-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_script.py
379 lines (321 loc) · 16.4 KB
/
run_script.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
# -*- coding:utf-8 -*-
import argparse
import glob
import logging
import os
import random
import copy
import math
import json
import numpy as np
import torch
from torch.nn import CrossEntropyLoss
from torch.utils.data import DataLoader, RandomSampler
from torch.utils.data.distributed import DistributedSampler
from tqdm import tqdm, trange
import sys
import pickle as pkl
from torch.nn import MSELoss
from transformers import (
WEIGHTS_NAME,
AdamW,
get_linear_schedule_with_warmup,
BertTokenizer,
BertConfig
)
from models.modeling_span import Span_Detector
from models.modeling_type import Type_Classifier
from utils.data_utils import load_and_cache_examples, get_labels
from utils.eval import evaluate
from utils.config import config
from utils.loss_utils import share_loss
logger = logging.getLogger(__name__)
MODEL_CLASSES = {
"span": (Span_Detector, BertConfig, BertTokenizer),
"type": (Type_Classifier, BertConfig, BertTokenizer),
}
torch.set_printoptions(profile="full")
def set_seed(args):
random.seed(args.seed)
np.random.seed(args.seed)
torch.manual_seed(args.seed)
if args.n_gpu > 0:
torch.cuda.manual_seed_all(args.seed)
def initialize(args, tokenizer, t_total, span_num_labels, type_num_labels_src, type_num_labels_tgt):
model_class, config_class, _ = MODEL_CLASSES["span"]
config = config_class.from_pretrained(
args.span_model_name_or_path,
cache_dir=args.cache_dir if args.cache_dir else None,
)
span_model = model_class.from_pretrained(
args.span_model_name_or_path,
config=config,
span_num_labels=span_num_labels,
device=args.device,
cache_dir=args.cache_dir if args.cache_dir else None,
)
span_model.to(args.device)
model_class, config_class, _ = MODEL_CLASSES["type"]
config = config_class.from_pretrained(
args.type_model_name_or_path,
cache_dir=args.cache_dir if args.cache_dir else None,
)
type_model = model_class.from_pretrained(
args.type_model_name_or_path,
config=config,
type_num_labels_src=type_num_labels_src,
type_num_labels_tgt=type_num_labels_tgt,
device=args.device,
domain=args.dataset,
cache_dir=args.cache_dir if args.cache_dir else None,
)
type_model.to(args.device)
no_decay = ["bias", "LayerNorm.weight"]
optimizer_grouped_parameters_span = [
{
"params": [p for n, p in span_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 span_model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0},
]
optimizer_span = AdamW(optimizer_grouped_parameters_span, lr=args.learning_rate, \
eps=args.adam_epsilon, betas=(args.adam_beta1, args.adam_beta2))
scheduler_span = get_linear_schedule_with_warmup(
optimizer_span, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
)
optimizer_grouped_parameters_type = [
{
"params": [p for n, p in type_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 type_model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0},
]
optimizer_type = AdamW(optimizer_grouped_parameters_type, lr=args.learning_rate, \
eps=args.adam_epsilon, betas=(args.adam_beta1, args.adam_beta2))
scheduler_type = get_linear_schedule_with_warmup(
optimizer_type, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
)
# Multi-gpu training (should be after apex fp16 initialization)
if args.n_gpu > 1:
span_model = torch.nn.DataParallel(span_model)
type_model = torch.nn.DataParallel(type_model)
# Distributed training (should be after apex fp16 initialization)
if args.local_rank != -1:
span_model = torch.nn.parallel.DistributedDataParallel(
span_model, device_ids=[args.local_rank], output_device=args.local_rank, find_unused_parameters=True
)
type_model = torch.nn.parallel.DistributedDataParallel(
type_model, device_ids=[args.local_rank], output_device=args.local_rank, find_unused_parameters=True
)
span_model.zero_grad()
type_model.zero_grad()
return span_model, type_model, optimizer_span, scheduler_span, optimizer_type, scheduler_type
def validation(args, span_model, type_model, tokenizer, id_to_label_span, pad_token_label_id, best_dev, test, best_dev_bio, test_bio,\
global_step, t_total, epoch, devs, tests):
best_dev, best_dev_bio, is_updated_dev = evaluate(devs, args, span_model, type_model, tokenizer, \
id_to_label_span, pad_token_label_id, best_dev, best_dev_bio, mode="dev", logger=logger, \
prefix='dev [Step {}/{} | Epoch {}/{}]'.format(global_step, t_total, epoch, args.num_train_epochs), verbose=False)
test, test_bio, _ = evaluate(tests, args, span_model, type_model, tokenizer, \
id_to_label_span, pad_token_label_id, test, test_bio, mode="test", logger=logger, \
prefix='test [Step {}/{} | Epoch {}/{}]'.format(global_step, t_total, epoch, args.num_train_epochs), verbose=False)
if args.local_rank in [-1, 0] and is_updated_dev:
path = os.path.join(args.output_dir, "checkpoint-best-span-dev")
logger.info("Saving span model checkpoint to %s", path)
if not os.path.exists(path):
os.makedirs(path)
model_to_save = (
span_model.module if hasattr(span_model, "module") else span_model
) # Take care of distributed/parallel training
model_to_save.save_pretrained(path)
path = os.path.join(args.output_dir, "checkpoint-best-type-dev")
logger.info("Saving type model checkpoint to %s", path)
if not os.path.exists(path):
os.makedirs(path)
model_to_save = (
type_model.module if hasattr(type_model, "module") else type_model
) # Take care of distributed/parallel training
model_to_save.save_pretrained(path)
tokenizer.save_pretrained(path)
return best_dev, test, best_dev_bio, test_bio, is_updated_dev
def cycle(iterable):
while True:
for x in iterable:
yield x
def train(args, train_dataset_src, train_dataset, id_to_label_span, id_to_label_type_src, id_to_label_type_tgt, tokenizer, pad_token_label_id):
""" Train the model """
# num_labels = len(labels)
span_num_labels = len(id_to_label_span)
type_num_labels_src = len(id_to_label_type_src)-1
type_num_labels_tgt = len(id_to_label_type_tgt)-1
args.train_batch_size_src = args.per_gpu_train_batch_size_src * max(1, args.n_gpu)
train_sampler_src = RandomSampler(train_dataset_src) if args.local_rank==-1 else DistributedSampler(train_dataset_src)
train_dataloader_src = DataLoader(train_dataset_src, sampler=train_sampler_src, batch_size=args.train_batch_size_src)
args.train_batch_size = args.per_gpu_train_batch_size * max(1, args.n_gpu)
train_sampler = RandomSampler(train_dataset) if args.local_rank==-1 else DistributedSampler(train_dataset)
train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.train_batch_size)
if args.max_steps > 0:
t_total = args.max_steps
args.num_train_epochs = args.max_steps//(len(train_dataloader_src)//args.gradient_accumulation_steps)+1
else:
t_total = len(train_dataloader_src)//args.gradient_accumulation_steps*50
span_model, type_model, optimizer_span, scheduler_span, \
optimizer_type, scheduler_type = initialize(args, tokenizer, t_total, span_num_labels, type_num_labels_src, type_num_labels_tgt)
logger.info("***** Running training *****")
logger.info(" Num examples of src = %d", len(train_dataset_src))
logger.info(" Num Epochs = %d", args.num_train_epochs)
logger.info(" Instantaneous batch size per GPU = %d", args.per_gpu_train_batch_size_src)
logger.info(
" Total train batch size (w. parallel, distributed & accumulation) = %d",
args.train_batch_size_src
* args.gradient_accumulation_steps
* (torch.distributed.get_world_size() if args.local_rank != -1 else 1),
)
logger.info(" Gradient Accumulation steps = %d", args.gradient_accumulation_steps)
logger.info(" Total optimization steps = %d", t_total)
global_step = 0
epochs_trained = 0
tr_loss, logging_loss = 0.0, 0.0
train_iterator = trange(
epochs_trained, int(args.num_train_epochs), desc="Epoch", disable=args.local_rank not in [-1, 0]
)
set_seed(args) # Added here for reproductibility
best_dev, test = [0, 0, 0], [0, 0, 0]
best_dev_bio, test_bio = [0, 0, 0], [0, 0, 0]
devs = []
tests = []
loss_funct = MSELoss()
iterator = iter(cycle(train_dataloader))
for epoch in train_iterator:
epoch_iterator = tqdm(train_dataloader_src, desc="Iteration", disable=args.local_rank not in [-1, 0])
for step, batch_src in enumerate(epoch_iterator):
span_model.train()
type_model.train()
batch_src = tuple(t.to(args.device) for t in batch_src)
batch = next(iterator)
batch = tuple(t.to(args.device) for t in batch)
inputs = {"input_ids": batch_src[0], "attention_mask": batch_src[1], "labels_bio": batch_src[2], "tgt": False, "reduction": "none"}
outputs_span_src = span_model(**inputs)
inputs = {"input_ids": batch_src[0], "attention_mask": batch_src[1], "labels_type": batch_src[3], "logits_bio": outputs_span_src[2], "tgt": False, "reduction": "none"}
outputs_type_src = type_model(**inputs)
loss1 = span_model.loss(outputs_span_src[0], outputs_type_src[1], tau=args.tau_span, eps=args.eps_span)
loss2 = type_model.loss(outputs_type_src[0], outputs_span_src[1], tau=args.tau_type, eps=args.eps_type)
loss6 = share_loss(outputs_span_src[4], outputs_type_src[4], loss_funct, args.L)
inputs = {"input_ids": batch[0], "attention_mask": batch[1], "labels_bio": batch[2], "tgt": True, "reduction": "none"}
outputs_span = span_model(**inputs)
inputs = {"input_ids": batch[0], "attention_mask": batch[1], "labels_type": batch[3], "logits_bio": outputs_span[2], "tgt": True, "reduction": "none"}
outputs_type = type_model(**inputs)
loss3 = span_model.loss(outputs_span[0], outputs_type[1], tau=args.tau_span, eps=args.eps_span)
permute_embed = span_model.adv_attack(outputs_span[4][0], loss3, mu=args.mu)
inputs = {"inputs_embeds": permute_embed, "attention_mask": batch[1], "labels_bio": batch[2], "tgt": True, "reduction": "mean"}
outputs_span_ = span_model(**inputs)
loss31 = outputs_span_[0]
loss4 = type_model.loss(outputs_type[0], outputs_span[1], tau=args.tau_type, eps=args.eps_type)
permute_embed = type_model.adv_attack(outputs_type[4][0], loss4, mu=args.mu)
inputs = {"inputs_embeds": permute_embed, "attention_mask": batch[1], "labels_type": batch[3], "tgt": True, "reduction": "mean"}
outputs_type_ = type_model(**inputs)
loss41 = outputs_type_[0]
loss7 = share_loss(outputs_span[4], outputs_type[4], loss_funct, args.L)
loss5 = type_model.mix_up(outputs_type_src[3], outputs_type[3], batch_src[3], batch[3], args.alpha, args.beta)
loss = loss1+loss2+loss3+loss4+0.1*loss5+0.1*(loss6+loss7)+0.1*(loss31+loss41) # ALL
if args.n_gpu > 1:
loss = loss.mean() # mean() to average on multi-gpu parallel training
if args.gradient_accumulation_steps > 1:
loss = loss/args.gradient_accumulation_steps
loss.backward()
tr_loss += loss.item()
if (step+1)%args.gradient_accumulation_steps == 0:
optimizer_span.step()
optimizer_type.step()
scheduler_span.step() # Update learning rate schedule
scheduler_type.step()
span_model.zero_grad()
type_model.zero_grad()
global_step += 1
if args.local_rank in [-1, 0] and args.logging_steps > 0 and global_step%args.logging_steps == 0:
# Log metrics
# iters.append(global_step)
if args.evaluate_during_training:
logger.info("***** training loss : %.4f*****", loss.item())
best_dev, test, best_dev_bio, test_bio, _ = validation(args, span_model, type_model, tokenizer, \
id_to_label_span, pad_token_label_id, best_dev, test, best_dev_bio, test_bio,\
global_step, t_total, epoch, devs, tests)
if args.max_steps > 0 and global_step > args.max_steps:
epoch_iterator.close()
break
if args.max_steps > 0 and global_step > args.max_steps:
train_iterator.close()
break
results = (best_dev, best_test)
return results
def main():
args = config()
args.do_train = args.do_train.lower()
# args.do_test = args.do_test.lower()
if (
os.path.exists(args.output_dir)
and os.listdir(args.output_dir)
and args.do_train
and not args.overwrite_output_dir
):
raise ValueError(
"Output directory ({}) already exists and is not empty. Use --overwrite_output_dir to overcome.".format(
args.output_dir
)
)
# Create output directory if needed
if not os.path.exists(args.output_dir) and args.local_rank in [-1, 0]:
os.makedirs(args.output_dir)
if args.local_rank == -1 or args.no_cuda:
device = torch.device("cuda" if torch.cuda.is_available() and not args.no_cuda else "cpu")
args.n_gpu = torch.cuda.device_count()
else: # Initializes the distributed backend which will take care of sychronizing nodes/GPUs
torch.cuda.set_device(args.local_rank)
device = torch.device("cuda", args.local_rank)
torch.distributed.init_process_group(backend="nccl")
args.n_gpu = 1
args.device = device
# Setup logging
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN,
)
formatter = logging.Formatter("%(asctime)s - %(levelname)s - %(name)s - %(message)s", "%m/%d/%Y %H:%M:%S")
logging_fh = logging.FileHandler(os.path.join(args.output_dir, 'log.txt'))
logging_fh.setLevel(logging.DEBUG)
logging_fh.setFormatter(formatter)
logger.addHandler(logging_fh)
logger.warning(
"Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s",
args.local_rank,
device,
args.n_gpu,
bool(args.local_rank != -1),
args.fp16,
)
# Set seed
set_seed(args)
id_to_label_span, id_to_label_type_tgt, id_to_label_type_src = get_labels(args.data_dir, args.src_dataset, args.dataset)
# num_labels = len(labels)
# Use cross entropy ignore index as padding label id so that only real label ids contribute to the loss later
pad_token_label_id = CrossEntropyLoss().ignore_index
# Load pretrained model and tokenizer
if args.local_rank not in [-1, 0]:
torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab
tokenizer = MODEL_CLASSES["span"][2].from_pretrained(
args.tokenizer_name_or_path,
do_lower_case=args.do_lower_case,
cache_dir=args.cache_dir if args.cache_dir else None,
)
if args.local_rank == 0:
torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab
logger.info("Training/evaluation parameters %s", args)
# Loss = CycleConsistencyLoss(non_entity_id, args.device)
# Training
if args.do_train == "true":
train_dataset_src, train_dataset = load_and_cache_examples(args, tokenizer, pad_token_label_id, mode="train")
best_results = train(args, train_dataset_src, train_dataset, \
id_to_label_span, id_to_label_type_src, id_to_label_type_tgt, tokenizer, pad_token_label_id)
# logger.info(" global_step = %s, average loss = %s", global_step, tr_loss)
if __name__ == "__main__":
main()