-
Notifications
You must be signed in to change notification settings - Fork 0
/
pretrain_Bert_distributed_horovod.py
501 lines (409 loc) · 19.8 KB
/
pretrain_Bert_distributed_horovod.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
# coding=utf-8
'''
@Software:PyCharm
@Time:2024/04/07 2:58 下午
@Author: fffan
'''
from __future__ import absolute_import, division, print_function
import os
#os.environ["CUDA_VISIBLE_DEVICES"] = "4"
import re
import argparse
import logging
import random
import pickle
import numpy as np
import torch
from collections import namedtuple
from torch.utils.data import (DataLoader, RandomSampler, Dataset)
from torch.utils.data.distributed import DistributedSampler
from tqdm import tqdm
import horovod.torch as hvd
from torch.nn import CrossEntropyLoss
import collections
from torchinfo import summary
from model_info.file_utils import WEIGHTS_NAME, CONFIG_NAME
from model_info.modeling import BertForPreTraining
from model_info.tokenization import BertTokenizer
from model_info.optimization import BertAdam
logging.basicConfig(format='%(asctime)s - %(levelname)s - %(name)s - %(message)s',
datefmt='%m/%d/%Y %H:%M:%S',
level=logging.INFO)
logger = logging.getLogger(__name__)
InputFeatures = namedtuple("InputFeatures", "input_ids input_masks segment_ids masked_lm_positions masked_lm_ids masked_lm_weights")
hvd.init()
logger.info("hvd.local_rank:{} ".format(hvd.local_rank()))
logger.info("hvd.rank:{} ".format(hvd.rank()))
logger.info("hvd.local_size:{} ".format(hvd.local_size()))
logger.info("hvd.size:{} ".format(hvd.size()))
torch.cuda.set_device(hvd.local_rank())
os.environ["CUDA_VISIBLE_DEVICES"] = str(hvd.local_rank())
device = 'cuda'
MaskedLmInstance = collections.namedtuple("MaskedLmInstance",
["index", "label"])
def create_masked_lm_predictions(tokens, masked_lm_prob,
max_predictions_per_seq, vocab_words, rng):
"""Creates the predictions for the masked LM objective."""
cand_indexes = []
for (i, token) in enumerate(tokens):
if token == "[CLS]" or token == "[SEP]":
continue
# Whole Word Masking means that if we mask all of the wordpieces
# corresponding to an original word. When a word has been split into
# WordPieces, the first token does not have any marker and any subsequence
# tokens are prefixed with ##. So whenever we see the ## token, we
# append it to the previous set of word indexes.
#
# Note that Whole Word Masking does *not* change the training code
# at all -- we still predict each WordPiece independently, softmaxed
# over the entire vocabulary.
do_whole_word_mask = True
if (do_whole_word_mask and len(cand_indexes) >= 1 and
token.startswith("##")):
cand_indexes[-1].append(i)
else:
cand_indexes.append([i])
rng.shuffle(cand_indexes)
output_tokens = [t[2:] if len(re.findall('##[\u4E00-\u9FA5]', t))>0 else t for t in tokens]
num_to_predict = min(max_predictions_per_seq,
max(1, int(round(len(tokens) * masked_lm_prob))))
masked_lms = []
covered_indexes = set()
for index_set in cand_indexes:
if len(masked_lms) >= num_to_predict:
break
# If adding a whole-word mask would exceed the maximum number of
# predictions, then just skip this candidate.
if len(masked_lms) + len(index_set) > num_to_predict:
continue
is_any_index_covered = False
for index in index_set:
if index in covered_indexes:
is_any_index_covered = True
break
if is_any_index_covered:
continue
for index in index_set:
covered_indexes.add(index)
masked_token = None
# 80% of the time, replace with [MASK]
if rng.random() < 0.8:
masked_token = "[MASK]"
else:
# 10% of the time, keep original
if rng.random() < 0.5:
masked_token = tokens[index][2:] if len(re.findall('##[\u4E00-\u9FA5]', tokens[index]))>0 else tokens[index]
# 10% of the time, replace with random word
else:
masked_token = vocab_words[rng.randint(0, len(vocab_words) - 1)]
output_tokens[index] = masked_token
masked_lms.append(MaskedLmInstance(index=index, label=tokens[index]))
assert len(masked_lms) <= num_to_predict
masked_lms = sorted(masked_lms, key=lambda x: x.index)
masked_lm_positions = []
masked_lm_labels = []
for p in masked_lms:
masked_lm_positions.append(p.index)
masked_lm_labels.append(p.label)
return (output_tokens, masked_lm_positions, masked_lm_labels)
def convert_example_to_features(text, tokenizer, max_seq_len):
"""输入text格式:
1): 单句
2): 双句,以\t分隔,并且分隔后这两个句子的0,1索引位置
"""
# 本项目为了方便,以单句的语料为例
sents = text.split('\t')[:1]
tokens = ['[CLS]'] + tokenizer.tokenize(sents[0])[:max_seq_len - 2] + ['[SEP]']
vocab_words = list(tokenizer.vocab.keys())
rng = random.Random()
(tokens, masked_lm_positions,
masked_lm_labels) = create_masked_lm_predictions(tokens, masked_lm_prob=0.1,
max_predictions_per_seq=23, vocab_words=vocab_words, rng=rng)
input_ids = tokenizer.convert_tokens_to_ids(tokens)
segment_ids = len(input_ids) * [0]
if len(sents) > 1:
token_b = tokenizer.tokenize(sents[1])[:max_seq_len - 2] + ['[SEP]']
input_ids += tokenizer.convert_tokens_to_ids(token_b)
segment_ids += len(token_b) * [1]
input_array = np.zeros(max_seq_len, dtype=np.int)
input_array[:len(input_ids)] = input_ids
mask_array = np.zeros(max_seq_len, dtype=np.bool)
mask_array[:len(input_ids)] = 1
segment_array = np.zeros(max_seq_len, dtype=np.bool)
segment_array[:len(segment_ids)] = segment_ids
masked_lm_positions = list(masked_lm_positions)
masked_lm_ids = tokenizer.convert_tokens_to_ids(masked_lm_labels)
masked_lm_weights = [1.0] * len(masked_lm_ids)
max_predictions_per_seq = 23
while len(masked_lm_positions) < max_predictions_per_seq:
masked_lm_positions.append(0)
masked_lm_ids.append(0)
masked_lm_weights.append(0.0)
masked_lm_positions = np.array(masked_lm_positions)
masked_lm_ids = np.array(masked_lm_ids)
masked_lm_weights = np.array(masked_lm_weights)
feature = InputFeatures(input_ids=input_array,
input_masks=mask_array,
segment_ids=segment_array,
masked_lm_positions=masked_lm_positions,
masked_lm_ids=masked_lm_ids,
masked_lm_weights=masked_lm_weights
)
return feature
class PreTrainingDataset(Dataset):
""" 实际的供给 Dataloader 的数据类 """
def __init__(self):
self.data = []
def add_instance(self, features: collections.OrderedDict):
self.data.append((
features["input_ids"],
features["segment_ids"],
features["input_mask"],
features["masked_lm_ids"],
features["next_sentence_labels"]
))
def __len__(self):
return len(self.data)
def __getitem__(self, index):
input_ids, token_type_ids, attention_mask, masked_lm_labels, next_sentence_label = self.data[index]
return input_ids, token_type_ids, attention_mask, masked_lm_labels, next_sentence_label
class PregeneratedDataset(object):
def __init__(self, training_path, tokenizer, max_seq_len):
self.vocab = tokenizer.vocab
self.tokenizer = tokenizer
logger.info('training_path: {}'.format(training_path))
self.input_ids = []
self.segment_ids = []
self.input_masks = []
self.masked_lm_ids = []
self.next_sentence_labels = []
print("##### 开始读取数据:",training_path)
with open(training_path, 'rb') as f:
one_pkl_dict = pickle.load(f)
for i, feature in enumerate(tqdm(one_pkl_dict.data)):
if not feature:
continue
"""
self.input_ids.append(feature.input_ids)
self.segment_ids.append(feature.segment_ids)
self.input_masks.append(feature.input_masks)
self.masked_lm_ids.append(feature.masked_lm_ids)
self.next_sentence_labels.append(feature.next_sentence_labels)
"""
self.input_ids.append(feature[0])
self.segment_ids.append(feature[1])
self.input_masks.append(feature[2])
self.masked_lm_ids.append(feature[3])
self.next_sentence_labels.append(feature[4])
self.data_size = len(self.input_ids)
def __len__(self):
return self.data_size
def __getitem__(self, item):
return (
torch.tensor(self.input_ids[item]),
torch.tensor(self.input_masks[item]),
torch.tensor(self.segment_ids[item]),
torch.tensor(self.masked_lm_ids[item]),
torch.tensor(self.next_sentence_labels[item]),
)
def save_model(prefix, model, path):
logging.info("** ** * Saving model ** ** * ")
model_name = "{}_{}".format(prefix, WEIGHTS_NAME)
model_to_save = model.module if hasattr(model, 'module') else model
output_model_file = os.path.join(path, model_name)
output_config_file = os.path.join(path, CONFIG_NAME)
torch.save(model_to_save.state_dict(), output_model_file)
model_to_save.config.to_json_file(output_config_file)
def get_parser():
parser = argparse.ArgumentParser()
parser.add_argument("--train_file_path", default="./data/bert_chinese_fffan_data.eval", type=str)
# Required parameters
parser.add_argument("--model_path", default="./pretrain_models/bert-base-chinese", type=str)
parser.add_argument("--output_dir", default="./output_dir/pretrain_fffan", type=str)
# Other parameters
parser.add_argument("--save_model_number",
default=5,
type=int, help="The maximum total input sequence length ")
parser.add_argument("--max_seq_len",
default=128,
type=int,
help="The maximum total input sequence length after WordPiece \n"
" tokenization. Sequences longer than this will be truncated, \n"
"and sequences shorter than this will be padded.")
parser.add_argument("--do_eval",
action='store_true',
help="Whether to run eval on the dev set.")
parser.add_argument("--do_lower_case",
action='store_true',
# default=True,
help="Set this flag if you are using an uncased model.")
parser.add_argument("--train_batch_size",
default=8,
type=int,
help="Total batch size for training.")
parser.add_argument("--eval_batch_size", default=1, type=int, help="Total batch size for eval.")
parser.add_argument("--learning_rate",
default=5e-5,
type=float,
help="The initial learning rate for Adam.")
parser.add_argument('--weight_decay',
'--wd',
default=1e-1,
type=float,
metavar='W',
help='weight decay')
parser.add_argument("--num_train_epochs",
default=3.0,
type=float,
help="Total number of training epochs to perform.")
parser.add_argument("--warmup_proportion",
default=0.1,
type=float,
help="Proportion of training to perform linear learning rate warmup for. "
"E.g., 0.1 = 10%% of training.")
parser.add_argument("--no_cuda",
action='store_true',
help="Whether not to use CUDA when available")
parser.add_argument("--local_rank",
type=int,
default=-1,
help="local_rank for distributed training on gpus")
parser.add_argument('--seed', type=int, default=42, help="random seed for initialization")
parser.add_argument('--gradient_accumulation_steps',
type=int,
default=1,
help="Number of updates steps to accumulate before performing \n"
"a backward/update pass.")
parser.add_argument('--fp16',
action='store_true',
help="Whether to use 16-bit float precision instead of 32-bit")
parser.add_argument('--continue_train',
action='store_true',
help='Whether to train from checkpoints')
# Additional arguments
parser.add_argument('--eval_step', type=int, default=5)
# This is used for running on Huawei Cloud.
parser.add_argument('--data_url', type=str, default="")
args = parser.parse_args()
return args
def main():
args = get_parser()
output_dir = args.output_dir
os.makedirs(output_dir, exist_ok=True)
torch.backends.cudnn.deterministic = True
logger.info('args:{}'.format(args))
tokenizer = BertTokenizer.from_pretrained(args.model_path, do_lower_case=args.do_lower_case)
if os.path.exists(os.path.join(args.model_path, "vocab.txt")):
os.system("cp " + os.path.join(args.model_path, "vocab.txt") + " " + args.output_dir)
model = BertForPreTraining.from_scratch(args.model_path)
model.to(device)
# logging
logger.info(f"[Data]: Reading data...\n")
dataset = PregeneratedDataset(args.train_file_path, tokenizer, max_seq_len=args.max_seq_len)
total_train_examples = len(dataset)
print("##### 训练数据量:", total_train_examples)
num_train_optimization_steps = int(total_train_examples / args.train_batch_size /
args.gradient_accumulation_steps * args.num_train_epochs)
print("##### 训练数据总步数:", num_train_optimization_steps)
train_sampler = torch.utils.data.distributed.DistributedSampler(dataset, num_replicas=hvd.size(),
rank=hvd.rank(), shuffle=True)
train_params = {
"batch_size": args.train_batch_size,
"num_workers": 0,
}
# Creation of Dataloaders for testing and validation.
training_loader = DataLoader(dataset, batch_size=train_params["batch_size"], shuffle=False,
sampler=train_sampler)
#######################################################
size = 0
for n, p in model.named_parameters():
logger.info('n: {}'.format(n))
logger.info('p: {}'.format(p.nelement()))
size += p.nelement()
logger.info('Total parameters: {}'.format(size))
# Prepare optimizer
param_optimizer = list(model.named_parameters())
no_decay = ['bias', 'LayerNorm.bias', 'LayerNorm.weight']
optimizer_grouped_parameters = [{
'params': [p for n, p in param_optimizer if not any(nd in n for nd in no_decay)],
'weight_decay': 0.01
}, {
'params': [p for n, p in param_optimizer if any(nd in n for nd in no_decay)],
'weight_decay': 0.0
}]
optimizer = BertAdam(optimizer_grouped_parameters,
lr=args.learning_rate,
warmup=args.warmup_proportion,
t_total=num_train_optimization_steps)
optimizer = hvd.DistributedOptimizer(optimizer, backward_passes_per_step=args.gradient_accumulation_steps)
hvd.broadcast_parameters(model.state_dict(), root_rank=0)
# Training loop
logger.info(f"[Initiating Fine Tuning]...\n")
logger.info("the length of dataloader is: {}".format(len(training_loader)))
logging.info("***** Running training *****")
logging.info(" Num examples = {}".format(total_train_examples))
logging.info(" Batch size = %d", args.train_batch_size)
logging.info(" Num steps = %d", num_train_optimization_steps)
model.train()
global_step = 0
all_save_model_list = []
output_model_file = None
for epoch in tqdm(range(int(args.num_train_epochs)), desc="## Epoch", ascii=True):
####
for step, batch in enumerate(tqdm(training_loader, desc="# Iteration", ascii=True)):
#####
batch = tuple(t.to(device) for t in batch)
input_ids, input_mask, segment_ids,masked_lm_ids,next_sentence_labels = batch
if input_ids.size()[0] != args.train_batch_size:
continue
loss = model(input_ids, segment_ids, input_mask, masked_lm_ids,next_sentence_labels)
loss = loss.mean() # mean() to average on multi-gpu.
if args.fp16:
optimizer.backward(loss)
else:
loss.backward()
if epoch == 0 and step == 0 and hvd.rank() == 0: ### 只在训练开头打印参数信息,其他时候不打印
summary(model=model, input_ids=input_ids, segment_ids=segment_ids,input_mask=input_mask, masked_lm_ids=masked_lm_ids, next_sentence_labels=next_sentence_labels,
device=device)
if step % 100 == 0:
logger.info(f'loss = {loss}')
if (step + 1) % args.gradient_accumulation_steps == 0:
optimizer.step()
optimizer.zero_grad()
global_step += 1
if (global_step + 1) % args.eval_step == 0 :
result = {}
result['global_step'] = global_step
result['loss'] = loss
output_eval_file = os.path.join(args.output_dir, "log.txt")
with open(output_eval_file, "a") as writer:
logger.info("***** Eval results *****")
for key in sorted(result.keys()):
logger.info(" %s = %s", key, str(result[key]))
writer.write("%s = %s\n" % (key, str(result[key])))
if hvd.rank() == 0:
# Save a trained model
########################################################################
prefix = f"step_{global_step}"
logging.info("** ** * Saving model ** ** * ")
model_name = "{}_{}".format(prefix, WEIGHTS_NAME)
model_to_save = model.module if hasattr(model, 'module') else model
output_model_file = os.path.join(args.output_dir, model_name)
output_config_file = os.path.join(args.output_dir, CONFIG_NAME)
torch.save(model_to_save.state_dict(), output_model_file)
model_to_save.config.to_json_file(output_config_file)
##### 保存的模型数量超过指定数量,删除模型 #############
if len(all_save_model_list) == args.save_model_number:
os.system("rm -rf " + all_save_model_list[0])
print("#### 删除模型:", all_save_model_list[0])
del all_save_model_list[0]
######################################################
all_save_model_list.append(output_model_file)
if output_model_file:
output_list = output_model_file.split("/")[:-1] + ["pytorch_model.bin"]
cp_path = "/".join(output_list)
cp_line = "mv " + output_model_file + " " + cp_path
print("##### mv model path: ", cp_line)
os.system(cp_line)
if __name__ == "__main__":
main()