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
| # basic python packages
import argparse
import glob
import itertools
import json
import logging
import pickle
import random
from argparse import Namespace
from typing import Dict
import numpy as np
logger = logging.getLogger(__name__)
# pytorch lightning packages
import pytorch_lightning as pl
# pytorch packages
import torch
import torch.nn as nn
import torch.nn.functional as F
from pytorch_lightning.callbacks import ModelCheckpoint
from pytorch_lightning.loggers import TensorBoardLogger
from torch.utils.data import DataLoader, IterableDataset
# huggingface transformers packages
from transformers import AdamW, AutoModel, AutoTokenizer
from transformers.optimization import (
Adafactor,
get_cosine_schedule_with_warmup,
get_cosine_with_hard_restarts_schedule_with_warmup,
get_linear_schedule_with_warmup,
get_polynomial_decay_schedule_with_warmup,
)
# allennlp dataloading packages
from allennlp.data.dataset_readers.dataset_reader import DatasetReader
from allennlp.data.token_indexers import TokenIndexer
from allennlp.data.tokenizers import Tokenizer
from allennlp.data.tokenizers.token import Token
from allennlp.data.tokenizers.word_splitter import WordSplitter
# Globe constants
training_size = 684100
# validation_size = 145375
# log_every_n_steps how frequently pytorch lightning logs.
# By default, Lightning logs every 50 rows, or 50 training steps.
log_every_n_steps = 1
arg_to_scheduler = {
"linear": get_linear_schedule_with_warmup,
"cosine": get_cosine_schedule_with_warmup,
"cosine_w_restarts": get_cosine_with_hard_restarts_schedule_with_warmup,
"polynomial": get_polynomial_decay_schedule_with_warmup,
# '': get_constant_schedule, # not supported for now
# '': get_constant_schedule_with_warmup, # not supported for now
}
arg_to_scheduler_choices = sorted(arg_to_scheduler.keys())
arg_to_scheduler_metavar = "{" + ", ".join(arg_to_scheduler_choices) + "}"
class DataReaderFromPickled(DatasetReader):
"""
This is copied from https://github.com/allenai/specter/blob/673346f9f76bcf422b38e0d1b448ef4414bcd4df/specter/data.py#L61:L109 without any change
"""
def __init__(self,
lazy: bool = False,
word_splitter: WordSplitter = None,
tokenizer: Tokenizer = None,
token_indexers: Dict[str, TokenIndexer] = None,
max_sequence_length: int = 256,
concat_title_abstract: bool = None
) -> None:
"""
Dataset reader that uses pickled preprocessed instances
Consumes the output resulting from data_utils/create_training_files.py
the additional arguments are not used here and are for compatibility with
the other data reader at prediction time
"""
self.max_sequence_length = max_sequence_length
self.token_indexers = token_indexers
self._concat_title_abstract = concat_title_abstract
super().__init__(lazy)
def _read(self, file_path: str):
"""
Args:
file_path: path to the pickled instances
"""
with open(file_path, 'rb') as f_in:
unpickler = pickle.Unpickler(f_in)
while True:
try:
instance = unpickler.load()
# compatibility with old models:
# for field in instance.fields:
# if hasattr(instance.fields[field], '_token_indexers') and 'bert' in instance.fields[field]._token_indexers:
# if not hasattr(instance.fields['source_title']._token_indexers['bert'], '_truncate_long_sequences'):
# instance.fields[field]._token_indexers['bert']._truncate_long_sequences = True
# instance.fields[field]._token_indexers['bert']._token_min_padding_length = 0
if self.max_sequence_length:
for paper_type in ['source', 'pos', 'neg']:
if self._concat_title_abstract:
tokens = []
title_field = instance.fields.get(f'{paper_type}_title')
abst_field = instance.fields.get(f'{paper_type}_abstract')
if title_field:
tokens.extend(title_field.tokens)
if tokens:
tokens.extend([Token('[SEP]')])
if abst_field:
tokens.extend(abst_field.tokens)
if title_field:
title_field.tokens = tokens
instance.fields[f'{paper_type}_title'] = title_field
elif abst_field:
abst_field.tokens = tokens
instance.fields[f'{paper_type}_title'] = abst_field
else:
yield None
# title_tokens = get_text_tokens(query_title_tokens, query_abstract_tokens, abstract_delimiter)
# pos_title_tokens = get_text_tokens(pos_title_tokens, pos_abstract_tokens, abstract_delimiter)
# neg_title_tokens = get_text_tokens(neg_title_tokens, neg_abstract_tokens, abstract_delimiter)
# query_abstract_tokens = pos_abstract_tokens = neg_abstract_tokens = []
for field_type in ['title', 'abstract', 'authors', 'author_positions']:
field = paper_type + '_' + field_type
if instance.fields.get(field):
instance.fields[field].tokens = instance.fields[field].tokens[
:self.max_sequence_length]
if field_type == 'abstract' and self._concat_title_abstract:
instance.fields.pop(field, None)
yield instance
except EOFError:
break
class IterableDataSetMultiWorker(IterableDataset):
def __init__(self, file_path, tokenizer, size, block_size=100):
self.datareaderfp = DataReaderFromPickled(max_sequence_length=512)
self.data_instances = self.datareaderfp._read(file_path)
self.tokenizer = tokenizer
self.size = size
self.block_size = block_size
def __iter__(self):
worker_info = torch.utils.data.get_worker_info()
if worker_info is None:
iter_end = self.size
for data_instance in itertools.islice(self.data_instances, iter_end):
data_input = self.ai2_to_transformers(data_instance, self.tokenizer)
yield data_input
else:
# when num_worker is greater than 1. we implement multiple process data loading.
iter_end = self.size
worker_id = worker_info.id
num_workers = worker_info.num_workers
i = 0
for data_instance in itertools.islice(self.data_instances, iter_end):
if int(i / self.block_size) % num_workers != worker_id:
i = i + 1
pass
else:
i = i + 1
data_input = self.ai2_to_transformers(data_instance, self.tokenizer)
yield data_input
def ai2_to_transformers(self, data_instance, tokenizer):
"""
Args:
data_instance: ai2 data instance
tokenizer: huggingface transformers tokenizer
"""
source_tokens = data_instance["source_title"].tokens
source_title = tokenizer(' '.join([str(token) for token in source_tokens]),
truncation=True, padding="max_length", return_tensors="pt",
max_length=512)
source_input = {'input_ids': source_title['input_ids'][0],
'token_type_ids': source_title['token_type_ids'][0],
'attention_mask': source_title['attention_mask'][0]}
pos_tokens = data_instance["pos_title"].tokens
pos_title = tokenizer(' '.join([str(token) for token in pos_tokens]),
truncation=True, padding="max_length", return_tensors="pt", max_length=512)
pos_input = {'input_ids': pos_title['input_ids'][0],
'token_type_ids': pos_title['token_type_ids'][0],
'attention_mask': pos_title['attention_mask'][0]}
neg_tokens = data_instance["neg_title"].tokens
neg_title = tokenizer(' '.join([str(token) for token in neg_tokens]),
truncation=True, padding="max_length", return_tensors="pt", max_length=512)
neg_input = {'input_ids': neg_title['input_ids'][0],
'token_type_ids': neg_title['token_type_ids'][0],
'attention_mask': neg_title['attention_mask'][0]}
return source_input, pos_input, neg_input
class IterableDataSetMultiWorkerTestStep(IterableDataset):
def __init__(self, file_path, tokenizer, size, block_size=100):
self.datareaderfp = DataReaderFromPickled(max_sequence_length=512)
self.data_instances = self.datareaderfp._read(file_path)
self.tokenizer = tokenizer
self.size = size
self.block_size = block_size
def __iter__(self):
worker_info = torch.utils.data.get_worker_info()
if worker_info is None:
iter_end = self.size
for data_instance in itertools.islice(self.data_instances, iter_end):
data_input = self.ai2_to_transformers(data_instance, self.tokenizer)
yield data_input
else:
# when num_worker is greater than 1. we implement multiple process data loading.
iter_end = self.size
worker_id = worker_info.id
num_workers = worker_info.num_workers
i = 0
for data_instance in itertools.islice(self.data_instances, iter_end):
if int(i / self.block_size) % num_workers != worker_id:
i = i + 1
pass
else:
i = i + 1
data_input = self.ai2_to_transformers(data_instance, self.tokenizer)
yield data_input
def ai2_to_transformers(self, data_instance, tokenizer):
"""
Args:
data_instance: ai2 data instance
tokenizer: huggingface transformers tokenizer
"""
source_tokens = data_instance["source_title"].tokens
source_title = tokenizer(' '.join([str(token) for token in source_tokens]),
truncation=True, padding="max_length", return_tensors="pt",
max_length=512)
source_input = {'input_ids': source_title['input_ids'][0],
'token_type_ids': source_title['token_type_ids'][0],
'attention_mask': source_title['attention_mask'][0]}
source_paper_id = data_instance['source_paper_id'].metadata
return source_input, source_paper_id
class TripletLoss(nn.Module):
"""
Triplet loss: copied from https://github.com/allenai/specter/blob/673346f9f76bcf422b38e0d1b448ef4414bcd4df/specter/model.py#L159 without any change
"""
def __init__(self, margin=1.0, distance='l2-norm', reduction='mean'):
"""
Args:
margin: margin (float, optional): Default: `1`.
distance: can be `l2-norm` or `cosine`, or `dot`
reduction (string, optional): Specifies the reduction to apply to the output:
'none' | 'mean' | 'sum'. 'none': no reduction will be applied,
'mean': the sum of the output will be divided by the number of
elements in the output, 'sum': the output will be summed. Note: :attr:`size_average`
and :attr:`reduce` are in the process of being deprecated, and in the meantime,
specifying either of those two args will override :attr:`reduction`. Default: 'mean'
"""
super(TripletLoss, self).__init__()
self.margin = margin
self.distance = distance
self.reduction = reduction
def forward(self, query, positive, negative):
if self.distance == 'l2-norm':
distance_positive = F.pairwise_distance(query, positive)
distance_negative = F.pairwise_distance(query, negative)
losses = F.relu(distance_positive - distance_negative + self.margin)
elif self.distance == 'cosine': # independent of length
distance_positive = F.cosine_similarity(query, positive)
distance_negative = F.cosine_similarity(query, negative)
losses = F.relu(-distance_positive + distance_negative + self.margin)
elif self.distance == 'dot': # takes into account the length of vectors
shapes = query.shape
# batch dot product
distance_positive = torch.bmm(
query.view(shapes[0], 1, shapes[1]),
positive.view(shapes[0], shapes[1], 1)
).reshape(shapes[0], )
distance_negative = torch.bmm(
query.view(shapes[0], 1, shapes[1]),
negative.view(shapes[0], shapes[1], 1)
).reshape(shapes[0], )
losses = F.relu(-distance_positive + distance_negative + self.margin)
else:
raise TypeError(f"Unrecognized option for `distance`:{self.distance}")
if self.reduction == 'mean':
return losses.mean()
elif self.reduction == 'sum':
return losses.sum()
elif self.reduction == 'none':
return losses
else:
raise TypeError(f"Unrecognized option for `reduction`:{self.reduction}")
class Specter(pl.LightningModule):
def __init__(self, init_args):
super().__init__()
if isinstance(init_args, dict):
# for loading the checkpoint, pl passes a dict (hparams are saved as dict)
init_args = Namespace(**init_args)
checkpoint_path = init_args.checkpoint_path
logger.info(f'loading model from checkpoint: {checkpoint_path}')
self.hparams = init_args
self.model = AutoModel.from_pretrained("allenai/scibert_scivocab_cased")
self.tokenizer = AutoTokenizer.from_pretrained("allenai/scibert_scivocab_cased")
self.tokenizer.model_max_length = self.model.config.max_position_embeddings
self.hparams.seqlen = self.model.config.max_position_embeddings
self.triple_loss = TripletLoss()
# number of training instances
self.training_size = None
# number of testing instances
self.validation_size = None
# number of test instances
self.test_size = None
# This is a dictionary to save the embeddings for source papers in test step.
self.embedding_output = {}
def forward(self, input_ids, token_type_ids, attention_mask):
# in lightning, forward defines the prediction/inference actions
source_embedding = self.model(input_ids=input_ids, token_type_ids=token_type_ids, attention_mask=attention_mask)
return source_embedding[1]
def _get_loader(self, split):
if split == 'train':
fname = self.hparams.train_file
size = self.training_size
elif split == 'dev':
fname = self.hparams.dev_file
size = self.validation_size
elif split == 'test':
fname = self.hparams.test_file
size = self.test_size
else:
assert False
if split == 'test':
dataset = IterableDataSetMultiWorkerTestStep(file_path=fname, tokenizer=self.tokenizer, size=size)
else:
dataset = IterableDataSetMultiWorker(file_path=fname, tokenizer=self.tokenizer, size=size)
# pin_memory enables faster data transfer to CUDA-enabled GPU.
loader = DataLoader(dataset, batch_size=self.hparams.batch_size, num_workers=self.hparams.num_workers,
shuffle=False, pin_memory=False)
return loader
def setup(self, mode):
self.train_loader = self._get_loader("train")
def train_dataloader(self):
return self.train_loader
def val_dataloader(self):
self.val_dataloader_obj = self._get_loader('dev')
return self.val_dataloader_obj
def test_dataloader(self):
return self._get_loader('test')
@property
def total_steps(self) -> int:
"""The number of total training steps that will be run. Used for lr scheduler purposes."""
num_devices = max(1, self.hparams.total_gpus) # TODO: consider num_tpu_cores
effective_batch_size = self.hparams.batch_size * self.hparams.grad_accum * num_devices
# dataset_size = len(self.train_loader.dataset)
"""The size of the training data need to be coded with more accurate number"""
dataset_size = training_size
return (dataset_size / effective_batch_size) * self.hparams.num_epochs
def get_lr_scheduler(self):
get_schedule_func = arg_to_scheduler[self.hparams.lr_scheduler]
scheduler = get_schedule_func(
self.opt, num_warmup_steps=self.hparams.warmup_steps, num_training_steps=self.total_steps
)
scheduler = {"scheduler": scheduler, "interval": "step", "frequency": 1}
return scheduler
def configure_optimizers(self):
"""Prepare optimizer and schedule (linear warmup and decay)"""
model = self.model
no_decay = ["bias", "LayerNorm.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": self.hparams.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,
},
]
if self.hparams.adafactor:
optimizer = Adafactor(
optimizer_grouped_parameters, lr=self.hparams.lr, scale_parameter=False, relative_step=False
)
else:
optimizer = AdamW(
optimizer_grouped_parameters, lr=self.hparams.lr, eps=self.hparams.adam_epsilon
)
self.opt = optimizer
scheduler = self.get_lr_scheduler()
return [optimizer], [scheduler]
def training_step(self, batch, batch_idx):
source_embedding = self.model(**batch[0])[1]
pos_embedding = self.model(**batch[1])[1]
neg_embedding = self.model(**batch[2])[1]
loss = self.triple_loss(source_embedding, pos_embedding, neg_embedding)
lr_scheduler = self.trainer.lr_schedulers[0]["scheduler"]
self.log('train_loss', loss, on_step=True, on_epoch=False, prog_bar=True, logger=True)
self.log('rate', lr_scheduler.get_last_lr()[-1], on_step=True, on_epoch=False, prog_bar=True, logger=True)
return {"loss": loss}
def validation_step(self, batch, batch_idx):
source_embedding = self.model(**batch[0])[1]
pos_embedding = self.model(**batch[1])[1]
neg_embedding = self.model(**batch[2])[1]
loss = self.triple_loss(source_embedding, pos_embedding, neg_embedding)
self.log('val_loss', loss, on_step=True, on_epoch=False, prog_bar=True)
return {'val_loss': loss}
def _eval_end(self, outputs) -> tuple:
avg_loss = torch.stack([x['val_loss'] for x in outputs]).mean()
if self.trainer.use_ddp:
torch.distributed.all_reduce(avg_loss, op=torch.distributed.ReduceOp.SUM)
avg_loss /= self.trainer.world_size
results = {"avg_val_loss": avg_loss}
for k, v in results.items():
if isinstance(v, torch.Tensor):
results[k] = v.detach().cpu().item()
return results
def validation_epoch_end(self, outputs: list) -> dict:
ret = self._eval_end(outputs)
self.log('avg_val_loss', ret["avg_val_loss"], on_epoch=True, prog_bar=True)
def test_epoch_end(self, outputs: list):
# convert the dictionary of {id1:embedding1, id2:embedding2, ...} to a
# list of dictionaries [{'id':'id1', 'embedding': 'embedding1'},{'id':'id2', 'embedding': 'embedding2'}, ...]
embedding_output_list = [{'id': key, 'embedding': value.detach().cpu().numpy().tolist()}
for key, value in self.embedding_output.items()]
with open(self.hparams.save_dir+'/embedding_result.jsonl', 'w') as fp:
fp.write('\n'.join(json.dumps(i) for i in embedding_output_list))
def test_step(self, batch, batch_nb):
source_embedding = self.model(**batch[0])[1]
source_paper_id = batch[1]
batch_embedding_output = dict(zip(source_paper_id, source_embedding))
# .update() will automatically remove duplicates.
self.embedding_output.update(batch_embedding_output)
# return self.validation_step(batch, batch_nb)
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument('--checkpoint_path', default=None, help='path to the model (if not setting checkpoint)')
parser.add_argument('--train_file')
parser.add_argument('--dev_file')
parser.add_argument('--test_file')
parser.add_argument('--input_dir', default=None, help='optionally provide a directory of the data and train/test/dev files will be automatically detected')
parser.add_argument('--batch_size', default=1, type=int)
parser.add_argument('--grad_accum', default=1, type=int)
parser.add_argument('--gpus', default='1')
parser.add_argument('--seed', default=1918, type=int)
parser.add_argument('--fp16', default=False, action='store_true')
parser.add_argument('--test_only', default=False, action='store_true')
parser.add_argument('--test_checkpoint', default=None)
parser.add_argument('--limit_test_batches', default=1.0, type=float)
parser.add_argument('--limit_val_batches', default=1.0, type=float)
parser.add_argument('--val_check_interval', default=1.0, type=float)
parser.add_argument('--num_epochs', default=1, type=int)
parser.add_argument("--lr", type=float, default=2e-5)
parser.add_argument("--weight_decay", default=0.0, type=float, help="Weight decay if we apply some.")
parser.add_argument("--adam_epsilon", default=1e-8, type=float, help="Epsilon for Adam optimizer.")
parser.add_argument("--warmup_steps", default=0, type=int, help="Linear warmup over warmup_steps.")
parser.add_argument("--num_workers", default=4, type=int, help="kwarg passed to DataLoader")
parser.add_argument("--adafactor", action="store_true")
parser.add_argument('--save_dir', required=True)
parser.add_argument('--num_samples', default=None, type=int)
parser.add_argument("--lr_scheduler",
default="linear",
choices=arg_to_scheduler_choices,
metavar=arg_to_scheduler_metavar,
type=str,
help="Learning rate scheduler")
args = parser.parse_args()
if args.input_dir is not None:
files = glob.glob(args.input_dir + '/*')
for f in files:
fname = f.split('/')[-1]
if 'train' in fname:
args.train_file = f
elif 'dev' in fname or 'val' in fname:
args.dev_file = f
elif 'test' in fname:
args.test_file = f
return args
def get_train_params(args):
train_params = {}
train_params["precision"] = 16 if args.fp16 else 32
if (isinstance(args.gpus, int) and args.gpus > 1) or (isinstance(args.gpus, list ) and len(args.gpus) > 1):
train_params["distributed_backend"] = "ddp"
else:
train_params["distributed_backend"] = None
train_params["accumulate_grad_batches"] = args.grad_accum
train_params['track_grad_norm'] = -1
train_params['limit_val_batches'] = args.limit_val_batches
train_params['val_check_interval'] = args.val_check_interval
train_params['gpus'] = args.gpus
train_params['max_epochs'] = args.num_epochs
train_params['log_every_n_steps'] = log_every_n_steps
return train_params
def main():
args = parse_args()
random.seed(args.seed)
np.random.seed(args.seed)
torch.manual_seed(args.seed)
if args.num_workers ==0:
print("num_workers cannot be less than 1")
return
if torch.cuda.is_available():
torch.cuda.manual_seed_all(args.seed)
if ',' in args.gpus:
args.gpus = list(map(int, args.gpus.split(',')))
args.total_gpus = len(args.gpus)
else:
args.gpus = int(args.gpus)
args.total_gpus = args.gpus
if args.test_only:
print('loading model...')
model = Specter.load_from_checkpoint(args.test_checkpoint)
trainer = pl.Trainer(gpus=args.gpus, limit_val_batches=args.limit_val_batches)
trainer.test(model)
else:
model = Specter(args)
# default logger used by trainer
logger = TensorBoardLogger(
save_dir=args.save_dir,
version=0,
name='pl-logs'
)
# second part of the path shouldn't be f-string
filepath = f'{args.save_dir}/version_{logger.version}/checkpoints/' + 'ep-{epoch}_avg_val_loss-{avg_val_loss:.3f}'
checkpoint_callback = ModelCheckpoint(
filepath=filepath,
save_top_k=1,
verbose=True,
monitor='avg_val_loss', # monitors metrics logged by self.log.
mode='min',
prefix=''
)
extra_train_params = get_train_params(args)
trainer = pl.Trainer(logger=logger,
checkpoint_callback=checkpoint_callback,
**extra_train_params)
trainer.fit(model)
if __name__ == '__main__':
main()
|