run_classifier.py 18.5 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
15
"""BERT classification or regression finetuning runner in TF 2.x."""
16

17
import functools
18
19
import json
import math
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
20
import os
21

Hongkun Yu's avatar
Hongkun Yu committed
22
# Import libraries
23
24
25
from absl import app
from absl import flags
from absl import logging
Le Hou's avatar
Le Hou committed
26
import gin
27
import tensorflow as tf
28
from official.common import distribute_utils
29
from official.modeling import performance
30
from official.nlp import optimization
31
from official.nlp.bert import bert_models
32
from official.nlp.bert import common_flags
33
from official.nlp.bert import configs as bert_configs
34
35
from official.nlp.bert import input_pipeline
from official.nlp.bert import model_saving_utils
36
from official.utils.misc import keras_utils
37
38

flags.DEFINE_enum(
Hongkun Yu's avatar
Hongkun Yu committed
39
40
    'mode', 'train_and_eval', ['train_and_eval', 'export_only', 'predict'],
    'One of {"train_and_eval", "export_only", "predict"}. `train_and_eval`: '
41
42
    'trains the model and evaluates in the meantime. '
    '`export_only`: will take the latest checkpoint inside '
Hongkun Yu's avatar
Hongkun Yu committed
43
44
    'model_dir and export a `SavedModel`. `predict`: takes a checkpoint and '
    'restores the model to output predictions on the test set.')
45
46
47
48
49
50
51
52
flags.DEFINE_string('train_data_path', None,
                    'Path to training data for BERT classifier.')
flags.DEFINE_string('eval_data_path', None,
                    'Path to evaluation data for BERT classifier.')
flags.DEFINE_string(
    'input_meta_data_path', None,
    'Path to file that contains meta data about input '
    'to be used for training and evaluation.')
53
54
55
flags.DEFINE_integer('train_data_size', None, 'Number of training samples '
                     'to use. If None, uses the full train data. '
                     '(default: None).')
Hongkun Yu's avatar
Hongkun Yu committed
56
57
flags.DEFINE_string('predict_checkpoint_path', None,
                    'Path to the checkpoint for predictions.')
Tianqi Liu's avatar
Tianqi Liu committed
58
59
60
61
62
63
flags.DEFINE_integer(
    'num_eval_per_epoch', 1,
    'Number of evaluations per epoch. The purpose of this flag is to provide '
    'more granular evaluation scores and checkpoints. For example, if original '
    'data has N samples and num_eval_per_epoch is n, then each epoch will be '
    'evaluated every N/n samples.')
64
flags.DEFINE_integer('train_batch_size', 32, 'Batch size for training.')
65
flags.DEFINE_integer('eval_batch_size', 32, 'Batch size for evaluation.')
66
67

common_flags.define_common_bert_flags()
68
69
70

FLAGS = flags.FLAGS

71
72
LABEL_TYPES_MAP = {'int': tf.int64, 'float': tf.float32}

73

74
def get_loss_fn(num_classes):
75
76
77
78
79
80
81
82
83
84
  """Gets the classification loss function."""

  def classification_loss_fn(labels, logits):
    """Classification loss."""
    labels = tf.squeeze(labels)
    log_probs = tf.nn.log_softmax(logits, axis=-1)
    one_hot_labels = tf.one_hot(
        tf.cast(labels, dtype=tf.int32), depth=num_classes, dtype=tf.float32)
    per_example_loss = -tf.reduce_sum(
        tf.cast(one_hot_labels, dtype=tf.float32) * log_probs, axis=-1)
85
    return tf.reduce_mean(per_example_loss)
86
87
88
89

  return classification_loss_fn


Tianqi Liu's avatar
Tianqi Liu committed
90
91
92
93
def get_dataset_fn(input_file_pattern,
                   max_seq_length,
                   global_batch_size,
                   is_training,
94
                   label_type=tf.int64,
95
96
                   include_sample_weights=False,
                   num_samples=None):
Hongkun Yu's avatar
Hongkun Yu committed
97
98
99
100
101
102
103
  """Gets a closure to create a dataset."""

  def _dataset_fn(ctx=None):
    """Returns tf.data.Dataset for distributed BERT pretraining."""
    batch_size = ctx.get_per_replica_batch_size(
        global_batch_size) if ctx else global_batch_size
    dataset = input_pipeline.create_classifier_dataset(
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
104
        tf.io.gfile.glob(input_file_pattern),
Hongkun Yu's avatar
Hongkun Yu committed
105
106
107
        max_seq_length,
        batch_size,
        is_training=is_training,
108
        input_pipeline_context=ctx,
109
        label_type=label_type,
110
111
        include_sample_weights=include_sample_weights,
        num_samples=num_samples)
Hongkun Yu's avatar
Hongkun Yu committed
112
113
114
115
116
    return dataset

  return _dataset_fn


A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
117
118
119
120
121
122
123
124
125
126
127
def run_bert_classifier(strategy,
                        bert_config,
                        input_meta_data,
                        model_dir,
                        epochs,
                        steps_per_epoch,
                        steps_per_loop,
                        eval_steps,
                        warmup_steps,
                        initial_lr,
                        init_checkpoint,
Rajagopal Ananthanarayanan's avatar
Rajagopal Ananthanarayanan committed
128
129
                        train_input_fn,
                        eval_input_fn,
130
                        training_callbacks=True,
131
132
                        custom_callbacks=None,
                        custom_metrics=None):
133
134
  """Run BERT classifier training using low-level API."""
  max_seq_length = input_meta_data['max_seq_length']
135
136
  num_classes = input_meta_data.get('num_labels', 1)
  is_regression = num_classes == 1
137
138

  def _get_classifier_model():
139
    """Gets a classifier model."""
140
    classifier_model, core_model = (
141
142
143
144
        bert_models.classifier_model(
            bert_config,
            num_classes,
            max_seq_length,
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
145
146
            hub_module_url=FLAGS.hub_module_url,
            hub_module_trainable=FLAGS.hub_module_trainable))
Hongkun Yu's avatar
Hongkun Yu committed
147
148
149
150
    optimizer = optimization.create_optimizer(initial_lr,
                                              steps_per_epoch * epochs,
                                              warmup_steps, FLAGS.end_lr,
                                              FLAGS.optimizer_type)
151
152
153
154
    classifier_model.optimizer = performance.configure_optimizer(
        optimizer,
        use_float16=common_flags.use_float16(),
        use_graph_rewrite=common_flags.use_graph_rewrite())
155
156
    return classifier_model, core_model

157
158
159
160
161
162
  # tf.keras.losses objects accept optional sample_weight arguments (eg. coming
  # from the dataset) to compute weighted loss, as used for the regression
  # tasks. The classification tasks, using the custom get_loss_fn don't accept
  # sample weights though.
  loss_fn = (tf.keras.losses.MeanSquaredError() if is_regression
             else get_loss_fn(num_classes))
163
164
165

  # Defines evaluation metrics function, which will create metrics in the
  # correct device and strategy scope.
166
167
168
  if custom_metrics:
    metric_fn = custom_metrics
  elif is_regression:
Tianqi Liu's avatar
Tianqi Liu committed
169
170
171
172
    metric_fn = functools.partial(
        tf.keras.metrics.MeanSquaredError,
        'mean_squared_error',
        dtype=tf.float32)
173
  else:
Tianqi Liu's avatar
Tianqi Liu committed
174
175
176
177
    metric_fn = functools.partial(
        tf.keras.metrics.SparseCategoricalAccuracy,
        'accuracy',
        dtype=tf.float32)
178
179
180

  # Start training using Keras compile/fit API.
  logging.info('Training using TF 2.x Keras compile/fit API with '
Rajagopal Ananthanarayanan's avatar
Rajagopal Ananthanarayanan committed
181
               'distribution strategy.')
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
  return run_keras_compile_fit(
      model_dir,
      strategy,
      _get_classifier_model,
      train_input_fn,
      eval_input_fn,
      loss_fn,
      metric_fn,
      init_checkpoint,
      epochs,
      steps_per_epoch,
      steps_per_loop,
      eval_steps,
      training_callbacks=training_callbacks,
      custom_callbacks=custom_callbacks)
197
198


A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
199
200
201
202
203
204
205
206
207
208
def run_keras_compile_fit(model_dir,
                          strategy,
                          model_fn,
                          train_input_fn,
                          eval_input_fn,
                          loss_fn,
                          metric_fn,
                          init_checkpoint,
                          epochs,
                          steps_per_epoch,
Hongkun Yu's avatar
Hongkun Yu committed
209
                          steps_per_loop,
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
210
                          eval_steps,
211
                          training_callbacks=True,
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
212
213
214
215
216
                          custom_callbacks=None):
  """Runs BERT classifier model using Keras compile/fit API."""

  with strategy.scope():
    training_dataset = train_input_fn()
Le Hou's avatar
Le Hou committed
217
    evaluation_dataset = eval_input_fn() if eval_input_fn else None
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
218
219
220
221
222
223
224
    bert_model, sub_model = model_fn()
    optimizer = bert_model.optimizer

    if init_checkpoint:
      checkpoint = tf.train.Checkpoint(model=sub_model)
      checkpoint.restore(init_checkpoint).assert_existing_objects_matched()

225
226
    if not isinstance(metric_fn, (list, tuple)):
      metric_fn = [metric_fn]
Hongkun Yu's avatar
Hongkun Yu committed
227
228
229
    bert_model.compile(
        optimizer=optimizer,
        loss=loss_fn,
230
        metrics=[fn() for fn in metric_fn],
Hongkun Yu's avatar
Hongkun Yu committed
231
        experimental_steps_per_execution=steps_per_loop)
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
232

233
234
    summary_dir = os.path.join(model_dir, 'summaries')
    summary_callback = tf.keras.callbacks.TensorBoard(summary_dir)
Hongkun Yu's avatar
Hongkun Yu committed
235
236
237
238
239
240
241
242
    checkpoint = tf.train.Checkpoint(model=bert_model, optimizer=optimizer)
    checkpoint_manager = tf.train.CheckpointManager(
        checkpoint,
        directory=model_dir,
        max_to_keep=None,
        step_counter=optimizer.iterations,
        checkpoint_interval=0)
    checkpoint_callback = keras_utils.SimpleCheckpoint(checkpoint_manager)
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
243

244
245
246
247
248
    if training_callbacks:
      if custom_callbacks is not None:
        custom_callbacks += [summary_callback, checkpoint_callback]
      else:
        custom_callbacks = [summary_callback, checkpoint_callback]
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
249

250
    history = bert_model.fit(
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
251
252
253
254
255
256
        x=training_dataset,
        validation_data=evaluation_dataset,
        steps_per_epoch=steps_per_epoch,
        epochs=epochs,
        validation_steps=eval_steps,
        callbacks=custom_callbacks)
257
258
259
260
261
262
    stats = {'total_training_steps': steps_per_epoch * epochs}
    if 'loss' in history.history:
      stats['train_loss'] = history.history['loss'][-1]
    if 'val_accuracy' in history.history:
      stats['eval_metrics'] = history.history['val_accuracy'][-1]
    return bert_model, stats
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
263
264


Hongkun Yu's avatar
Hongkun Yu committed
265
266
267
def get_predictions_and_labels(strategy,
                               trained_model,
                               eval_input_fn,
268
                               is_regression=False,
Hongkun Yu's avatar
Hongkun Yu committed
269
                               return_probs=False):
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
270
271
272
273
274
275
276
277
278
  """Obtains predictions of trained model on evaluation data.

  Note that list of labels is returned along with the predictions because the
  order changes on distributing dataset over TPU pods.

  Args:
    strategy: Distribution strategy.
    trained_model: Trained model with preloaded weights.
    eval_input_fn: Input function for evaluation data.
279
    is_regression: Whether it is a regression task.
Hongkun Yu's avatar
Hongkun Yu committed
280
    return_probs: Whether to return probabilities of classes.
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
281
282
283
284
285
286
287
288
289
290
291
292
293

  Returns:
    predictions: List of predictions.
    labels: List of gold labels corresponding to predictions.
  """

  @tf.function
  def test_step(iterator):
    """Computes predictions on distributed devices."""

    def _test_step_fn(inputs):
      """Replicated predictions."""
      inputs, labels = inputs
Hongkun Yu's avatar
Hongkun Yu committed
294
      logits = trained_model(inputs, training=False)
295
      if not is_regression:
296
297
298
299
        probabilities = tf.nn.softmax(logits)
        return probabilities, labels
      else:
        return logits, labels
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
300

Hongkun Yu's avatar
Hongkun Yu committed
301
    outputs, labels = strategy.run(_test_step_fn, args=(next(iterator),))
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
302
303
304
305
306
307
308
309
310
    # outputs: current batch logits as a tuple of shard logits
    outputs = tf.nest.map_structure(strategy.experimental_local_results,
                                    outputs)
    labels = tf.nest.map_structure(strategy.experimental_local_results, labels)
    return outputs, labels

  def _run_evaluation(test_iterator):
    """Runs evaluation steps."""
    preds, golds = list(), list()
Hongkun Yu's avatar
Hongkun Yu committed
311
312
313
314
315
316
317
318
319
320
321
322
    try:
      with tf.experimental.async_scope():
        while True:
          probabilities, labels = test_step(test_iterator)
          for cur_probs, cur_labels in zip(probabilities, labels):
            if return_probs:
              preds.extend(cur_probs.numpy().tolist())
            else:
              preds.extend(tf.math.argmax(cur_probs, axis=1).numpy())
            golds.extend(cur_labels.numpy().tolist())
    except (StopIteration, tf.errors.OutOfRangeError):
      tf.experimental.async_clear_error()
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
323
324
    return preds, golds

Chenkai Kuang's avatar
Chenkai Kuang committed
325
  test_iter = iter(strategy.distribute_datasets_from_function(eval_input_fn))
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
326
327
328
329
330
  predictions, labels = _run_evaluation(test_iter)

  return predictions, labels


Hongkun Yu's avatar
Hongkun Yu committed
331
332
def export_classifier(model_export_path, input_meta_data, bert_config,
                      model_dir):
333
334
335
336
337
  """Exports a trained model as a `SavedModel` for inference.

  Args:
    model_export_path: a string specifying the path to the SavedModel directory.
    input_meta_data: dictionary containing meta data about input and model.
Rajagopal Ananthanarayanan's avatar
Rajagopal Ananthanarayanan committed
338
339
340
    bert_config: Bert configuration file to define core bert layers.
    model_dir: The directory where the model weights and training/evaluation
      summaries are stored.
341
342
343
344
345
346

  Raises:
    Export path is not specified, got an empty string or None.
  """
  if not model_export_path:
    raise ValueError('Export path is not specified: %s' % model_export_path)
Rajagopal Ananthanarayanan's avatar
Rajagopal Ananthanarayanan committed
347
348
  if not model_dir:
    raise ValueError('Export path is not specified: %s' % model_dir)
349

Zongwei Zhou's avatar
Zongwei Zhou committed
350
351
  # Export uses float32 for now, even if training uses mixed precision.
  tf.keras.mixed_precision.experimental.set_policy('float32')
352
  classifier_model = bert_models.classifier_model(
353
354
355
356
      bert_config,
      input_meta_data.get('num_labels', 1),
      hub_module_url=FLAGS.hub_module_url,
      hub_module_trainable=False)[0]
357

358
  model_saving_utils.export_bert_model(
Hongkun Yu's avatar
Hongkun Yu committed
359
      model_export_path, model=classifier_model, checkpoint_dir=model_dir)
360
361


Hongkun Yu's avatar
Hongkun Yu committed
362
363
def run_bert(strategy,
             input_meta_data,
364
             model_config,
Hongkun Yu's avatar
Hongkun Yu committed
365
             train_input_fn=None,
Le Hou's avatar
Le Hou committed
366
             eval_input_fn=None,
367
             init_checkpoint=None,
368
369
             custom_callbacks=None,
             custom_metrics=None):
370
  """Run BERT training."""
371
  # Enables XLA in Session Config. Should not be set for TPU.
372
  keras_utils.set_session_config(FLAGS.enable_xla)
373
  performance.set_mixed_precision_policy(common_flags.dtype())
374

Tianqi Liu's avatar
Tianqi Liu committed
375
376
377
  epochs = FLAGS.num_train_epochs * FLAGS.num_eval_per_epoch
  train_data_size = (
      input_meta_data['train_data_size'] // FLAGS.num_eval_per_epoch)
378
379
380
  if FLAGS.train_data_size:
    train_data_size = min(train_data_size, FLAGS.train_data_size)
    logging.info('Updated train_data_size: %s', train_data_size)
381
382
383
384
385
386
387
  steps_per_epoch = int(train_data_size / FLAGS.train_batch_size)
  warmup_steps = int(epochs * train_data_size * 0.1 / FLAGS.train_batch_size)
  eval_steps = int(
      math.ceil(input_meta_data['eval_data_size'] / FLAGS.eval_batch_size))

  if not strategy:
    raise ValueError('Distribution strategy has not been specified.')
A. Unique TensorFlower's avatar
A. Unique TensorFlower committed
388

389
390
391
  if not custom_callbacks:
    custom_callbacks = []

392
  if FLAGS.log_steps:
Hongkun Yu's avatar
Hongkun Yu committed
393
394
395
396
397
    custom_callbacks.append(
        keras_utils.TimeHistory(
            batch_size=FLAGS.train_batch_size,
            log_steps=FLAGS.log_steps,
            logdir=FLAGS.model_dir))
398

399
  trained_model, _ = run_bert_classifier(
400
      strategy,
401
      model_config,
402
403
404
405
      input_meta_data,
      FLAGS.model_dir,
      epochs,
      steps_per_epoch,
406
      FLAGS.steps_per_loop,
407
408
409
      eval_steps,
      warmup_steps,
      FLAGS.learning_rate,
Le Hou's avatar
Le Hou committed
410
      init_checkpoint or FLAGS.init_checkpoint,
Rajagopal Ananthanarayanan's avatar
Rajagopal Ananthanarayanan committed
411
412
      train_input_fn,
      eval_input_fn,
413
414
      custom_callbacks=custom_callbacks,
      custom_metrics=custom_metrics)
415

416
  if FLAGS.model_export_path:
417
    model_saving_utils.export_bert_model(
Hongkun Yu's avatar
Hongkun Yu committed
418
        FLAGS.model_export_path, model=trained_model)
419
420
  return trained_model

421

422
def custom_main(custom_callbacks=None, custom_metrics=None):
423
  """Run classification or regression.
424

425
426
  Args:
    custom_callbacks: list of tf.keras.Callbacks passed to training loop.
427
    custom_metrics: list of metrics passed to the training loop.
428
  """
Le Hou's avatar
Le Hou committed
429
430
  gin.parse_config_files_and_bindings(FLAGS.gin_file, FLAGS.gin_param)

431
432
  with tf.io.gfile.GFile(FLAGS.input_meta_data_path, 'rb') as reader:
    input_meta_data = json.loads(reader.read().decode('utf-8'))
433
  label_type = LABEL_TYPES_MAP[input_meta_data.get('label_type', 'int')]
434
  include_sample_weights = input_meta_data.get('has_sample_weights', False)
435
436
437
438

  if not FLAGS.model_dir:
    FLAGS.model_dir = '/tmp/bert20/'

Hongkun Yu's avatar
Hongkun Yu committed
439
440
441
442
443
444
445
  bert_config = bert_configs.BertConfig.from_json_file(FLAGS.bert_config_file)

  if FLAGS.mode == 'export_only':
    export_classifier(FLAGS.model_export_path, input_meta_data, bert_config,
                      FLAGS.model_dir)
    return

446
  strategy = distribute_utils.get_distribution_strategy(
447
448
449
      distribution_strategy=FLAGS.distribution_strategy,
      num_gpus=FLAGS.num_gpus,
      tpu_address=FLAGS.tpu)
Hongkun Yu's avatar
Hongkun Yu committed
450
  eval_input_fn = get_dataset_fn(
Rajagopal Ananthanarayanan's avatar
Rajagopal Ananthanarayanan committed
451
      FLAGS.eval_data_path,
Hongkun Yu's avatar
Hongkun Yu committed
452
      input_meta_data['max_seq_length'],
Hongkun Yu's avatar
Hongkun Yu committed
453
      FLAGS.eval_batch_size,
454
      is_training=False,
455
456
      label_type=label_type,
      include_sample_weights=include_sample_weights)
Hongkun Yu's avatar
Hongkun Yu committed
457

Hongkun Yu's avatar
Hongkun Yu committed
458
  if FLAGS.mode == 'predict':
459
    num_labels = input_meta_data.get('num_labels', 1)
Hongkun Yu's avatar
Hongkun Yu committed
460
461
    with strategy.scope():
      classifier_model = bert_models.classifier_model(
462
          bert_config, num_labels)[0]
Hongkun Yu's avatar
Hongkun Yu committed
463
464
465
466
467
468
469
470
471
472
      checkpoint = tf.train.Checkpoint(model=classifier_model)
      latest_checkpoint_file = (
          FLAGS.predict_checkpoint_path or
          tf.train.latest_checkpoint(FLAGS.model_dir))
      assert latest_checkpoint_file
      logging.info('Checkpoint file %s found and restoring from '
                   'checkpoint', latest_checkpoint_file)
      checkpoint.restore(
          latest_checkpoint_file).assert_existing_objects_matched()
      preds, _ = get_predictions_and_labels(
473
474
475
476
477
          strategy,
          classifier_model,
          eval_input_fn,
          is_regression=(num_labels == 1),
          return_probs=True)
Hongkun Yu's avatar
Hongkun Yu committed
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
    output_predict_file = os.path.join(FLAGS.model_dir, 'test_results.tsv')
    with tf.io.gfile.GFile(output_predict_file, 'w') as writer:
      logging.info('***** Predict results *****')
      for probabilities in preds:
        output_line = '\t'.join(
            str(class_probability)
            for class_probability in probabilities) + '\n'
        writer.write(output_line)
    return

  if FLAGS.mode != 'train_and_eval':
    raise ValueError('Unsupported mode is specified: %s' % FLAGS.mode)
  train_input_fn = get_dataset_fn(
      FLAGS.train_data_path,
      input_meta_data['max_seq_length'],
      FLAGS.train_batch_size,
494
      is_training=True,
495
      label_type=label_type,
496
497
      include_sample_weights=include_sample_weights,
      num_samples=FLAGS.train_data_size)
Hongkun Yu's avatar
Hongkun Yu committed
498
499
500
501
502
503
  run_bert(
      strategy,
      input_meta_data,
      bert_config,
      train_input_fn,
      eval_input_fn,
504
505
      custom_callbacks=custom_callbacks,
      custom_metrics=custom_metrics)
506
507
508


def main(_):
509
  custom_main(custom_callbacks=None, custom_metrics=None)
510
511
512
513
514


if __name__ == '__main__':
  flags.mark_flag_as_required('bert_config_file')
  flags.mark_flag_as_required('input_meta_data_path')
515
  flags.mark_flag_as_required('model_dir')
516
  app.run(main)