:
from torch.distributed.checkpoint.state_dict import StateDictOptions
from torch.distributed.fsdp.fully_sharded_data_parallel import StateDictType
sd_options = StateDictOptions(
full_state_dict=fsdp_plugin.state_dict_type == StateDictType.FULL_STATE_DICT,
cpu_offload=getattr(fsdp_plugin.state_dict_config, "offload_to_cpu", False),
broadcast_from_rank0=getattr(fsdp_plugin.state_dict_config, "rank0_only", False),
)
return sd_options
def save_fsdp_model(fsdp_plugin, accelerator, model, output_dir, model_index=0, adapter_only=False):
# Note: We import here to reduce import time from general modules, and isolate outside dependencies
import torch.distributed.checkpoint as dist_cp
from torch.distributed.checkpoint.default_planner import DefaultSavePlanner
from torch.distributed.fsdp.fully_sharded_data_parallel import FullyShardedDataParallel as FSDP
from torch.distributed.fsdp.fully_sharded_data_parallel import StateDictType
os.makedirs(output_dir, exist_ok=True)
if fsdp_plugin.state_dict_type == StateDictType.FULL_STATE_DICT:
# FSDP raises error when single GPU is used with `offload_to_cpu=True` for FULL_STATE_DICT
# so, only enable it when num_processes>1
is_multi_process = accelerator.num_processes > 1
fsdp_plugin.state_dict_config.offload_to_cpu = is_multi_process
fsdp_plugin.state_dict_config.rank0_only = is_multi_process
ctx = (
FSDP.state_dict_type(
model, fsdp_plugin.state_dict_type, fsdp_plugin.state_dict_config, fsdp_plugin.optim_state_dict_config
)
if fsdp_plugin.fsdp_version == 1
else nullcontext()
)
sd_options = _prepare_sd_options(fsdp_plugin)
with ctx:
state_dict = _get_model_state_dict(model, adapter_only=adapter_only, sd_options=sd_options)
if fsdp_plugin.state_dict_type == StateDictType.FULL_STATE_DICT:
weights_name = f"{FSDP_MODEL_NAME}.bin" if model_index == 0 else f"{FSDP_MODEL_NAME}_{model_index}.bin"
output_model_file = os.path.join(output_dir, weights_name)
if accelerator.process_index == 0:
logger.info(f"Saving model to {output_model_file}")
torch.save(state_dict, output_model_file)
logger.info(f"Model saved to {output_model_file}")
# Invariant: `LOCAL_STATE_DICT` is never possible with `FSDP2`
elif fsdp_plugin.state_dict_type == StateDictType.LOCAL_STATE_DICT:
weights_name = (
f"{FSDP_MODEL_NAME}_rank{accelerator.process_index}.bin"
if model_index == 0
else f"{FSDP_MODEL_NAME}_{model_index}_rank{accelerator.process_index}.bin"
)
output_model_file = os.path.join(output_dir, weights_name)
logger.info(f"Saving model to {output_model_file}")
torch.save(state_dict, output_model_file)
logger.info(f"Model saved to {output_model_file}")
elif fsdp_plugin.state_dict_type == StateDictType.SHARDED_STATE_DICT:
ckpt_dir = os.path.join(output_dir, f"{FSDP_MODEL_NAME}_{model_index}")
os.makedirs(ckpt_dir, exist_ok=True)
logger.info(f"Saving model to {ckpt_dir}")
state_dict = {"model": state_dict}
dist_cp.save(
state_dict=state_dict,
storage_writer=dist_cp.FileSystemWriter(ckpt_dir),
planner=DefaultSavePlanner(),
)
logger.info(f"Model saved to {ckpt_dir}")
def load_fsdp_model(fsdp_plugin, accelerator, model, input_dir, model_index=0, adapter_only=False):
# Note: We import here to reduce import time from general modules, and isolate outside dependencies
import torch.distributed.checkpoint as dist_cp
from torch.distributed.checkpoint.default_planner import DefaultLoadPlanner
from torch.distributed.fsdp.fully_sharded_data_parallel import FullyShardedDataParallel as FSDP
from torch.distributed.fsdp.fully_sharded_data_parallel import StateDictType
accelerator.wait_for_everyone()
if fsdp_plugin.state_dict_type == StateDictType.FULL_STATE_DICT:
# FSDP raises error when single GPU is used with `offload_to_cpu=True` for FULL_STATE_DICT
# so, only enable it when num_processes>1
is_multi_process = accelerator.num_processes > 1
fsdp_plugin.state_dict_config.offload_to_cpu = is_multi_process
fsdp_plugin.state_dict_config.rank0_only = is_multi_process
ctx = (
FSDP.state_dict_type(
model, fsdp_plugin.state_dict_type, fsdp_plugin.state_dict_config, fsdp_plugin.optim_state_dict_config
)
if fsdp_plugin.fsdp_version == 1
else nullcontext()
)
sd_options = _prepare_sd_options(fsdp_plugin)
with ctx:
if fsdp_plugin.state_dict_type == StateDictType.FULL_STATE_DICT:
if type(model) is not FSDP and accelerator.process_index != 0:
if not fsdp_plugin.sync_module_states and fsdp_plugin.fsdp_version == 1:
raise ValueError(
"Set the `sync_module_states` flag to `True` so that model states are synced across processes when "
"initializing FSDP object"
)
return
weights_name = f"{FSDP_MODEL_NAME}.bin" if model_index == 0 else f"{FSDP_MODEL_NAME}_{model_index}.bin"
input_model_file = os.path.join(input_dir, weights_name)
logger.info(f"Loading model from {input_model_file}")
state_dict = torch.load(input_model_file, weights_only=True)
logger.info(f"Model loaded from {input_model_file}")
elif fsdp_plugin.state_dict_type == StateDictType.LOCAL_STATE_DICT:
weights_name = (
f"{FSDP_MODEL_NAME}_rank{accelerator.process_index}.bin"
if model_index == 0
else f"{FSDP_MODEL_NAME}_{model_index}_rank{accelerator.process_index}.bin"
)
input_model_file = os.path.join(input_dir, weights_name)
logger.info(f"Loading model from {input_model_file}")
state_dict = torch.load(input_model_file, weights_only=True)
logger.info(f"Model loaded from {input_model_file}")
elif fsdp_plugin.state_dict_type == StateDictType.SHARDED_STATE_DICT:
ckpt_dir = (
os.path.join(input_dir, f"{FSDP_MODEL_NAME}_{model_index}")
if f"{FSDP_MODEL_NAME}" not in input_dir
else input_dir
)
logger.info(f"Loading model from {ckpt_dir}")
state_dict = {"model": _get_model_state_dict(model, adapter_only=adapter_only, sd_options=sd_options)}
dist_cp.load(
state_dict=state_dict,
storage_reader=dist_cp.FileSystemReader(ckpt_dir),
planner=DefaultLoadPlanner(),
)
state_dict = state_dict["model"]
logger.info(f"Model loaded from {ckpt_dir}")
load_result = _set_model_state_dict(model, state_dict, adapter_only=adapter_only, sd_options=sd_options)
return load_result
def save_fsdp_optimizer(fsdp_plugin, accelerator, optimizer, model, output_dir, optimizer_index=0):
# Note: We import here to reduce import time from general modules, and isolate outside dependencies
import torch.distributed.checkpoint as dist_cp
from torch.distributed.checkpoint.default_planner import DefaultSavePlanner
from torch.distributed.fsdp.fully_sharded_data_parallel import FullyShardedDataParallel as FSDP
from torch.distributed.fsdp.fully_sharded_data_parallel import StateDictType
os.makedirs(output_dir, exist_ok=True)
ctx = (
FSDP.state_dict_type(
model, fsdp_plugin.state_dict_type, fsdp_plugin.state_dict_config, fsdp_plugin.optim_state_dict_config
)
if fsdp_plugin.fsdp_version == 1
else nullcontext()
)
sd_options = _prepare_sd_options(fsdp_plugin)
with ctx:
if fsdp_plugin.fsdp_version == 2:
from torch.distributed.checkpoint.state_dict import get_optimizer_state_dict
optim_state = get_optimizer_state_dict(model, optimizer, options=sd_options)
else:
optim_state = FSDP.optim_state_dict(model, optimizer)
if fsdp_plugin.state_dict_type == StateDictType.FULL_STATE_DICT:
if accelerator.process_index == 0:
optim_state_name = (
f"{OPTIMIZER_NAME}.bin" if optimizer_index == 0 else f"{OPTIMIZER_NAME}_{optimizer_index}.bin"
)
output_optimizer_file = os.path.join(output_dir, optim_state_name)
logger.info(f"Saving Optimizer state to {output_optimizer_file}")
torch.save(optim_state, output_optimizer_file)
logger.info(f"Optimizer state saved in {output_optimizer_file}")
else:
ckpt_dir = os.path.join(output_dir, f"{OPTIMIZER_NAME}_{optimizer_index}")
os.makedirs(ckpt_dir, exist_ok=True)
logger.info(f"Saving Optimizer state to {ckpt_dir}")
dist_cp.save(
state_dict={"optimizer": optim_state},
storage_writer=dist_cp.FileSystemWriter(ckpt_dir),
planner=DefaultSavePlanner(),
)
logger.info(f"Optimizer state saved in {ckpt_dir}")
def load_fsdp_optimizer(fsdp_plugin, accelerator, optimizer, model, input_dir, optimizer_index=0, adapter_only=False):
# Note: We import here to reduce import time from general modules, and isolate outside dependencies
import torch.distributed.checkpoint as dist_cp
from torch.distributed.fsdp.fully_sharded_data_parallel import FullyShardedDataParallel as FSDP
from torch.distributed.fsdp.fully_sharded_data_parallel import StateDictType
accelerator.wait_for_everyone()
ctx = (
FSDP.state_dict_type(
model, fsdp_plugin.state_dict_type, fsdp_plugin.state_dict_config, fsdp_plugin.optim_state_dict_config
)
if fsdp_plugin.fsdp_version == 1
else nullcontext()
)
sd_options = _prepare_sd_options(fsdp_plugin)
with ctx:
if fsdp_plugin.state_dict_type == StateDictType.FULL_STATE_DICT:
optim_state = None
if accelerator.process_index == 0 or not fsdp_plugin.optim_state_dict_config.rank0_only:
optimizer_name = (
f"{OPTIMIZER_NAME}.bin" if optimizer_index == 0 else f"{OPTIMIZER_NAME}_{optimizer_index}.bin"
)
input_optimizer_file = os.path.join(input_dir, optimizer_name)
logger.info(f"Loading Optimizer state from {input_optimizer_file}")
optim_state = torch.load(input_optimizer_file, weights_only=True)
logger.info(f"Optimizer state loaded from {input_optimizer_file}")
else:
ckpt_dir = (
os.path.join(input_dir, f"{OPTIMIZER_NAME}_{optimizer_index}")
if f"{OPTIMIZER_NAME}" not in input_dir
else input_dir
)
logger.info(f"Loading Optimizer from {ckpt_dir}")
optim_state = {"optimizer": optimizer.state_dict()}
dist_cp.load(
optim_state,
checkpoint_id=ckpt_dir,
storage_reader=dist_cp.FileSystemReader(ckpt_dir),
)
optim_state = optim_state["optimizer"]
logger.info(f"Optimizer loaded from {ckpt_dir}")
if fsdp_plugin.fsdp_version == 1:
flattened_osd = FSDP.optim_state_dict_to_load(model=model, optim=optimizer, optim_state_dict=optim_state)
optimizer.load_state_dict(flattened_osd)
else:
from torch.distributed.checkpoint.state_dict import set_optimizer_state_dict
set_optimizer_state_dict(model, optimizer, optim_state, options=sd_options)
def _distributed_checkpoint_to_merged_weights(checkpoint_dir: str, save_path: str, safe_serialization: bool = True):
"""
Passthrough to `torch.distributed.checkpoint.format_utils.dcp_to_torch_save`
Will save under `save_path` as either `model.safetensors` or `pytorch_model.bin`.
"""
# Note: We import here to reduce import time from general modules, and isolate outside dependencies
import torch.distributed.checkpoint as dist_cp
import torch.distributed.checkpoint.format_utils as dist_cp_format_utils
state_dict = {}
save_path = Path(save_path)
save_path.mkdir(exist_ok=True)
dist_cp_format_utils._load_state_dict(
state_dict,
storage_reader=dist_cp.FileSystemReader(checkpoint_dir),
planner=dist_cp_format_utils._EmptyStateDictLoadPlanner(),
no_dist=True,
)
save_path = save_path / SAFE_WEIGHTS_NAME if safe_serialization else save_path / WEIGHTS_NAME
# To handle if state is a dict like {model: {...}}
if len(state_dict.keys()) == 1:
state_dict = state_dict[list(state_dict)[0]]
save(state_dict, save_path, safe_serialization=safe_serialization)
return save_path
def merge_fsdp_weights(
checkpoint_dir: str, output_path: str, safe_serialization: bool = True, remove_checkpoint_dir: bool = False
):
"""
Merge the weights from sharded FSDP model checkpoints into a single combined checkpoint. Should be used if
`SHARDED_STATE_DICT` was used for the model. Weights will be saved to `{output_path}/model.safetensors` if
`safe_serialization` else `pytorch_model.bin`.
Note: this is a CPU-bound process.
Args:
checkpoint_dir (`str`):
The directory containing the FSDP checkpoints (can be either the model or optimizer).
output_path (`str`):
The path to save the merged checkpoint.
safe_serialization (`bool`, *optional*, defaults to `True`):
Whether to save the merged weights with safetensors (recommended).
remove_checkpoint_dir (`bool`, *optional*, defaults to `False`):
Whether to remove the checkpoint directory after merging.
"""
checkpoint_dir = Path(checkpoint_dir)
from accelerate.state import PartialState
if not is_torch_version(">=", "2.3.0"):
raise ValueError("`merge_fsdp_weights` requires PyTorch >= 2.3.0`")
# Verify that the checkpoint directory exists
if not checkpoint_dir.exists():
model_path_exists = (checkpoint_dir / "pytorch_model_fsdp_0").exists()
optimizer_path_exists = (checkpoint_dir / "optimizer_0").exists()
err = f"Tried to load from {checkpoint_dir} but couldn't find a valid metadata file."
if model_path_exists and optimizer_path_exists:
err += " However, potential model and optimizer checkpoint directories exist."
err += f"Please pass in either {checkpoint_dir}/pytorch_model_fsdp_0 or {checkpoint_dir}/optimizer_0"
err += "instead."
elif model_path_exists:
err += " However, a potential model checkpoint directory exists."
err += f"Please try passing in {checkpoint_dir}/pytorch_model_fsdp_0 instead."
elif optimizer_path_exists:
err += " However, a potential optimizer checkpoint directory exists."
err += f"Please try passing in {checkpoint_dir}/optimizer_0 instead."
raise ValueError(err)
# To setup `save` to work
state = PartialState()
if state.is_main_process:
logger.info(f"Merging FSDP weights from {checkpoint_dir}")
save_path = _distributed_checkpoint_to_merged_weights(checkpoint_dir, output_path, safe_serialization)
logger.info(f"Successfully merged FSDP weights and saved to {save_path}")
if remove_checkpoint_dir:
logger.info(f"Removing old checkpoint directory {checkpoint_dir}")
shutil.rmtree(checkpoint_dir)
state.wait_for_everyone()
def ensure_weights_retied(param_init_fn, model: torch.nn.Module, device: torch.device):
_tied_names = getattr(model, "_tied_weights_keys", None)
if not _tied_names:
# if no tied names just passthrough
return param_init_fn
# get map of parameter instances to params.
# - needed for replacement later
_tied_params = {}
for name in _tied_names:
name = name.split(".")
name, param_name = ".".join(name[:-1]), name[-1]
mod = model.get_submodule(name)
param = getattr(mod, param_name)
_tied_params[id(param)] = None # placeholder for the param first
# build param_init_fn for the case with tied params
def param_init_fn_tied_param(module: torch.nn.Module):
# track which params to tie
# - usually only 1, but for completeness consider > 1
params_to_tie = defaultdict(list)
for n, param in module.named_parameters(recurse=False):
if id(param) in _tied_params:
params_to_tie[id(param)].append(n)
# call the param init fn, which potentially re-allocates the
# parameters
module = param_init_fn(module)
# search the parameters again and tie them up again
for id_key, _param_names in params_to_tie.items():
for param_name in _param_names:
param = _tied_params[id_key]
if param is None:
# everything will be tied to the first time the
# param is observed
_tied_params[id_key] = getattr(module, param_name)
else:
setattr(module, param_name, param) # tie
return module
return param_init_fn_tied_param
def fsdp2_load_full_state_dict(accelerator, model: torch.nn.Module, full_sd: dict):
"""
Loads the full state dict (could be only on rank 0) into the sharded model. This is done by broadcasting the
parameters from rank 0 to all other ranks. This function modifies the model in-place.
Args:
accelerator (`Accelerator`): The accelerator instance
model (`torch.nn.Module`):
The model to load the state dict into, expected to be on meta device or a VRAM spike can occur
full_sd (`dict`): The full state dict to load, can only be on rank 0
"""
import torch.distributed as dist
from torch.distributed.tensor import distribute_tensor
# Model was previously copied to meta device
meta_sharded_sd = model.state_dict()
sharded_sd = {}
# Rank 0 distributes the full state dict to other ranks
def _infer_parameter_dtype(model, param_name, empty_param):
try:
old_param = model.get_parameter_or_buffer(param_name)
except AttributeError:
# Need this for LORA, as there some params are not *parameters* of sorts
base_param_name, local_param_name = param_name.rsplit(".", 1)
submodule = model.get_submodule(base_param_name)
old_param = getattr(submodule, local_param_name)
is_torch_e4m3fn_available = hasattr(torch, "float8_e4m3fn")
casting_dtype = None
is_param_float8_e4m3fn = is_torch_e4m3fn_available and empty_param.dtype == torch.float8_e4m3fn
if empty_param.dtype.is_floating_point and not is_param_float8_e4m3fn:
casting_dtype = old_param.dtype
return old_param is not None and old_param.is_contiguous(), casting_dtype
def _cast_and_contiguous(tensor, to_contiguous, dtype):
if dtype is not None:
tensor = tensor.to(dtype=dtype)
if to_contiguous:
tensor = tensor.contiguous()
return tensor
if accelerator.is_main_process:
for (param_name, full_param), sharded_param in zip(full_sd.items(), meta_sharded_sd.values()):
device_mesh = sharded_param.device_mesh
full_param = full_param.detach().to(device_mesh.device_type)
dist.broadcast(full_param, src=0, group=device_mesh.get_group())
sharded_tensor = distribute_tensor(full_param, device_mesh, sharded_param.placements)
to_contiguous, casting_dtype = _infer_parameter_dtype(
model,
param_name,
full_param,
)
sharded_tensor = _cast_and_contiguous(sharded_tensor, to_contiguous, casting_dtype)
sharded_sd[param_name] = sharded_tensor
# We need this else to have a matching `broadcast` for all of the ranks, else we deadlock
else:
for param_name, sharded_param in meta_sharded_sd.items():
device_mesh = sharded_param.device_mesh
full_tensor = torch.empty(sharded_param.size(), device=device_mesh.device_type, dtype=sharded_param.dtype)
dist.broadcast(full_tensor, src=0, group=device_mesh.get_group())
sharded_tensor = distribute_tensor(full_tensor, device_mesh, sharded_param.placements)
to_contiguous, casting_dtype = _infer_parameter_dtype(
model,
param_name,
full_tensor,
)
sharded_tensor = _cast_and_contiguous(sharded_tensor, to_contiguous, casting_dtype)
sharded_sd[param_name] = sharded_tensor
# we set `assign=True` because our params are on meta device
model.load_state_dict(sharded_sd, assign=True)
return model
def fsdp2_switch_optimizer_parameters(optimizer: torch.optim.Optimizer, mapping: dict):
"""
Switches the parameters of the optimizer to new ones (sharded parameters in usual case). This function modifies the
optimizer in-place.
Args:
optimizer (`torch.optim.Optimizer`): Optimizer instance which contains the original model parameters
mapping (`dict`): Mapping from the original parameter (specified by `data_ptr`) to the sharded parameter
Raises:
KeyError:
If a parameter in the optimizer couldn't be switched to its sharded version. This should never happen and
indicates a bug. If we kept the original params instead of raising, the training wouldn't be numerically
correct and weights wouldn't get updated.
"""
try:
for param_group in optimizer.param_groups:
param_group["params"] = [mapping[p.data_ptr] for p in param_group["params"]]
except KeyError:
# This shouldn't ever happen, but we want to fail here else training wouldn't be numerically correct
# This basically means that we're missing a mapping from the original parameter to the sharded parameter
raise KeyError(
"A parameter in the optimizer couldn't be switched to its sharded version. This breaks the training. Please raise an issue on GitHub."
)
def fsdp2_apply_ac(accelerator, model: torch.nn.Module):
"""
Applies the activation checkpointing to the model.
Args:
accelerator (`Accelerator`): The accelerator instance
model (`torch.nn.Module`): The model to apply the activation checkpointing to
Returns:
`torch.nn.Module`: The model with the activation checkpointing applied
"""
from torch.distributed.algorithms._checkpoint.checkpoint_wrapper import (
checkpoint_wrapper,
)
auto_wrap_policy_func = fsdp2_prepare_auto_wrap_policy(accelerator.state.fsdp_plugin, model)
for layer_name, layer in get_module_children_bottom_up(model, return_fqns=True)[:-1]:
if len(layer_name.split(".")) > 1:
parent_name, child_name = layer_name.rsplit(".", 1)
else:
parent_name = None
child_name = layer_name
parent_module = model.get_submodule(parent_name) if parent_name else model
if auto_wrap_policy_func(parent_module):
layer = checkpoint_wrapper(layer, preserve_rng_state=False)
parent_module.register_module(child_name, layer)
return model
def fsdp2_prepare_model(accelerator, model: torch.nn.Module) -> torch.nn.Module:
"""Prepares the model for FSDP2 in-place. Also returns the model to avoid misuse of the original model.
Args:
accelerator (`Accelerator`): The accelerator instance
model (`torch.nn.Module`): The model to prepare
Returns:
`torch.nn.Module`: Prepared model
"""
from torch.distributed.fsdp import FSDPModule, MixedPrecisionPolicy, fully_shard
is_type_fsdp = isinstance(model, FSDPModule) or (
is_compiled_module(model) and isinstance(model._orig_mod, FSDPModule)
)
if is_type_fsdp:
return model
fsdp2_plugin = accelerator.state.fsdp_plugin
fsdp2_plugin.set_auto_wrap_policy(model)
original_sd = model.state_dict()
fsdp2_kwargs = {
"reshard_after_forward": fsdp2_plugin.reshard_after_forward,
"offload_policy": fsdp2_plugin.cpu_offload,
# `fully_shard` doesn't accept `None` in case of `MixedPrecisionPolicy`
"mp_policy": fsdp2_plugin.mixed_precision_policy or MixedPrecisionPolicy(),
}
model_has_params4bit = False
for name, param in model.named_parameters():
# this is a temporary fix whereby loading models with bnb params cannot be moved from
# GPU to a meta device due with FSDP2 because torch operations don't return the original class type
# bypassing the move to meta will still cause the VRAM spike, but at least it still will load
if param.__class__.__name__ == "Params4bit":
model_has_params4bit = True
break
if fsdp2_plugin.cpu_ram_efficient_loading and not model_has_params4bit:
# Context: `fully_shard` moves the model to GPU if it was on CPU, however it can also be on `meta` and then it stays there even after `fully_shard`
# For this reason, we need to move the model to `meta` device, as then sharding happens on `meta` device
# If we kept the model on CPU (`cpu_ram_efficient_loading` has model be on CPU on all ranks, though non-main ranks only have `torch.emtpy`), `fully_shard` would move it to GPU
# Afterwards, when we call `fsdp2_load_full_state_dict`, us creating the state_dict would result into briefly having two copies of model state_dict on the GPU -> VRAM spike
# We need to keep the original non-persistent buffers, as those MAY not be in the state_dict, resulting in them staying on meta device
# Also, these buffers aren't getting sharded by default
# We get the FQNs of all non-persistent buffers, to re-register them after
non_persistent_buffer_fqns = get_non_persistent_buffers(model, recurse=True, fqns=True)
original_non_persistent_buffers = copy.deepcopy(
{k: v for k, v in model.named_buffers() if k in non_persistent_buffer_fqns}
)
# We move the model to meta device, as then sharding happens on meta device
model = model.to(torch.device("meta"))
# We need to re-tie the weights, not exactly sure why, but if we don't do this, reference to `lm_head/embed_tokens` stay hanging -> more VRAM usage
# We assume `transformers` models have a `tie_weights` method if they support it
if hasattr(model, "tie_weights"):
model.tie_weights()
auto_wrap_policy_func = fsdp2_prepare_auto_wrap_policy(fsdp2_plugin, model)
if auto_wrap_policy_func is not None:
# We skip the model itself, as that one is always wrapped
for module in get_module_children_bottom_up(model)[:-1]:
if auto_wrap_policy_func(module) and not isinstance(module, FSDPModule):
fully_shard(module, **fsdp2_kwargs)
if not isinstance(model, FSDPModule):
fully_shard(model, **fsdp2_kwargs)
if fsdp2_plugin.cpu_ram_efficient_loading:
# If `cpu_ram_efficient_loading` is enabled, only rank 0 loads the weights
# Other ranks have an empty model on `meta` device, so we need to distribute the weights properly
fsdp2_load_full_state_dict(accelerator, model, original_sd)
if fsdp2_plugin.cpu_ram_efficient_loading and not model_has_params4bit:
# We re-register the buffers, as they may not be in the state_dict
for fqn, buffer_tensor in original_non_persistent_buffers.items():
buffer_tensor = buffer_tensor.to(accelerator.device)
if "." in fqn:
parent_fqn, local_buffer_name = fqn.rsplit(".", 1)
parent_module = model.get_submodule(parent_fqn)
else:
local_buffer_name = fqn
parent_module = model
parent_module.register_buffer(local_buffer_name, buffer_tensor, persistent=False)
# We need to tie the weights again, as call to `load_full_state_dict` breaks the tie
# Needs to be called both here and above
# removing this call makes the have slightly different loss
# removing the call above leads to extra memory usage as explained in the comment above
if hasattr(model, "tie_weights"):
model.tie_weights()
# There is no `dtype` attribution for nn.Module
# Set it to None if it doesn't exist and do the upcast always
model_dtype = getattr(model, "dtype", None)
if accelerator.mixed_precision != "no" and (model_dtype is None or model_dtype != torch.float32):
# We upcast the model according to `deepspeed`'s implementation
# More info about this can be found in `accelerator.py:prepare_model`s FSDP1 section
model = model.to(torch.float32)
if accelerator.is_main_process:
# TODO(siro1): Add a warning for each parameter that was upcasted
warnings.warn(
"FSDP upcast of low precision parameters to fp32 (since mixed_precision != 'no') may affect the precision of model checkpoints."
)
return model
def fsdp2_prepare_auto_wrap_policy(fsdp2_plugin, model: torch.nn.Module) -> Callable[[torch.nn.Module], bool]:
"""Prepares the auto wrap policy based on its type, done to mimic the behaviour of FSDP1 auto wrap policy.
Args:
fsdp2_plugin (`FullyShardedDataParallelPlugin`):
Instance of `FullyShardedDataParallelPlugin` containing the configuration options
auto_wrap_policy_type (`str`):
Either `transformer` or `size`
model (`torch.nn.Module`):
The model to wrap
Returns:
`Callable[[torch.nn.Module], bool]`:
The auto wrap policy function to be applied to the model
"""
from torch.distributed.fsdp.wrap import size_based_auto_wrap_policy, transformer_auto_wrap_policy
fn = fsdp2_plugin.auto_wrap_policy
if isinstance(fn, functools.partial):
fn = fn.func
if fn is transformer_auto_wrap_policy:
no_split_modules = getattr(model, "_no_split_modules", None)
if no_split_modules is None:
no_split_modules = []
transformer_cls_names_to_wrap = list(no_split_modules)
if fsdp2_plugin.transformer_cls_names_to_wrap is not None:
transformer_cls_names_to_wrap = fsdp2_plugin.transformer_cls_names_to_wrap
transformer_cls_to_wrap = set()
for layer_class in transformer_cls_names_to_wrap:
transformer_cls = get_module_class_from_name(model, layer_class)
if transformer_cls is None:
raise ValueError(f"Could not find the transformer layer class {layer_class} in the model.")
transformer_cls_to_wrap.add(transformer_cls)
def policy(module: torch.nn.Module) -> bool:
if fsdp2_plugin.transformer_cls_names_to_wrap is None:
return False
return isinstance(module, tuple(transformer_cls_to_wrap))
elif fn is size_based_auto_wrap_policy:
def policy(module: torch.nn.Module) -> bool:
module_num_params = sum(p.numel() for p in module.parameters())
return module_num_params > fsdp2_plugin.min_num_params
else:
return None
return policy
def get_fsdp2_grad_scaler(**kwargs):
"""
Returns a `GradScaler` for FSDP2, as the current implementation of `get_grad_scaler` doesn't accept other args. We
need this as current `get_grad_scaler` accepts only `distributed_type` as arg, which doesn't differentiate between
FSDP1 and FSDP2
"""
from torch.amp.grad_scaler import GradScaler
return GradScaler(**kwargs)
def fsdp2_canonicalize_names(named_params: dict) -> dict:
"""Removes parameter name modifiers in order to map them back to their original names.
See huggingface/accelerate#3554 for more context.
Args:
named_params (`dict`): The named parameters dictionary to canonicalize.
Returns:
`dict`: The canonicalized named parameters dictionary
"""
named_params = {k.replace("._checkpoint_wrapped_module", ""): v for k, v in named_params.items()}
named_params = {
k.replace("_orig_mod.", "") if k.startswith("_orig_mod.") else k: v for k, v in named_params.items()
}
named_params = {k.replace("._orig_mod", ""): v for k, v in named_params.items()}
return named_params
|