-
Notifications
You must be signed in to change notification settings - Fork 0
/
trainer.py
583 lines (472 loc) · 24.8 KB
/
trainer.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
import os
import torch
import numpy as np
import editdistance
import torch.nn as nn
from typing import List
from typing import Tuple
from torch import Tensor
from logger import logger
from dataset import Batch
from typing import Optional
from torch.optim import SGD
from torch.optim import Adam
from torch.optim import AdamW
from settings import Settings
from dataset import RawDataset
from collections import namedtuple
from loss import soft_attention_loss
from models import SoftAttentionModel
from models.base import TransducerModel
from torch.utils.data import DataLoader
from models import NonAutoregressiveLSTM
from dataset import TransducerDatasetTrain
from models import LSTMEncoderDecoderModel
from torch.nn.utils import clip_grad_value_
from torch.optim.lr_scheduler import OneCycleLR
from torch.optim.lr_scheduler import ExponentialLR
from loss import autoregressive_transduction_loss
from inference import non_autoregressive_inference
from inference import soft_attention_greedy_sampling
from inference import autoregressive_greedy_sampling
from loss import non_autoregressive_transduction_loss
from dataset import AutoregressiveTransducerDatasetTrain
from dataset import NonAutoregressiveTransducerDatasetTrain
from vocabulary import SourceVocabulary, TransducerVocabulary
from loss import fast_autoregressive_transduction_loss
Sequence = List[str]
Sequences = List[Sequence]
TrainData = Tuple[Sequences, Sequences]
TrainedModel = namedtuple(
"TrainedModel",
["model", "source_vocabulary", "target_vocabulary", "feature_vocabulary", "metrics", "checkpoint", "settings"]
)
def _prepare_datasets(settings: Settings, train_data: RawDataset, development_data: Optional[RawDataset],
autoregressive: bool = True, use_features: bool = False):
# Build vocabularies
source_vocabulary = SourceVocabulary.build_vocabulary(
train_data.sources, min_frequency=settings.min_source_frequency
)
target_vocabulary = TransducerVocabulary.build_vocabulary(
train_data.targets, min_frequency=settings.min_target_frequency
)
if use_features:
feature_vocabulary = SourceVocabulary.build_vocabulary(train_data.features)
else:
feature_vocabulary = None
dataset_class = AutoregressiveTransducerDatasetTrain if autoregressive else NonAutoregressiveTransducerDatasetTrain
train_dataset = dataset_class(
dataset=train_data, source_vocabulary=source_vocabulary, target_vocabulary=target_vocabulary,
feature_vocabulary=feature_vocabulary, use_features=use_features
)
if development_data is not None:
development_dataset = dataset_class(
dataset=development_data, source_vocabulary=source_vocabulary, target_vocabulary=target_vocabulary,
feature_vocabulary=feature_vocabulary, use_features=use_features
)
else:
development_dataset = None
return source_vocabulary, target_vocabulary, feature_vocabulary, train_dataset, development_dataset
def _build_model(source_vocab_size: int, target_vocab_size: int, settings: Settings,
feature_vocab_size: Optional[int] = None) -> TransducerModel:
if settings.model == "autoregressive":
return LSTMEncoderDecoderModel(
source_vocab_size=source_vocab_size, target_vocab_size=target_vocab_size,
embedding_dim=settings.embedding_size, hidden_size=settings.hidden_size,
num_layers=settings.hidden_layers, dropout=settings.dropout, device=settings.device,
scorer=settings.scorer, temperature=settings.temperature, use_features=settings.use_features,
feature_vocab_size=feature_vocab_size, feature_encoder_hidden=settings.hidden_size,
feature_encoder_pooling=settings.features_pooling, feature_encoder_layers=settings.features_num_layers,
encoder_bridge=settings.encoder_bridge
)
elif settings.model == "non-autoregressive":
return NonAutoregressiveLSTM(
source_vocab_size=source_vocab_size, target_vocab_size=target_vocab_size,
embedding_dim=settings.embedding_size, hidden_size=settings.hidden_size,
num_layers=settings.hidden_layers, dropout=settings.dropout, device=settings.device,
tau=settings.tau, scorer=settings.scorer, temperature=settings.temperature,
use_features=settings.use_features, feature_vocab_size=feature_vocab_size,
feature_encoder_layers=settings.features_num_layers, feature_encoder_hidden=settings.hidden_size,
feature_encoder_pooling=settings.features_pooling, decoder_type=settings.non_autoregressive_decoder,
max_targets_per_symbol=settings.max_targets_per_symbol
)
elif settings.model == "soft-attention":
return SoftAttentionModel(
source_vocab_size=source_vocab_size, target_vocab_size=target_vocab_size,
embedding_dim=settings.embedding_size, hidden_size=settings.hidden_size,
num_layers=settings.hidden_layers, dropout=settings.dropout, device=settings.device,
scorer=settings.scorer, temperature=settings.temperature, use_features=settings.use_features,
feature_vocab_size=feature_vocab_size, feature_encoder_hidden=settings.hidden_size,
feature_encoder_pooling=settings.features_pooling, feature_encoder_layers=settings.features_num_layers,
encoder_bridge=settings.encoder_bridge
)
else:
raise ValueError(f"Unknown model type: {settings.model}")
def _build_optimizer(model: nn.Module, optimizer: str, lr: float, weight_decay: float):
if optimizer == "adamw":
return AdamW(model.parameters(), lr=lr, weight_decay=weight_decay)
elif optimizer == "adam":
return Adam(model.parameters(), lr=lr, weight_decay=weight_decay)
elif optimizer == "sgd":
return SGD(model.parameters(), lr=lr, weight_decay=weight_decay)
else:
raise ValueError(f"Unknown optimizer: {optimizer}")
def _build_scheduler(optimizer, scheduler: str, gamma: float = 1.0, max_learning_rate: float = 0.1,
total_steps: int = None):
if scheduler == "exponential":
scheduler_model = ExponentialLR(optimizer, gamma=gamma)
elif scheduler == "one-cycle":
scheduler_model = OneCycleLR(optimizer, max_lr=max_learning_rate, total_steps=total_steps, div_factor=100)
else:
raise ValueError(f"Unknown scheduler: {scheduler}")
def scheduler_step(epoch_end: bool):
if scheduler == "exponential" and epoch_end:
scheduler_model.step()
elif scheduler == "one-cycle" and not epoch_end:
scheduler_model.step()
else:
pass
return scheduler_step
def _count_model_parameters(model: nn.Module) -> int:
return sum(parameter.numel() for parameter in model.parameters())
def moving_avg_loss(old_loss: float, new_loss: float, gamma: float = 0.95) -> float:
if old_loss is None:
return new_loss
else:
return gamma * old_loss + (1 - gamma) * new_loss
def save_model(model: TrainedModel, name: str, path: str) -> str:
os.makedirs(path, exist_ok=True)
model_save_info = dict()
model_save_info["model_class"] = type(model.model)
model_save_info["parameters"] = model.model.get_params()
model_save_info["state_dict"] = model.model.state_dict()
model_save_info["source_vocabulary"] = model.source_vocabulary
model_save_info["target_vocabulary"] = model.target_vocabulary
model_save_info["feature_vocabulary"] = model.feature_vocabulary
model_save_info["metrics"] = model.metrics
model_save_info["checkpoint"] = model.checkpoint
model_save_info["settings"] = model.settings
save_model_path = os.path.join(path, name + ".pt")
torch.save(model_save_info, save_model_path)
return save_model_path
def load_model(path: str) -> TrainedModel:
model_save_info = torch.load(path)
model = model_save_info["model_class"](**model_save_info["parameters"])
model.load_state_dict(model_save_info["state_dict"])
source_vocabulary = model_save_info["source_vocabulary"]
target_vocabulary = model_save_info["target_vocabulary"]
feature_vocabulary = model_save_info["feature_vocabulary"]
metrics = model_save_info["metrics"]
checkpoint = model_save_info["checkpoint"]
settings = model_save_info["settings"]
return TrainedModel(
model=model, source_vocabulary=source_vocabulary, target_vocabulary=target_vocabulary,
feature_vocabulary=feature_vocabulary, metrics=metrics, checkpoint=checkpoint, settings=settings
)
def evaluate_on_development_set(model_name: str, model: TrainedModel, development_data: TransducerDatasetTrain,
batch_size: int, fast_autoregressive_loss: bool, device: torch.device,
max_decoding_length: int):
assert development_data is not None
source_vocabulary = model.source_vocabulary
target_vocabulary = model.target_vocabulary
feature_vocabulary = model.feature_vocabulary
model: TransducerModel = model.model
model = model.eval()
development_dataloader = DataLoader(
development_data, batch_size=batch_size, shuffle=False, collate_fn=development_data.collate_fn
)
losses = []
predictions = []
targets = []
with torch.no_grad():
for batch in development_dataloader:
batch_sources = [
[symbol for symbol in source if symbol not in source_vocabulary.get_specials()]
for source in batch.raw_sources
]
if model.use_features:
batch_features = [
[symbol for symbol in feats if symbol not in feature_vocabulary.get_specials()]
for feats in batch.raw_features
]
else:
batch_features = None
if model_name == "autoregressive":
batch_losses = _get_autoregressive_loss(
model=model, batch=batch, device=device, allow_copy=True, enforce_copy=False, reduction='none',
fast=fast_autoregressive_loss
)
batch_predictions = autoregressive_greedy_sampling(
model=model, source_vocabulary=source_vocabulary, target_vocabulary=target_vocabulary,
sequences=batch_sources, max_decoding_length=max_decoding_length, features=batch_features,
feature_vocabulary=feature_vocabulary
)
elif model_name == "non-autoregressive":
batch_losses = _get_non_autoregressive_loss(
model=model, batch=batch, device=device, allow_copy=True, enforce_copy=False, noop_discount=1.0,
reduction='none'
)
batch_predictions = non_autoregressive_inference(
model=model, source_vocabulary=source_vocabulary, target_vocabulary=target_vocabulary,
sequences=batch_sources, features=batch_features, feature_vocabulary=feature_vocabulary,
max_decoding_length=max_decoding_length
)
elif model_name == "soft-attention":
batch_losses = _get_soft_attention_loss(model=model, batch=batch, reduction="none")
batch_predictions = soft_attention_greedy_sampling(
model=model, source_vocabulary=source_vocabulary, target_vocabulary=target_vocabulary,
sequences=batch_sources, max_decoding_length=max_decoding_length, features=batch_features,
feature_vocabulary=feature_vocabulary
)
else:
raise ValueError(f"Unknown model name: {model_name}")
batch_losses = batch_losses.detach().cpu().tolist()
losses.extend(batch_losses)
batch_predictions = [prediction for prediction, _ in batch_predictions]
batch_predictions = [
[symbol for symbol in prediction if symbol not in target_vocabulary.get_special_symbols()]
for prediction in batch_predictions
]
predictions.extend(batch_predictions)
batch_targets = [
[symbol for symbol in target if symbol not in target_vocabulary.get_special_symbols()]
for target in batch.raw_targets
]
targets.extend(batch_targets)
# Calculate metrics
loss = np.mean(losses)
wer = 100 * (1 - np.mean([prediction == target for prediction, target in zip(predictions, targets)]))
edit_distances = [editdistance.distance(prediction, target) for prediction, target in zip(predictions, targets)]
edit_distance = np.mean(edit_distances)
normalised_edit_distance = np.mean([distance / len(target) for distance, target in zip(edit_distances, targets)])
return {
'loss': loss,
'wer': wer,
'edit_distance': edit_distance,
"normalised_edit_distance": normalised_edit_distance
}
def _autoregressive_get_scores(model: TransducerModel, batch: Batch) -> Tensor:
source_encodings = model.encode(batch.sources, batch.source_lengths)
target_encodings, _ = model.decode(
batch.targets, batch.target_lengths, source_encodings, batch.source_lengths
)
return model.get_transduction_scores(
source_encodings, target_encodings, batch.features, batch.feature_lengths
)
def _get_autoregressive_loss(model: TransducerModel, batch: Batch, device: torch.device, allow_copy: bool,
enforce_copy: bool, reduction: str, fast: bool) -> Tensor:
scores = _autoregressive_get_scores(model=model, batch=batch)
criterion = fast_autoregressive_transduction_loss if fast else autoregressive_transduction_loss
"""
fast_loss = fast_autoregressive_transduction_loss(
scores=scores, source_lengths=batch.source_lengths, target_lengths=batch.target_lengths,
insertion_labels=batch.insertion_labels, substitution_labels=batch.substitution_labels,
copy_index=batch.copy_index, copy_shift_index=batch.copy_shift_index, deletion_index=batch.deletion_index,
copy_matrix=batch.copy_matrix, device=device, allow_copy=allow_copy, enforce_copy=enforce_copy,
reduction=reduction
)
slow_loss = autoregressive_transduction_loss(
scores=scores, source_lengths=batch.source_lengths, target_lengths=batch.target_lengths,
insertion_labels=batch.insertion_labels, substitution_labels=batch.substitution_labels,
copy_index=batch.copy_index, copy_shift_index=batch.copy_shift_index, deletion_index=batch.deletion_index,
copy_matrix=batch.copy_matrix, device=device, allow_copy=allow_copy, enforce_copy=enforce_copy,
reduction=reduction
)
print(fast_loss)
print(slow_loss)
raise
"""
loss = criterion(
scores=scores, source_lengths=batch.source_lengths, target_lengths=batch.target_lengths,
insertion_labels=batch.insertion_labels, substitution_labels=batch.substitution_labels,
copy_index=batch.copy_index, copy_shift_index=batch.copy_shift_index, deletion_index=batch.deletion_index,
copy_matrix=batch.copy_matrix, device=device, allow_copy=allow_copy, enforce_copy=enforce_copy,
reduction=reduction
)
return loss
def _get_soft_attention_loss(model: TransducerModel, batch: Batch, reduction: str) -> Tensor:
targets = batch.targets[:, 1:]
target_lengths = batch.target_lengths - 1
source_encodings = model.encode(batch.sources, batch.source_lengths)
target_encodings, _ = model.decode(targets, target_lengths, source_encodings, batch.source_lengths)
if model.use_features:
scores = model.get_transduction_scores(target_encodings, batch.features, batch.feature_lengths)
else:
scores = model.get_transduction_scores(target_encodings)
scores = scores[:, :-1]
labels = batch.targets[:, 2:] # Remove SOS symbols from labels
loss = soft_attention_loss(scores=scores, target_labels=labels, reduction=reduction)
return loss
def _non_autoregressive_get_scores(model: TransducerModel, batch: Batch) -> Tuple[Tensor, int]:
tau = model.tau
if tau is None:
tau = batch.target_lengths.max().detach().cpu().item()
if tau > model.max_targets_per_symbol:
tau = model.max_targets_per_symbol
if model.use_features:
scores = model(
sources=batch.sources, lengths=batch.source_lengths, features=batch.features,
feature_lengths=batch.feature_lengths, tau=tau
)
else:
scores = model(
sources=batch.sources, lengths=batch.source_lengths, features=None, feature_lengths=None, tau=tau
)
return scores, tau
def _get_non_autoregressive_loss(model: TransducerModel, batch: Batch, device: torch.device, allow_copy: bool,
enforce_copy: bool, noop_discount: float, reduction: str) -> Tensor:
scores, tau = _non_autoregressive_get_scores(model=model, batch=batch)
loss = non_autoregressive_transduction_loss(
scores=scores, source_lengths=batch.source_lengths, target_lengths=batch.target_lengths,
insertion_labels=batch.insertion_labels, substitution_labels=batch.substitution_labels,
copy_index=batch.copy_index, copy_shift_index=batch.copy_shift_index,
deletion_index=batch.deletion_index, noop_index=batch.noop_index,
copy_matrix=batch.copy_matrix, device=device, allow_copy=allow_copy, enforce_copy=enforce_copy, tau=tau,
noop_discount=noop_discount, reduction=reduction, return_backpointers=False
)
return loss
def train(train_data: RawDataset, development_data: Optional[RawDataset], settings: Settings) -> TrainedModel:
if settings.verbose:
logger.info("Prepare for Training")
logger.info("Build vocabulary and datasets")
is_non_autoregressive = settings.model == "non-autoregressive"
is_autoregressive = not is_non_autoregressive
source_vocabulary, target_vocabulary, feature_vocabulary, train_dataset, dev_dataset = _prepare_datasets(
settings=settings, train_data=train_data, development_data=development_data,
autoregressive=is_autoregressive, use_features=settings.use_features
)
max_development_decoding_length = max([len(datapoint.target) for datapoint in train_dataset]) + 10
if settings.verbose:
logger.info(f"Train data contains {len(train_dataset)} datapoints")
if dev_dataset is not None:
logger.info(f"Dev data contains {len(dev_dataset)} datapoints")
logger.info(f"Source vocabulary contains {len(source_vocabulary)} items")
logger.info(f"Target vocabulary contains {len(target_vocabulary)} actions")
train_dataloader = DataLoader(
train_dataset, batch_size=settings.batch, shuffle=True, collate_fn=train_dataset.collate_fn
)
if settings.verbose:
logger.info("Build model")
feature_vocab_size = None if feature_vocabulary is None else len(feature_vocabulary)
if settings.model == "soft-attention":
target_vocab_size = len(target_vocabulary.symbols)
else:
target_vocab_size = len(target_vocabulary)
model = _build_model(
source_vocab_size=len(source_vocabulary), target_vocab_size=target_vocab_size, settings=settings,
feature_vocab_size=feature_vocab_size
)
if settings.verbose:
num_model_parameters = _count_model_parameters(model)
logger.info(f"Model has {num_model_parameters} parameters")
logger.info(f"Device: {settings.device}")
model = model.to(device=settings.device)
model = model.train()
if settings.verbose:
logger.info("Build optimizer")
optimizer = _build_optimizer(
model=model, optimizer=settings.optimizer, lr=settings.lr, weight_decay=settings.weight_decay
)
if settings.verbose:
logger.info("Build scheduler")
scheduler_step = _build_scheduler(
optimizer, scheduler=settings.scheduler, gamma=settings.gamma, max_learning_rate=settings.lr,
total_steps=settings.epochs * len(train_dataloader)
)
if settings.verbose:
logger.info("Start Training")
running_loss = None
step_counter = 0
best_model_metric = np.inf
best_checkpoint_path = None
total_num_steps = settings.epochs * len(train_dataloader)
for epoch in range(1, settings.epochs + 1):
# Train epoch
model = model.train()
epoch_losses = []
for batch in train_dataloader:
optimizer.zero_grad()
if settings.model == "autoregressive":
loss = _get_autoregressive_loss(
model=model, batch=batch, allow_copy=settings.allow_copy, enforce_copy=settings.enforce_copy,
device=settings.device, reduction='mean', fast=settings.fast_autoregressive
)
elif settings.model == "non-autoregressive":
loss = _get_non_autoregressive_loss(
model=model, batch=batch, allow_copy=settings.allow_copy, enforce_copy=settings.enforce_copy,
device=settings.device, noop_discount=settings.noop_discount, reduction='mean'
)
elif settings.model == "soft-attention":
loss = _get_soft_attention_loss(model=model, batch=batch, reduction="mean")
else:
raise ValueError(f"Unknown model: {settings.model}")
# Update parameters
loss.backward()
if settings.grad_clip is not None:
clip_grad_value_(model.parameters(), settings.grad_clip)
optimizer.step()
scheduler_step(epoch_end=False)
# Display loss
step_counter += 1
loss_item = loss.detach().cpu().item()
running_loss = moving_avg_loss(running_loss, loss_item)
epoch_losses.append(loss_item)
if settings.verbose:
if step_counter % settings.report_progress_every == 0 or step_counter == 1:
progress = 100 * step_counter / total_num_steps
current_learning_rate = optimizer.param_groups[0]['lr']
logger.info(
f"[{progress:.2f}%]" +
f" Loss: {running_loss:.3f}" +
f" || LR: {current_learning_rate:.6f}" +
f" || Step {step_counter} / {total_num_steps}"
)
scheduler_step(epoch_end=True)
# Evaluate on dev set
epoch_model = TrainedModel(
model=model, source_vocabulary=source_vocabulary, target_vocabulary=target_vocabulary,
feature_vocabulary=feature_vocabulary, metrics=None, checkpoint=None, settings=settings
)
if dev_dataset is not None and (epoch % settings.evaluate_every == 0 or epoch == settings.epochs):
development_metrics = evaluate_on_development_set(
model_name=settings.model, model=epoch_model, development_data=dev_dataset,
batch_size=settings.batch, device=settings.device, max_decoding_length=max_development_decoding_length,
fast_autoregressive_loss=settings.fast_autoregressive
)
if settings.verbose:
logger.info(
f"[Development metrics] " +
f"Loss: {development_metrics['loss']:.4f}" +
f" || WER: {development_metrics['wer']:.2f}" +
f" || Edit-Distance: {development_metrics['edit_distance']:.2f}"
)
elif dev_dataset is None:
development_metrics = None
else:
continue
if development_metrics is not None:
epoch_model_metric = development_metrics[settings.main_metric]
else:
epoch_model_metric = np.mean(epoch_losses)
model_improved = epoch_model_metric < best_model_metric
best_model_metric = epoch_model_metric if model_improved else best_model_metric
save_metrics = development_metrics if development_metrics is not None else {'loss': np.mean(epoch_losses)}
epoch_model = TrainedModel(
model=model, source_vocabulary=source_vocabulary, target_vocabulary=target_vocabulary,
feature_vocabulary=feature_vocabulary, metrics=save_metrics, checkpoint=epoch, settings=settings
)
if settings.keep_only_best_checkpoint:
if model_improved or epoch == 1:
if settings.verbose:
logger.info(f"Saving Model after epoch {epoch}")
checkpoint_path = save_model(model=epoch_model, name=settings.name, path=settings.save_path)
else:
checkpoint_path = best_checkpoint_path
else:
if settings.verbose:
logger.info(f"Saving Model after epoch {epoch}")
checkpoint_path = save_model(model=epoch_model, name=settings.name + f"_{epoch}", path=settings.save_path)
if model_improved or epoch == 1:
best_checkpoint_path = checkpoint_path
model = load_model(best_checkpoint_path)
return model