-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
481 lines (400 loc) · 17.8 KB
/
train.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
# coding: utf-8
import time
import math
import sys
import itertools
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
from data_utils import get_lm_corpus
from models.mem_transformer import MemTransformerLM
from utils.exp_utils import create_exp_dir, scale_grad, checkpoint_paths, optimize_model
# from utils.data_parallel import BalancedDataParallel
from utils.scheduler import InvSqrtAnnealingLR
import apex.amp as amp
import apex
from option import create_parser
import os
# os.environ['CUDA_LAUNCH_BLOCKING'] = "1"
parser = create_parser()
args = parser.parse_args()
args.tied = not args.not_tied
if args.d_embed < 0:
args.d_embed = args.d_model
assert args.ext_len >= 0, 'extended context length must be non-negative'
assert args.batch_size % args.batch_chunk == 0
# args.work_dir = '{}-{}'.format(args.work_dir, args.dataset)
# args.work_dir = os.path.join(args.work_dir, time.strftime('%Y%m%d-%H%M%S'))
logging = create_exp_dir(args.work_dir,
scripts_to_save=[], debug=args.debug)
# Set the random seed manually for reproducibility.
np.random.seed(args.seed)
torch.manual_seed(args.seed)
if torch.cuda.is_available():
if not args.cuda:
print('WARNING: You have a CUDA device, so you should probably run with --cuda')
else:
torch.cuda.manual_seed_all(args.seed)
# Validate `--fp16` option
if args.fp16:
if not args.cuda:
print('WARNING: --fp16 requires --cuda, ignoring --fp16 option')
args.fp16 = False
else:
try:
from apex.fp16_utils import FP16_Optimizer
except:
print('WARNING: apex not installed, ignoring --fp16 option')
args.fp16 = False
device = torch.device('cuda' if args.cuda else 'cpu')
###############################################################################
# Load data
###############################################################################
# corpus = get_lm_corpus(args.data, args.dataset)
corpus = get_lm_corpus(args.data)
ntokens = len(corpus.vocab)
vocab = corpus.vocab
args.n_token = ntokens
eval_batch_size = 1 if (not args.no_order) else args.batch_size
bos_id = corpus.vocab.get_idx('<bos>')
eos_id = corpus.vocab.get_idx('<eos>')
args.bos_id = bos_id
args.eos_id = eos_id
tr_iter = corpus.get_iterator('train', args.batch_size, args.tgt_len, order=(not args.no_order),
device=device, ext_len=args.ext_len,
bos_id=bos_id, eos_id=eos_id,
switchout=args.switchout)
va_iter = corpus.get_iterator('valid', eval_batch_size, args.eval_tgt_len, order=(not args.no_order),
device=device, ext_len=args.ext_len, bos_id=bos_id, eos_id=eos_id)
te_iter = corpus.get_iterator('test', eval_batch_size, args.eval_tgt_len, order=(not args.no_order),
device=device, ext_len=args.ext_len, bos_id=bos_id, eos_id=eos_id)
# adaptive softmax / embedding
cutoffs, tie_projs = [], [False]
###############################################################################
# Build the model
###############################################################################
def init_weight(weight):
if args.init == 'uniform':
nn.init.uniform_(weight, -args.init_range, args.init_range)
elif args.init == 'normal':
nn.init.normal_(weight, 0.0, args.init_std)
def init_embed(weight):
nn.init.normal_(model.decoder.word_lut.weight, mean=0, std=opt.model_size ** -0.5)
def init_bias(bias):
nn.init.constant_(bias, 0.0)
def weights_init(m):
classname = m.__class__.__name__
if classname.find('Linear') != -1:
if hasattr(m, 'weight') and m.weight is not None:
init_weight(m.weight)
if hasattr(m, 'bias') and m.bias is not None:
init_bias(m.bias)
elif classname.find('AdaptiveEmbedding') != -1:
if hasattr(m, 'emb_projs'):
for i in range(len(m.emb_projs)):
if m.emb_projs[i] is not None:
nn.init.normal_(m.emb_projs[i], 0.0, args.proj_init_std)
elif classname.find('Embedding') != -1:
if hasattr(m, 'weight'):
init_weight(m.weight)
elif classname.find('LayerNorm') != -1 or classname.find('FusedLayerNorm') != -1:
if hasattr(m, 'weight'):
nn.init.normal_(m.weight, 1.0, args.init_std)
if hasattr(m, 'bias') and m.bias is not None:
init_bias(m.bias)
elif classname.find('TransformerLM') != -1:
if hasattr(m, 'r_emb'):
init_weight(m.r_emb)
if hasattr(m, 'r_w_bias'):
init_weight(m.r_w_bias)
if hasattr(m, 'r_r_bias'):
init_weight(m.r_r_bias)
if hasattr(m, 'r_bias'):
init_bias(m.r_bias)
def update_dropout(m):
classname = m.__class__.__name__
if classname.find('Dropout') != -1:
if hasattr(m, 'p'):
m.p = args.dropout
def update_dropatt(m):
if hasattr(m, 'dropatt'):
m.dropatt.p = args.dropatt
model = MemTransformerLM(vocab, args.n_layer, args.n_head, args.d_model,
args.d_head, args.d_inner, args.dropout, args.dropatt,
tie_weight=args.tied, d_embed=args.d_embed, div_val=args.div_val,
tie_projs=tie_projs, pre_lnorm=args.pre_lnorm, tgt_len=args.tgt_len,
ext_len=args.ext_len, mem_len=args.mem_len, cutoffs=cutoffs,
same_length=args.same_length, attn_type=args.attn_type,
clamp_len=args.clamp_len, sample_softmax=args.sample_softmax,
word_dropout=args.word_dropout, label_smoothing=args.label_smoothing,
death_rate=args.layer_drop)
optimize_model(model)
model.apply(weights_init)
model.word_emb.apply(weights_init) # ensure embedding init is not overridden by out_layer in case of weight sharing
args.n_all_param = sum([p.nelement() for p in model.parameters()])
args.n_nonemb_param = sum([p.nelement() for p in model.layers.parameters()])
# if args.fp16:
# model = model.half()
if args.multi_gpu:
model = model.to(device)
if args.gpu0_bsz >= 0:
para_model = BalancedDataParallel(args.gpu0_bsz // args.batch_chunk,
model, dim=1).to(device)
else:
para_model = nn.DataParallel(model, dim=1).to(device)
else:
para_model = model.to(device)
#### optimizer
if args.optim.lower() == 'sgd':
optimizer = optim.SGD(model.parameters(), lr=args.lr,
momentum=args.mom)
elif args.optim.lower() == 'adam':
optimizer = optim.Adam(model.parameters(), lr=args.lr)
elif args.optim.lower() == 'fused_adam':
optimizer = apex.optimizers.FusedAdam(model.parameters(), lr=args.lr)
elif args.optim.lower() == 'adagrad':
optimizer = optim.Adagrad(model.parameters(), lr=args.lr)
#### Apex
model, optimizer = amp.initialize(model, optimizer, opt_level=args.opt_level,
keep_batchnorm_fp32=False, loss_scale="dynamic")
#### scheduler
if args.scheduler == 'cosine':
# here we do not set eta_min to lr_min to be backward compatible
# because in previous versions eta_min is default to 0
# rather than the default value of lr_min 1e-6
scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer,
args.max_step, eta_min=args.eta_min) # should use eta_min arg
elif args.scheduler == 'inv_sqrt':
# originally used for Transformer (in Attention is all you need)
# def lr_lambda(step):
# # return a multiplier instead of a learning rate
# init_lr = (512 ** (-0.5)) * args.lr
#
# if step == 0 and args.warmup_step == 0:
# return 0.0
# else:
# return init_lr * (step ** (-0.5)) if step > args.warmup_step \
# else init_lr * step * (args.warmup_step ** (-1.5))
scheduler = InvSqrtAnnealingLR(optimizer, args.warmup_step)
# scheduler = optim.lr_scheduler.LambdaLR(optimizer, lr_lambda=lr_lambda)
elif args.scheduler == 'dev_perf':
scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer,
factor=args.decay_rate, patience=args.patience, min_lr=args.lr_min)
elif args.scheduler == 'constant':
pass
# if args.cuda and args.fp16:
# # If args.dynamic_loss_scale is False, static_loss_scale will be used.
# # If args.dynamic_loss_scale is True, it will take precedence over static_loss_scale.
# optimizer = FP16_Optimizer(optimizer,
# static_loss_scale = args.static_loss_scale,
# dynamic_loss_scale = args.dynamic_loss_scale,
# dynamic_loss_args = {'init_scale': 2 ** 16})
if len(args.pretrain) > 0:
print("loading pretrained model from %s" % args.pretrain)
pretrained_checkpoint = torch.load(args.pretrain)
model.load_state_dict(pretrained_checkpoint['model'])
del pretrained_checkpoint
if args.restart:
# if os.path.exists(os.path.join(args.restart_dir, 'optimizer.pt')):
# with open(os.path.join(args.restart_dir, 'optimizer.pt'), 'rb') as f:
# opt_state_dict = torch.load(f)
# optimizer.load_state_dict(opt_state_dict)
# else:
# if os.path.exists(os.path.join(args.restart_dir, 'checkpoint.pt')):
if os.path.exists(args.restart_checkpoint):
with open(args.restart_checkpoint, 'rb') as f:
checkpoint = torch.load(f)
optimizer.load_state_dict(checkpoint['optimizer'])
model.load_state_dict(checkpoint['model'])
amp.load_state_dict(checkpoint['amp'])
else:
print('Optimizer was not saved. Start from scratch.')
logging('=' * 100)
for k, v in args.__dict__.items():
logging(' - {} : {}'.format(k, v))
logging('=' * 100)
logging('#params = {}'.format(args.n_all_param))
logging('#non emb params = {}'.format(args.n_nonemb_param))
###############################################################################
# Training code
###############################################################################
def evaluate(eval_iter):
# Turn on evaluation mode which disables dropout.
model.eval()
eval_iter.reset_order()
# If the model does not use memory at all, make the ext_len longer.
# Otherwise, make the mem_len longer and keep the ext_len the same.
if args.mem_len == 0:
model.reset_length(args.eval_tgt_len,
args.ext_len + args.tgt_len - args.eval_tgt_len, args.mem_len)
else:
model.reset_length(args.eval_tgt_len,
args.ext_len, args.mem_len + args.tgt_len - args.eval_tgt_len)
# Evaluation
total_len, total_loss = 0, 0.
with torch.no_grad():
mems = tuple()
for i, (data, target, seq_len, weight) in enumerate(eval_iter):
if 0 < args.max_eval_steps <= i:
break
ret = model(data, target, weight, *mems)
loss, nll, mems = ret[0], ret[1], ret[2:]
if args.no_order:
mems = tuple()
total_loss += nll
total_len += weight.sum().item()
# Switch back to the training mode
model.reset_length(args.tgt_len, args.ext_len, args.mem_len)
model.train()
return total_loss / total_len
def train():
# Turn on training mode which enables dropout.
global train_step, best_val_loss, eval_start_time, log_start_time
model.train()
mems = tuple()
tr_iter.reset_order()
train_iter = tr_iter.get_varlen_iter() if args.varlen else tr_iter
n_accumulated_words = 0
denom = 8000
total_words = 0
train_loss = 0
for batch, (data, target, seq_len, weight) in enumerate(train_iter):
model.zero_grad()
ret = para_model(data, target, weight, *mems)
loss, nll, mems = ret[0], ret[1], ret[2:]
ntarget = weight.float().sum().item()
loss = loss.float().sum().type_as(loss)
with amp.scale_loss(loss, optimizer) as scaled_loss:
scaled_loss.backward()
train_loss += (loss.float().item())
n_accumulated_words += ntarget
total_words += ntarget
# reset the memory if we don't care about order
if args.no_order:
mems = tuple()
if n_accumulated_words >= args.batch_size_update:
# div gradients to the number of accumulated
scale_factor = n_accumulated_words
scale_grad(amp.master_params(optimizer), scale_factor)
torch.nn.utils.clip_grad_norm_(amp.master_params(optimizer), args.clip)
n_accumulated_words = 0
temp_accumulated = 0
optimizer.step()
model.zero_grad()
optimizer.zero_grad()
# step-wise learning rate annealing
train_step += 1
if args.scheduler in ['cosine', 'constant', 'dev_perf']:
# linear warmup stage
if train_step < args.warmup_step:
curr_lr = args.lr * train_step / args.warmup_step
optimizer.param_groups[0]['lr'] = curr_lr
else:
if args.scheduler == 'cosine':
scheduler.step(train_step)
elif args.scheduler == 'inv_sqrt':
# scheduler.step(train_step)
scheduler.step()
if train_step % args.log_interval == 0:
cur_loss = train_loss / total_words
elapsed = time.time() - log_start_time
log_str = '| epoch {:3d} step {:>8d} | {:>6d} batches | lr {:.3g} ' \
'| ms/update {:5.2f} | loss {:5.2f}'.format(
epoch, train_step, batch + 1, optimizer.param_groups[0]['lr'],
elapsed * 1000 / args.log_interval, cur_loss)
# if args.dataset in ['enwik8', 'text8']:
# log_str += ' | bpc {:9.5f}'.format(cur_loss / math.log(2))
# else:
log_str += ' | ppl {:9.3f}'.format(math.exp(cur_loss))
logging(log_str)
# train_loss = 0
log_start_time = time.time()
if train_step % args.eval_interval == 0:
val_loss = evaluate(va_iter)
logging('-' * 100)
log_str = '| Eval {:3d} at step {:>8d} | time: {:5.2f}s ' \
'| valid loss {:5.2f}'.format(
train_step // args.eval_interval, train_step,
(time.time() - eval_start_time), val_loss)
# if args.dataset in ['enwik8', 'text8']:
# log_str += ' | bpc {:9.5f}'.format(val_loss / math.log(2))
# else:
log_str += ' | valid ppl {:9.3f}'.format(math.exp(val_loss))
logging(log_str)
logging('-' * 100)
# Save the model (always)
checkpoint = dict()
checkpoint['model'] = model.state_dict()
checkpoint['optimizer'] = optimizer.state_dict()
checkpoint['amp'] = amp.state_dict()
checkpoint['args'] = args
checkpoint['vocab'] = vocab
val_ppl = math.exp(val_loss)
checkpoint_name = 'checkpoint_ppl_%.6f_xl.pt' % val_ppl
# checkpoint['vocab_size'] = args.n_token
checkpoint_dir = args.work_dir
with open(os.path.join(args.work_dir, checkpoint_name), 'wb') as f:
log_str = "Saving to file %s" % os.path.join(args.work_dir, checkpoint_name)
logging(log_str)
torch.save(checkpoint, f)
existed_save_files = checkpoint_paths(checkpoint_dir)
num_save_files = 5
for save_file in existed_save_files[num_save_files:]:
print(" * Deleting old save file %s ...." % save_file)
os.remove(save_file)
# now we have to delete the worst checkpoint:
if not best_val_loss or val_loss < best_val_loss:
if not args.debug:
continue
best_val_loss = val_loss
# dev-performance based learning rate annealing
if args.scheduler == 'dev_perf':
scheduler.step(val_loss)
eval_start_time = time.time()
if train_step == args.max_step:
break
# Loop over epochs.
train_step = 0
train_loss = 0
best_val_loss = None
log_start_time = time.time()
eval_start_time = time.time()
val_loss = evaluate(va_iter)
logging('-' * 100)
log_str = '| Eval at step {:>8d} | time: {:5.2f}s ' \
'| valid loss {:5.2f}'.format(
0,
(time.time() - eval_start_time), val_loss)
# if args.dataset in ['enwik8', 'text8']:
# log_str += ' | bpc {:9.5f}'.format(val_loss / math.log(2))
# else:
log_str += ' | valid ppl {:9.3f}'.format(math.exp(val_loss))
logging(log_str)
# At any point you can hit Ctrl + C to break out of training early.
try:
for epoch in itertools.count(start=1):
train()
if train_step == args.max_step:
logging('-' * 100)
logging('End of training')
break
except KeyboardInterrupt:
logging('-' * 100)
logging('Exiting from training early')
#
# # Load the best saved model.
# with open(os.path.join(args.work_dir, 'model.pt'), 'rb') as f:
# model = torch.load(f)
# para_model = model.to(device)
#
# # Run on test data.
# test_loss = evaluate(te_iter)
# logging('=' * 100)
# if args.dataset in ['enwik8', 'text8']:
# logging('| End of training | test loss {:5.2f} | test bpc {:9.5f}'.format(
# test_loss, test_loss / math.log(2)))
# else:
# logging('| End of training | test loss {:5.2f} | test ppl {:9.3f}'.format(
# test_loss, math.exp(test_loss)))
# logging('=' * 100)