-
Notifications
You must be signed in to change notification settings - Fork 2
/
finetune.py
804 lines (734 loc) · 27.6 KB
/
finetune.py
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
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
#!/usr/bin/env python
# coding=utf-8
import argparse
import logging
import math
import os
import random
import datasets
import torch
from functools import partial
from accelerate import Accelerator
from accelerate.logging import get_logger
from accelerate.utils import set_seed
from datasets import load_dataset
from torch.utils.data import DataLoader
from tqdm.auto import tqdm
import pdb
import transformers
from transformers import (
AutoConfig,
AutoModelForCausalLM,
AutoTokenizer,
SchedulerType,
DataCollatorForSeq2Seq,
get_scheduler,
GPTNeoXTokenizerFast,
GPT2Tokenizer,
OPTForCausalLM,
)
from peft import LoraConfig, TaskType, get_peft_model
logger = get_logger(__name__)
"""
@misc{wang2023far,
title={How Far Can Camels Go? Exploring the State of Instruction Tuning on Open Resources},
author={Yizhong Wang and Hamish Ivison and Pradeep Dasigi and Jack Hessel and Tushar Khot and Khyathi Raghavi Chandu and David Wadden and Kelsey MacMillan and Noah A. Smith and Iz Beltagy and Hannaneh Hajishirzi},
year={2023},
eprint={2306.04751},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
"""
def parse_args():
parser = argparse.ArgumentParser(
description="Finetune a transformers model on a causal language modeling task"
)
parser.add_argument(
"--dataset_name",
type=str,
default=None,
help="The name of the dataset to use (via the datasets library).",
)
parser.add_argument(
"--dataset_config_name",
type=str,
default=None,
help="The configuration name of the dataset to use (via the datasets library).",
)
parser.add_argument(
"--train_file",
type=str,
default=None,
help="A csv or a json file containing the training data.",
)
parser.add_argument(
"--model_name_or_path",
type=str,
help="Path to pretrained model or model identifier from huggingface.co/models.",
required=False,
)
parser.add_argument(
"--config_name",
type=str,
default=None,
help="Pretrained config name or path if not the same as model_name",
)
parser.add_argument(
"--use_lora",
action="store_true",
help="If passed, will use LORA (low-rank parameter-efficient training) to train the model.",
)
parser.add_argument(
"--lora_rank",
type=int,
default=64,
help="The rank of lora.",
)
parser.add_argument(
"--lora_alpha",
type=float,
default=16,
help="The alpha parameter of lora.",
)
parser.add_argument(
"--lora_dropout",
type=float,
default=0.1,
help="The dropout rate of lora modules.",
)
parser.add_argument(
"--save_merged_lora_model",
action="store_true",
help="If passed, will merge the lora modules and save the entire model.",
)
parser.add_argument(
"--tokenizer_name",
type=str,
default=None,
help="Pretrained tokenizer name or path if not the same as model_name",
)
parser.add_argument(
"--use_slow_tokenizer",
action="store_true",
help="If passed, will use a slow tokenizer (not backed by the 🤗 Tokenizers library).",
)
parser.add_argument(
"--max_seq_length",
type=int,
default=4096,
help="The maximum total sequence length (prompt+completion) of each training example.",
)
parser.add_argument(
"--per_device_train_batch_size",
type=int,
default=8,
help="Batch size (per device) for the training dataloader.",
)
parser.add_argument(
"--learning_rate",
type=float,
default=5e-5,
help="Initial learning rate (after the potential warmup period) to use.",
)
parser.add_argument(
"--weight_decay", type=float, default=0.0, help="Weight decay to use."
)
parser.add_argument(
"--num_train_epochs",
type=int,
default=3,
help="Total number of training epochs to perform.",
)
parser.add_argument(
"--max_train_steps",
type=int,
default=None,
help="Total number of training steps to perform. If provided, overrides num_train_epochs.",
)
parser.add_argument(
"--gradient_accumulation_steps",
type=int,
default=1,
help="Number of updates steps to accumulate before performing a backward/update pass.",
)
parser.add_argument(
"--lr_scheduler_type",
type=SchedulerType,
default="linear",
help="The scheduler type to use.",
choices=[
"linear",
"cosine",
"cosine_with_restarts",
"polynomial",
"constant",
"constant_with_warmup",
],
)
parser.add_argument(
"--warmup_ratio",
type=float,
default=0,
help="Ratio of total training steps used for warmup.",
)
parser.add_argument(
"--output_dir", type=str, default=None, help="Where to store the final model."
)
parser.add_argument(
"--seed", type=int, default=None, help="A seed for reproducible training."
)
parser.add_argument(
"--preprocessing_num_workers",
type=int,
default=None,
help="The number of processes to use for the preprocessing.",
)
parser.add_argument(
"--overwrite_cache",
action="store_true",
help="Overwrite the cached training and evaluation sets",
)
parser.add_argument(
"--checkpointing_steps",
type=str,
default=None,
help="Whether the various states should be saved at the end of every n steps, or 'epoch' for each epoch.",
)
parser.add_argument(
"--logging_steps",
type=int,
default=None,
help="Log the training loss and learning rate every logging_steps steps.",
)
parser.add_argument(
"--resume_from_checkpoint",
type=str,
default=None,
help="If the training should continue from a checkpoint folder.",
)
parser.add_argument(
"--with_tracking",
action="store_true",
help="Whether to enable experiment trackers for logging.",
)
parser.add_argument(
"--tracking_group",
type=str,
default=None,
help="Configure tracking group name.",
)
parser.add_argument(
"--tracking_project",
type=str,
default="test",
help="Configure tracking project name.",
)
parser.add_argument(
"--report_to",
type=str,
default="all",
help=(
'The integration to report the results and logs to. Supported platforms are `"tensorboard"`,'
' `"wandb"`, `"comet_ml"` and `"clearml"`. Use `"all"` (default) to report to all integrations.'
"Only applicable when `--with_tracking` is passed."
),
)
parser.add_argument(
"--low_cpu_mem_usage",
action="store_true",
help=(
"It is an option to create the model as an empty shell, then only materialize its parameters when the pretrained weights are loaded."
"If passed, LLM loading time and RAM consumption will be benefited."
),
)
parser.add_argument(
"--mask_observation_compl",
action="store_true",
)
args = parser.parse_args()
# Sanity checks
if args.dataset_name is None and args.train_file is None:
raise ValueError("Need either a dataset name or a training file.")
else:
if args.train_file is not None:
extension = args.train_file.split(".")[-1]
assert extension in [
"json",
"jsonl",
], "`train_file` should be a json/jsonl file."
return args
def encode_with_prompt_completion_format(example, tokenizer, max_seq_length):
"""
Here we assume each example has 'prompt' and 'completion' fields.
We concatenate prompt and completion and tokenize them together because otherwise prompt will be padded/trancated
and it doesn't make sense to follow directly with the completion.
"""
# if prompt doesn't end with space and completion doesn't start with space, add space
if not example["prompt"].endswith((" ", "\n", "\t")) and not example[
"completion"
].startswith((" ", "\n", "\t")):
example_text = example["prompt"] + " " + example["completion"]
else:
example_text = example["prompt"] + example["completion"]
example_text = example_text + tokenizer.eos_token
tokenized_example = tokenizer(
example_text, return_tensors="pt", max_length=max_seq_length, truncation=True
)
input_ids = tokenized_example.input_ids
labels = input_ids.clone()
tokenized_prompt = tokenizer(
example["prompt"],
return_tensors="pt",
max_length=max_seq_length,
truncation=True,
)
# mask the prompt part for avoiding loss
labels[:, : tokenized_prompt.input_ids.shape[1]] = -100
attention_mask = torch.ones_like(input_ids)
return {
"input_ids": input_ids.flatten(),
"labels": labels.flatten(),
"attention_mask": attention_mask.flatten(),
}
def encode_with_prompt_completion_format_observation_mask(
example,
tokenizer,
max_seq_length,
observation_token="<|observation|>",
action_token="<|action|>",
):
"""
Here we assume each example has 'prompt' and 'completion' fields.
We concatenate prompt and completion and tokenize them together because otherwise prompt will be padded/trancated
and it doesn't make sense to follow directly with the completion.
"""
# if prompt doesn't end with space and completion doesn't start with space, add space
if not example["prompt"].endswith((" ", "\n", "\t")) and not example[
"completion"
].startswith((" ", "\n", "\t")):
example_text = example["prompt"] + " " + example["completion"]
else:
example_text = example["prompt"] + example["completion"]
example_text = example_text + tokenizer.eos_token
tokenized_example = tokenizer(
example_text, return_tensors="pt", max_length=max_seq_length, truncation=True
)
input_ids = tokenized_example.input_ids
labels = input_ids.clone()
tokenized_prompt = tokenizer(
example["prompt"],
return_tensors="pt",
max_length=max_seq_length,
truncation=True,
)
# unpack reserved token ids
observation_id = tokenizer(observation_token)["input_ids"][0]
action_id = tokenizer(action_token)["input_ids"][0]
mask = torch.zeros_like(labels)
starts = (
(
(input_ids[0, :] == observation_id).int()
+ (input_ids[0, :] == action_id).int()
)
.nonzero()
.squeeze()
)
if len(starts.shape) == 0:
# mask the prompt part for avoiding loss
labels[:, : tokenized_prompt.input_ids.shape[1]] = -100
attention_mask = torch.ones_like(input_ids)
return {
"input_ids": input_ids.flatten(),
"labels": labels.flatten(),
"attention_mask": attention_mask.flatten(),
}
# mask the prompt + observation parts for avoiding loss
for i, s in enumerate(starts):
if i % 2 == 1:
mask[:, s + 1 :] += 1
else:
mask[:, s:] += 1
evens_mask = mask % 2 == 0
labels[evens_mask] = -100
labels[:, : tokenized_prompt.input_ids.shape[1]] = -100
attention_mask = torch.ones_like(input_ids)
return {
"input_ids": input_ids.flatten(),
"labels": labels.flatten(),
"attention_mask": attention_mask.flatten(),
}
def main():
args = parse_args()
import datetime
torch.distributed.init_process_group(
init_method=None,
timeout=datetime.timedelta(seconds=100000),
world_size=-1,
rank=-1,
store=None,
group_name="",
pg_options=None,
backend="nccl",
)
accelerator_log_kwargs = {}
if args.with_tracking:
accelerator_log_kwargs["log_with"] = args.report_to
accelerator_log_kwargs["project_dir"] = args.output_dir
accelerator = Accelerator(
gradient_accumulation_steps=args.gradient_accumulation_steps,
**accelerator_log_kwargs,
)
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
level=logging.INFO,
)
logger.info(accelerator.state, main_process_only=False)
if accelerator.is_local_main_process:
datasets.utils.logging.set_verbosity_warning()
transformers.utils.logging.set_verbosity_info()
else:
datasets.utils.logging.set_verbosity_error()
transformers.utils.logging.set_verbosity_error()
if args.seed is not None:
set_seed(args.seed)
if accelerator.is_main_process:
if args.output_dir is not None:
os.makedirs(args.output_dir, exist_ok=True)
accelerator.wait_for_everyone()
data_files = {}
dataset_args = {}
if args.train_file is not None:
data_files["train"] = args.train_file
raw_datasets = load_dataset(
"json",
data_files=data_files,
**dataset_args,
)
if args.config_name:
config = AutoConfig.from_pretrained(args.config_name)
elif args.model_name_or_path:
config = AutoConfig.from_pretrained(args.model_name_or_path)
else:
raise ValueError(
"You are instantiating a new config instance from scratch. This is not supported by this script."
)
if args.tokenizer_name:
tokenizer = AutoTokenizer.from_pretrained(
args.tokenizer_name, use_fast=not args.use_slow_tokenizer
)
elif args.model_name_or_path:
tokenizer = AutoTokenizer.from_pretrained(
args.model_name_or_path, use_fast=not args.use_slow_tokenizer
)
else:
raise ValueError(
"You are instantiating a new tokenizer from scratch. This is not supported by this script."
"You can do it from another script, save it, and load it from here, using --tokenizer_name."
)
if args.model_name_or_path:
model = AutoModelForCausalLM.from_pretrained(
args.model_name_or_path,
from_tf=bool(".ckpt" in args.model_name_or_path),
config=config,
low_cpu_mem_usage=args.low_cpu_mem_usage,
)
else:
logger.info("Training new model from scratch")
model = AutoModelForCausalLM.from_config(config)
model = model.cuda()
num_added_tokens = tokenizer.add_special_tokens(
{
"bos_token": "<s>",
"eos_token": "</s>",
"unk_token": "<unk>",
"pad_token": "<pad>",
}
)
from action_textmap import special_tokens_interaction_history
num_added_tokens += tokenizer.add_tokens(
[v for v in special_tokens_interaction_history.values()]
)
observation_token = special_tokens_interaction_history["observation"]
action_token = special_tokens_interaction_history["action"]
print(f"Adding {num_added_tokens} to tokenizer")
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer))
if args.use_lora:
logger.info("Initializing LORA model...")
peft_config = LoraConfig(
task_type=TaskType.CAUSAL_LM,
inference_mode=False,
r=args.lora_rank,
lora_alpha=args.lora_alpha,
lora_dropout=args.lora_dropout,
)
model = get_peft_model(model, peft_config)
model.print_trainable_parameters()
if (
"prompt" in raw_datasets["train"].column_names
and "completion" in raw_datasets["train"].column_names
):
if args.mask_observation_compl:
encode_function = partial(
encode_with_prompt_completion_format_observation_mask,
tokenizer=tokenizer,
max_seq_length=args.max_seq_length,
observation_token=observation_token,
action_token=action_token,
)
else:
encode_function = partial(
encode_with_prompt_completion_format,
tokenizer=tokenizer,
max_seq_length=args.max_seq_length,
)
else:
raise ValueError(
"You need to have either 'prompt'&'completion in your column names."
)
with accelerator.main_process_first():
lm_datasets = raw_datasets.map(
encode_function,
batched=False,
num_proc=args.preprocessing_num_workers,
load_from_cache_file=not args.overwrite_cache,
remove_columns=[
name
for name in raw_datasets["train"].column_names
if name not in ["input_ids", "labels", "attention_mask"]
],
desc="Tokenizing and reformatting instruction data",
)
lm_datasets.set_format(type="pt")
lm_datasets = lm_datasets.filter(
lambda example: (example["labels"] != -100).any()
)
train_dataset = lm_datasets["train"]
for index in random.sample(range(len(train_dataset)), 3):
logger.info(f"Sample {index} of the training set: {train_dataset[index]}.")
# DataLoaders creation:
train_dataloader = DataLoader(
train_dataset,
shuffle=True,
collate_fn=DataCollatorForSeq2Seq(
tokenizer=tokenizer, model=model, padding="longest"
),
batch_size=args.per_device_train_batch_size,
)
no_decay = ["bias", "layer_norm.weight"]
optimizer_grouped_parameters = [
{
"params": [
p
for n, p in model.named_parameters()
if not any(nd in n for nd in no_decay)
],
"weight_decay": args.weight_decay,
},
{
"params": [
p
for n, p in model.named_parameters()
if any(nd in n for nd in no_decay)
],
"weight_decay": 0.0,
},
]
optimizer = torch.optim.AdamW(optimizer_grouped_parameters, lr=args.learning_rate)
overrode_max_train_steps = False
num_update_steps_per_epoch = math.ceil(
len(train_dataloader) / args.gradient_accumulation_steps
)
if args.max_train_steps is None:
args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
overrode_max_train_steps = True
num_training_steps_for_scheduler = (
args.max_train_steps
if overrode_max_train_steps
else args.max_train_steps * accelerator.num_processes
)
lr_scheduler = get_scheduler(
name=args.lr_scheduler_type,
optimizer=optimizer,
num_training_steps=num_training_steps_for_scheduler,
num_warmup_steps=int(num_training_steps_for_scheduler * args.warmup_ratio),
)
# Prepare everything with `accelerator`.
model, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
model, optimizer, train_dataloader, lr_scheduler
)
# We need to recalculate our total training steps as the size of the training dataloader may have changed.
num_update_steps_per_epoch = math.ceil(
len(train_dataloader) / args.gradient_accumulation_steps
)
if overrode_max_train_steps:
args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
# Afterwards we recalculate our number of training epochs
args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
# Figure out how many steps we should save the Accelerator states
checkpointing_steps = args.checkpointing_steps
if checkpointing_steps is not None and checkpointing_steps.isdigit():
checkpointing_steps = int(checkpointing_steps)
# We need to initialize the trackers we use, and also store our configuration.
# The trackers initializes automatically on the main process.
if args.with_tracking:
experiment_config = vars(args)
# TensorBoard cannot log Enums, need the raw value
experiment_config["lr_scheduler_type"] = experiment_config[
"lr_scheduler_type"
].value
accelerator.init_trackers(
args.tracking_project,
experiment_config,
init_kwargs={"wandb": {"group": args.tracking_group}},
)
# Train!
total_batch_size = (
args.per_device_train_batch_size
* accelerator.num_processes
* args.gradient_accumulation_steps
)
logger.info("***** Running training *****")
logger.info(f" Num examples = {len(train_dataset)}")
logger.info(f" Num Epochs = {args.num_train_epochs}")
logger.info(
f" Instantaneous batch size per device = {args.per_device_train_batch_size}"
)
logger.info(
f" Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}"
)
logger.info(f" Gradient Accumulation steps = {args.gradient_accumulation_steps}")
logger.info(f" Total optimization steps = {args.max_train_steps}")
# Only show the progress bar once on each machine.
progress_bar = tqdm(
range(args.max_train_steps), disable=not accelerator.is_local_main_process
)
completed_steps = 0
starting_epoch = 0
if args.resume_from_checkpoint:
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
model.load_checkpoint(args.resume_from_checkpoint)
path = os.path.basename(args.resume_from_checkpoint)
else:
# Get the most recent checkpoint
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
dirs.sort(key=os.path.getctime)
path = dirs[
-1
] # Sorts folders by date modified, most recent checkpoint is the last
# Extract `epoch_{i}` or `step_{i}`
training_difference = os.path.splitext(path)[0]
if "epoch" in training_difference:
starting_epoch = int(training_difference.replace("epoch_", "")) + 1
resume_step = None
else:
# need to multiply `gradient_accumulation_steps` to reflect real steps
resume_step = (
int(training_difference.replace("step_", ""))
* args.gradient_accumulation_steps
)
starting_epoch = resume_step // len(train_dataloader)
resume_step -= starting_epoch * len(train_dataloader)
# update the progress_bar if load from checkpoint
progress_bar.update(starting_epoch * num_update_steps_per_epoch)
completed_steps = starting_epoch * num_update_steps_per_epoch
for epoch in range(starting_epoch, args.num_train_epochs):
model.train()
total_loss = 0
for step, batch in enumerate(train_dataloader):
# We need to skip steps until we reach the resumed step
if args.resume_from_checkpoint and epoch == starting_epoch:
if resume_step is not None and completed_steps < resume_step:
if step % args.gradient_accumulation_steps == 0:
progress_bar.update(1)
completed_steps += 1
continue
with accelerator.accumulate(model):
outputs = model(**batch, use_cache=False)
loss = outputs.loss
# We keep track of the loss at each logged step
total_loss += loss.detach().float()
accelerator.backward(loss)
optimizer.step()
optimizer.zero_grad()
lr_scheduler.step()
# # Checks if the accelerator has performed an optimization step behind the scenes
if accelerator.sync_gradients:
progress_bar.update(1)
completed_steps += 1
if args.logging_steps and completed_steps % args.logging_steps == 0:
avg_loss = (
accelerator.gather(total_loss).mean().item()
/ args.gradient_accumulation_steps
/ args.logging_steps
)
logger.info(
f" Step: {completed_steps}, LR: {lr_scheduler.get_last_lr()[0]}, Loss: {avg_loss}"
)
if args.with_tracking:
accelerator.log(
{
"learning_rate": lr_scheduler.get_last_lr()[0],
"train_loss": avg_loss,
},
step=completed_steps,
)
total_loss = 0
if isinstance(checkpointing_steps, int):
if completed_steps % checkpointing_steps == 0:
output_dir = f"step_{completed_steps}"
if args.output_dir is not None:
output_dir = os.path.join(args.output_dir, output_dir)
unwrapped_model = accelerator.unwrap_model(model)
state_dict = accelerator.get_state_dict(model)
unwrapped_model.save_pretrained(
output_dir,
is_main_process=accelerator.is_main_process,
save_function=accelerator.save,
state_dict=state_dict,
)
tokenizer.save_pretrained(output_dir)
if completed_steps >= args.max_train_steps:
break
if args.checkpointing_steps == "epoch":
output_dir = f"epoch_{epoch}"
if args.output_dir is not None:
output_dir = os.path.join(args.output_dir, output_dir)
unwrapped_model = accelerator.unwrap_model(model)
state_dict = accelerator.get_state_dict(model)
unwrapped_model.save_pretrained(
output_dir,
is_main_process=accelerator.is_main_process,
save_function=accelerator.save,
state_dict=state_dict,
)
tokenizer.save_pretrained(output_dir)
if args.with_tracking:
accelerator.end_training()
if args.output_dir is not None:
accelerator.wait_for_everyone()
if accelerator.is_main_process:
tokenizer.save_pretrained(args.output_dir)
unwrapped_model = accelerator.unwrap_model(model)
# When doing multi-gpu training, we need to use accelerator.get_state_dict(model) to get the state_dict.
# Otherwise, sometimes the model will be saved with only part of the parameters.
# Also, accelerator needs to use the wrapped model to get the state_dict.
state_dict = accelerator.get_state_dict(model)
if args.use_lora:
# When using lora, the unwrapped model is a PeftModel, which doesn't support the is_main_process
# and has its own save_pretrained function for only saving lora modules.
# We have to mannually specify the is_main_process outside the save_pretrained function.
if accelerator.is_main_process:
unwrapped_model.save_pretrained(args.output_dir, state_dict=state_dict)
else:
unwrapped_model.save_pretrained(
args.output_dir,
is_main_process=accelerator.is_main_process,
save_function=accelerator.save,
state_dict=state_dict,
)
if __name__ == "__main__":
main()