Python modeling.get_shape_list() Examples

The following are 30 code examples of modeling.get_shape_list(). You can vote up the ones you like or vote down the ones you don't like, and go to the original project or source file by following the links above each example. You may also want to check out all available functions/classes of the module modeling , or try the search function .
Example #1
Source File: layers.py    From Cross-Lingual-MRC with Apache License 2.0 6 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
  """
  Gathers the vectors at the specific positions over a minibatch.
  sequence_tensor: [batch, seq_length, width]
  positions: [batch, n]
  """
  sequence_shape  = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  batch_size      = sequence_shape[0]
  seq_length      = sequence_shape[1]
  width           = sequence_shape[2]

  flat_offsets    = tf.reshape(tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
  flat_positions  = tf.reshape(positions + flat_offsets, [-1])
  flat_sequence_tensor = tf.reshape(sequence_tensor, [batch_size * seq_length, width])
  output_tensor   = tf.gather(flat_sequence_tensor, flat_positions)

  return output_tensor


#
Example #2
Source File: layers.py    From Cross-Lingual-MRC with Apache License 2.0 6 votes vote down vote up 
def extract_span_tensor(bert_config, sequence_tensor, output_span_mask, 
                        start_positions, end_positions,   
                        scope=None):
  sequence_shape  = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  seq_length = sequence_shape[1]

  output_span_mask = tf.cast(output_span_mask, tf.float32)
  filtered_sequence_tensor = sequence_tensor * tf.expand_dims(output_span_mask, -1)

  with tf.variable_scope(scope, default_name="span_loss"):
    fw_output_tensor = gather_indexes(filtered_sequence_tensor, tf.expand_dims(start_positions, -1))
    bw_output_tensor = gather_indexes(filtered_sequence_tensor, tf.expand_dims(end_positions, -1))
    att_output_tensor = simple_attention(filtered_sequence_tensor, filtered_sequence_tensor, output_span_mask, self_adaptive=True)
    output_tensor = tf.concat([fw_output_tensor, bw_output_tensor, att_output_tensor], axis=-1)      

  return output_tensor


#
Example #3
Source File: run_squad.py    From QGforQA with MIT License 5 votes vote down vote up 
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
                 use_one_hot_embeddings):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  final_hidden = model.get_sequence_output()

  final_hidden_shape = modeling.get_shape_list(final_hidden, expected_rank=3)
  batch_size = final_hidden_shape[0]
  seq_length = final_hidden_shape[1]
  hidden_size = final_hidden_shape[2]

  output_weights = tf.get_variable(
      "cls/squad/output_weights", [2, hidden_size],
      initializer=tf.truncated_normal_initializer(stddev=0.02))

  output_bias = tf.get_variable(
      "cls/squad/output_bias", [2], initializer=tf.zeros_initializer())

  final_hidden_matrix = tf.reshape(final_hidden,
                                   [batch_size * seq_length, hidden_size])
  logits = tf.matmul(final_hidden_matrix, output_weights, transpose_b=True)
  logits = tf.nn.bias_add(logits, output_bias)

  logits = tf.reshape(logits, [batch_size, seq_length, 2])
  logits = tf.transpose(logits, [2, 0, 1])

  unstacked_logits = tf.unstack(logits, axis=0)

  (start_logits, end_logits) = (unstacked_logits[0], unstacked_logits[1])

  return (start_logits, end_logits)
Example #4
Source File: run_squad.py    From pynlp with MIT License 5 votes vote down vote up 
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
                 use_one_hot_embeddings):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  final_hidden = model.get_sequence_output()

  final_hidden_shape = modeling.get_shape_list(final_hidden, expected_rank=3)
  batch_size = final_hidden_shape[0]
  seq_length = final_hidden_shape[1]
  hidden_size = final_hidden_shape[2]

  output_weights = tf.get_variable(
      "cls/squad/output_weights", [2, hidden_size],
      initializer=tf.truncated_normal_initializer(stddev=0.02))

  output_bias = tf.get_variable(
      "cls/squad/output_bias", [2], initializer=tf.zeros_initializer())

  final_hidden_matrix = tf.reshape(final_hidden,
                                   [batch_size * seq_length, hidden_size])
  logits = tf.matmul(final_hidden_matrix, output_weights, transpose_b=True)
  logits = tf.nn.bias_add(logits, output_bias)

  logits = tf.reshape(logits, [batch_size, seq_length, 2])
  logits = tf.transpose(logits, [2, 0, 1])

  unstacked_logits = tf.unstack(logits, axis=0)

  (start_logits, end_logits) = (unstacked_logits[0], unstacked_logits[1])

  return (start_logits, end_logits)
Example #5
Source File: run_pretraining.py    From training with Apache License 2.0 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
  """Gathers the vectors at the specific positions over a minibatch."""
  sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  batch_size = sequence_shape[0]
  seq_length = sequence_shape[1]
  width = sequence_shape[2]

  flat_offsets = tf.reshape(
      tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
  flat_positions = tf.reshape(positions + flat_offsets, [-1])
  flat_sequence_tensor = tf.reshape(sequence_tensor,
                                    [batch_size * seq_length, width])
  output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
  return output_tensor
Example #6
Source File: run_squad.py    From BERT-for-Sequence-Labeling-and-Text-Classification with Apache License 2.0 5 votes vote down vote up 
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
                 use_one_hot_embeddings):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  final_hidden = model.get_sequence_output()

  final_hidden_shape = modeling.get_shape_list(final_hidden, expected_rank=3)
  batch_size = final_hidden_shape[0]
  seq_length = final_hidden_shape[1]
  hidden_size = final_hidden_shape[2]

  output_weights = tf.get_variable(
      "cls/squad/output_weights", [2, hidden_size],
      initializer=tf.truncated_normal_initializer(stddev=0.02))

  output_bias = tf.get_variable(
      "cls/squad/output_bias", [2], initializer=tf.zeros_initializer())

  final_hidden_matrix = tf.reshape(final_hidden,
                                   [batch_size * seq_length, hidden_size])
  logits = tf.matmul(final_hidden_matrix, output_weights, transpose_b=True)
  logits = tf.nn.bias_add(logits, output_bias)

  logits = tf.reshape(logits, [batch_size, seq_length, 2])
  logits = tf.transpose(logits, [2, 0, 1])

  unstacked_logits = tf.unstack(logits, axis=0)

  (start_logits, end_logits) = (unstacked_logits[0], unstacked_logits[1])

  return (start_logits, end_logits)
Example #7
Source File: run_pretraining.py    From BERT-for-Sequence-Labeling-and-Text-Classification with Apache License 2.0 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
  """Gathers the vectors at the specific positions over a minibatch."""
  sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  batch_size = sequence_shape[0]
  seq_length = sequence_shape[1]
  width = sequence_shape[2]

  flat_offsets = tf.reshape(
      tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
  flat_positions = tf.reshape(positions + flat_offsets, [-1])
  flat_sequence_tensor = tf.reshape(sequence_tensor,
                                    [batch_size * seq_length, width])
  output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
  return output_tensor
Example #8
Source File: run_squad.py    From MedicalRelationExtraction with MIT License 5 votes vote down vote up 
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
                 use_one_hot_embeddings):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  final_hidden = model.get_sequence_output()

  final_hidden_shape = modeling.get_shape_list(final_hidden, expected_rank=3)
  batch_size = final_hidden_shape[0]
  seq_length = final_hidden_shape[1]
  hidden_size = final_hidden_shape[2]

  output_weights = tf.get_variable(
      "cls/squad/output_weights", [2, hidden_size],
      initializer=tf.truncated_normal_initializer(stddev=0.02))

  output_bias = tf.get_variable(
      "cls/squad/output_bias", [2], initializer=tf.zeros_initializer())

  final_hidden_matrix = tf.reshape(final_hidden,
                                   [batch_size * seq_length, hidden_size])
  logits = tf.matmul(final_hidden_matrix, output_weights, transpose_b=True)
  logits = tf.nn.bias_add(logits, output_bias)

  logits = tf.reshape(logits, [batch_size, seq_length, 2])
  logits = tf.transpose(logits, [2, 0, 1])

  unstacked_logits = tf.unstack(logits, axis=0)

  (start_logits, end_logits) = (unstacked_logits[0], unstacked_logits[1])

  return (start_logits, end_logits)
Example #9
Source File: run_pretraining.py    From MedicalRelationExtraction with MIT License 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
  """Gathers the vectors at the specific positions over a minibatch."""
  sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  batch_size = sequence_shape[0]
  seq_length = sequence_shape[1]
  width = sequence_shape[2]

  flat_offsets = tf.reshape(
      tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
  flat_positions = tf.reshape(positions + flat_offsets, [-1])
  flat_sequence_tensor = tf.reshape(sequence_tensor,
                                    [batch_size * seq_length, width])
  output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
  return output_tensor
Example #10
Source File: run_pretraining_gpu_v2.py    From BERT-multi-gpu with Apache License 2.0 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
    """Gathers the vectors at the specific positions over a minibatch."""
    sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
    batch_size = sequence_shape[0]
    seq_length = sequence_shape[1]
    width = sequence_shape[2]

    flat_offsets = tf.reshape(
        tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
    flat_positions = tf.reshape(positions + flat_offsets, [-1])
    flat_sequence_tensor = tf.reshape(sequence_tensor,
                                      [batch_size * seq_length, width])
    output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
    return output_tensor
Example #11
Source File: run_pretraining_gpu.py    From BERT-multi-gpu with Apache License 2.0 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
    """Gathers the vectors at the specific positions over a minibatch."""
    sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
    batch_size = sequence_shape[0]
    seq_length = sequence_shape[1]
    width = sequence_shape[2]

    flat_offsets = tf.reshape(
        tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
    flat_positions = tf.reshape(positions + flat_offsets, [-1])
    flat_sequence_tensor = tf.reshape(sequence_tensor,
                                      [batch_size * seq_length, width])
    output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
    return output_tensor
Example #12
Source File: run_squad.py    From coref with Apache License 2.0 5 votes vote down vote up 
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
                 use_one_hot_embeddings):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  final_hidden = model.get_sequence_output()

  final_hidden_shape = modeling.get_shape_list(final_hidden, expected_rank=3)
  batch_size = final_hidden_shape[0]
  seq_length = final_hidden_shape[1]
  hidden_size = final_hidden_shape[2]

  output_weights = tf.get_variable(
      "cls/squad/output_weights", [2, hidden_size],
      initializer=tf.truncated_normal_initializer(stddev=0.02))

  output_bias = tf.get_variable(
      "cls/squad/output_bias", [2], initializer=tf.zeros_initializer())

  final_hidden_matrix = tf.reshape(final_hidden,
                                   [batch_size * seq_length, hidden_size])
  logits = tf.matmul(final_hidden_matrix, output_weights, transpose_b=True)
  logits = tf.nn.bias_add(logits, output_bias)

  logits = tf.reshape(logits, [batch_size, seq_length, 2])
  logits = tf.transpose(logits, [2, 0, 1])

  unstacked_logits = tf.unstack(logits, axis=0)

  (start_logits, end_logits) = (unstacked_logits[0], unstacked_logits[1])

  return (start_logits, end_logits)
Example #13
Source File: run_pretraining.py    From coref with Apache License 2.0 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
  """Gathers the vectors at the specific positions over a minibatch."""
  sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  batch_size = sequence_shape[0]
  seq_length = sequence_shape[1]
  width = sequence_shape[2]

  flat_offsets = tf.reshape(
      tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
  flat_positions = tf.reshape(positions + flat_offsets, [-1])
  flat_sequence_tensor = tf.reshape(sequence_tensor,
                                    [batch_size * seq_length, width])
  output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
  return output_tensor
Example #14
Source File: run_pretraining.py    From albert-chinese-ner with MIT License 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
  """Gathers the vectors at the specific positions over a minibatch."""
  sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  batch_size = sequence_shape[0]
  seq_length = sequence_shape[1]
  width = sequence_shape[2]

  flat_offsets = tf.reshape(
      tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
  flat_positions = tf.reshape(positions + flat_offsets, [-1])
  flat_sequence_tensor = tf.reshape(sequence_tensor,
                                    [batch_size * seq_length, width])
  output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
  return output_tensor
Example #15
Source File: run_pretraining.py    From bert-japanese with Apache License 2.0 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
  """Gathers the vectors at the specific positions over a minibatch."""
  sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  batch_size = sequence_shape[0]
  seq_length = sequence_shape[1]
  width = sequence_shape[2]

  flat_offsets = tf.reshape(
      tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
  flat_positions = tf.reshape(positions + flat_offsets, [-1])
  flat_sequence_tensor = tf.reshape(sequence_tensor,
                                    [batch_size * seq_length, width])
  output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
  return output_tensor
Example #16
Source File: run_pretraining.py    From BERT-sentiment--classification with Apache License 2.0 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
  """Gathers the vectors at the specific positions over a minibatch."""
  sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  batch_size = sequence_shape[0]
  seq_length = sequence_shape[1]
  width = sequence_shape[2]

  flat_offsets = tf.reshape(
      tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
  flat_positions = tf.reshape(positions + flat_offsets, [-1])
  flat_sequence_tensor = tf.reshape(sequence_tensor,
                                    [batch_size * seq_length, width])
  output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
  return output_tensor
Example #17
Source File: run_squad.py    From BERT-sentiment--classification with Apache License 2.0 5 votes vote down vote up 
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
                 use_one_hot_embeddings):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  final_hidden = model.get_sequence_output()

  final_hidden_shape = modeling.get_shape_list(final_hidden, expected_rank=3)
  batch_size = final_hidden_shape[0]
  seq_length = final_hidden_shape[1]
  hidden_size = final_hidden_shape[2]

  output_weights = tf.get_variable(
      "cls/squad/output_weights", [2, hidden_size],
      initializer=tf.truncated_normal_initializer(stddev=0.02))

  output_bias = tf.get_variable(
      "cls/squad/output_bias", [2], initializer=tf.zeros_initializer())

  final_hidden_matrix = tf.reshape(final_hidden,
                                   [batch_size * seq_length, hidden_size])
  logits = tf.matmul(final_hidden_matrix, output_weights, transpose_b=True)
  logits = tf.nn.bias_add(logits, output_bias)

  logits = tf.reshape(logits, [batch_size, seq_length, 2])
  logits = tf.transpose(logits, [2, 0, 1])

  unstacked_logits = tf.unstack(logits, axis=0)

  (start_logits, end_logits) = (unstacked_logits[0], unstacked_logits[1])

  return (start_logits, end_logits)
Example #18
Source File: run_squad.py    From training with Apache License 2.0 5 votes vote down vote up 
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
                 use_one_hot_embeddings):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  final_hidden = model.get_sequence_output()

  final_hidden_shape = modeling.get_shape_list(final_hidden, expected_rank=3)
  batch_size = final_hidden_shape[0]
  seq_length = final_hidden_shape[1]
  hidden_size = final_hidden_shape[2]

  output_weights = tf.get_variable(
      "cls/squad/output_weights", [2, hidden_size],
      initializer=tf.truncated_normal_initializer(stddev=0.02))

  output_bias = tf.get_variable(
      "cls/squad/output_bias", [2], initializer=tf.zeros_initializer())

  final_hidden_matrix = tf.reshape(final_hidden,
                                   [batch_size * seq_length, hidden_size])
  logits = tf.matmul(final_hidden_matrix, output_weights, transpose_b=True)
  logits = tf.nn.bias_add(logits, output_bias)

  logits = tf.reshape(logits, [batch_size, seq_length, 2])
  logits = tf.transpose(logits, [2, 0, 1])

  unstacked_logits = tf.unstack(logits, axis=0)

  (start_logits, end_logits) = (unstacked_logits[0], unstacked_logits[1])

  return (start_logits, end_logits)
Example #19
Source File: test_squad.py    From QGforQA with MIT License 5 votes vote down vote up 
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
                 use_one_hot_embeddings):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  final_hidden = model.get_sequence_output()

  final_hidden_shape = modeling.get_shape_list(final_hidden, expected_rank=3)
  batch_size = final_hidden_shape[0]
  seq_length = final_hidden_shape[1]
  hidden_size = final_hidden_shape[2]

  output_weights = tf.get_variable(
      "cls/squad/output_weights", [2, hidden_size],
      initializer=tf.truncated_normal_initializer(stddev=0.02))

  output_bias = tf.get_variable(
      "cls/squad/output_bias", [2], initializer=tf.zeros_initializer())

  final_hidden_matrix = tf.reshape(final_hidden,
                                   [batch_size * seq_length, hidden_size])
  logits = tf.matmul(final_hidden_matrix, output_weights, transpose_b=True)
  logits = tf.nn.bias_add(logits, output_bias)

  logits = tf.reshape(logits, [batch_size, seq_length, 2])
  logits = tf.transpose(logits, [2, 0, 1])

  unstacked_logits = tf.unstack(logits, axis=0)

  (start_logits, end_logits) = (unstacked_logits[0], unstacked_logits[1])

  return (start_logits, end_logits)
Example #20
Source File: run_pretraining.py    From BERT_STS-B with Apache License 2.0 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
  """Gathers the vectors at the specific positions over a minibatch."""
  sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  batch_size = sequence_shape[0]
  seq_length = sequence_shape[1]
  width = sequence_shape[2]

  flat_offsets = tf.reshape(
      tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
  flat_positions = tf.reshape(positions + flat_offsets, [-1])
  flat_sequence_tensor = tf.reshape(sequence_tensor,
                                    [batch_size * seq_length, width])
  output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
  return output_tensor
Example #21
Source File: run_squad.py    From BERT_STS-B with Apache License 2.0 5 votes vote down vote up 
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
                 use_one_hot_embeddings):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  final_hidden = model.get_sequence_output()

  final_hidden_shape = modeling.get_shape_list(final_hidden, expected_rank=3)
  batch_size = final_hidden_shape[0]
  seq_length = final_hidden_shape[1]
  hidden_size = final_hidden_shape[2]

  output_weights = tf.get_variable(
      "cls/squad/output_weights", [2, hidden_size],
      initializer=tf.truncated_normal_initializer(stddev=0.02))

  output_bias = tf.get_variable(
      "cls/squad/output_bias", [2], initializer=tf.zeros_initializer())

  final_hidden_matrix = tf.reshape(final_hidden,
                                   [batch_size * seq_length, hidden_size])
  logits = tf.matmul(final_hidden_matrix, output_weights, transpose_b=True)
  logits = tf.nn.bias_add(logits, output_bias)

  logits = tf.reshape(logits, [batch_size, seq_length, 2])
  logits = tf.transpose(logits, [2, 0, 1])

  unstacked_logits = tf.unstack(logits, axis=0)

  (start_logits, end_logits) = (unstacked_logits[0], unstacked_logits[1])

  return (start_logits, end_logits)
Example #22
Source File: run_pretraining.py    From KBQA-BERT with MIT License 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
  """Gathers the vectors at the specific positions over a minibatch."""
  sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  batch_size = sequence_shape[0]
  seq_length = sequence_shape[1]
  width = sequence_shape[2]

  flat_offsets = tf.reshape(
      tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
  flat_positions = tf.reshape(positions + flat_offsets, [-1])
  flat_sequence_tensor = tf.reshape(sequence_tensor,
                                    [batch_size * seq_length, width])
  output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
  return output_tensor
Example #23
Source File: run_squad.py    From KBQA-BERT with MIT License 5 votes vote down vote up 
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
                 use_one_hot_embeddings):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  final_hidden = model.get_sequence_output()

  final_hidden_shape = modeling.get_shape_list(final_hidden, expected_rank=3)
  batch_size = final_hidden_shape[0]
  seq_length = final_hidden_shape[1]
  hidden_size = final_hidden_shape[2]

  output_weights = tf.get_variable(
      "cls/squad/output_weights", [2, hidden_size],
      initializer=tf.truncated_normal_initializer(stddev=0.02))

  output_bias = tf.get_variable(
      "cls/squad/output_bias", [2], initializer=tf.zeros_initializer())

  final_hidden_matrix = tf.reshape(final_hidden,
                                   [batch_size * seq_length, hidden_size])
  logits = tf.matmul(final_hidden_matrix, output_weights, transpose_b=True)
  logits = tf.nn.bias_add(logits, output_bias)

  logits = tf.reshape(logits, [batch_size, seq_length, 2])
  logits = tf.transpose(logits, [2, 0, 1])

  unstacked_logits = tf.unstack(logits, axis=0)

  (start_logits, end_logits) = (unstacked_logits[0], unstacked_logits[1])

  return (start_logits, end_logits)
Example #24
Source File: run_pretraining.py    From bert-qa with MIT License 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
  """Gathers the vectors at the specific positions over a minibatch."""
  sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  batch_size = sequence_shape[0]
  seq_length = sequence_shape[1]
  width = sequence_shape[2]

  flat_offsets = tf.reshape(
      tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
  flat_positions = tf.reshape(positions + flat_offsets, [-1])
  flat_sequence_tensor = tf.reshape(sequence_tensor,
                                    [batch_size * seq_length, width])
  output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
  return output_tensor
Example #25
Source File: run_squad.py    From bert-qa with MIT License 5 votes vote down vote up 
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
                 use_one_hot_embeddings):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  final_hidden = model.get_sequence_output()

  final_hidden_shape = modeling.get_shape_list(final_hidden, expected_rank=3)
  batch_size = final_hidden_shape[0]
  seq_length = final_hidden_shape[1]
  hidden_size = final_hidden_shape[2]

  output_weights = tf.get_variable(
      "cls/squad/output_weights", [2, hidden_size],
      initializer=tf.truncated_normal_initializer(stddev=0.02))

  output_bias = tf.get_variable(
      "cls/squad/output_bias", [2], initializer=tf.zeros_initializer())

  final_hidden_matrix = tf.reshape(final_hidden,
                                   [batch_size * seq_length, hidden_size])
  logits = tf.matmul(final_hidden_matrix, output_weights, transpose_b=True)
  logits = tf.nn.bias_add(logits, output_bias)

  logits = tf.reshape(logits, [batch_size, seq_length, 2])
  logits = tf.transpose(logits, [2, 0, 1])

  unstacked_logits = tf.unstack(logits, axis=0)

  (start_logits, end_logits) = (unstacked_logits[0], unstacked_logits[1])

  return (start_logits, end_logits)
Example #26
Source File: run_pretraining.py    From delft with Apache License 2.0 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
  """Gathers the vectors at the specific positions over a minibatch."""
  sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  batch_size = sequence_shape[0]
  seq_length = sequence_shape[1]
  width = sequence_shape[2]

  flat_offsets = tf.reshape(
      tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
  flat_positions = tf.reshape(positions + flat_offsets, [-1])
  flat_sequence_tensor = tf.reshape(sequence_tensor,
                                    [batch_size * seq_length, width])
  output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
  return output_tensor
Example #27
Source File: run_squad.py    From delft with Apache License 2.0 5 votes vote down vote up 
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
                 use_one_hot_embeddings):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  final_hidden = model.get_sequence_output()

  final_hidden_shape = modeling.get_shape_list(final_hidden, expected_rank=3)
  batch_size = final_hidden_shape[0]
  seq_length = final_hidden_shape[1]
  hidden_size = final_hidden_shape[2]

  output_weights = tf.get_variable(
      "cls/squad/output_weights", [2, hidden_size],
      initializer=tf.truncated_normal_initializer(stddev=0.02))

  output_bias = tf.get_variable(
      "cls/squad/output_bias", [2], initializer=tf.zeros_initializer())

  final_hidden_matrix = tf.reshape(final_hidden,
                                   [batch_size * seq_length, hidden_size])
  logits = tf.matmul(final_hidden_matrix, output_weights, transpose_b=True)
  logits = tf.nn.bias_add(logits, output_bias)

  logits = tf.reshape(logits, [batch_size, seq_length, 2])
  logits = tf.transpose(logits, [2, 0, 1])

  unstacked_logits = tf.unstack(logits, axis=0)

  (start_logits, end_logits) = (unstacked_logits[0], unstacked_logits[1])

  return (start_logits, end_logits)
Example #28
Source File: run_pretraining.py    From uai-sdk with Apache License 2.0 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
  """Gathers the vectors at the specific positions over a minibatch."""
  sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  batch_size = sequence_shape[0]
  seq_length = sequence_shape[1]
  width = sequence_shape[2]

  flat_offsets = tf.reshape(
      tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
  flat_positions = tf.reshape(positions + flat_offsets, [-1])
  flat_sequence_tensor = tf.reshape(sequence_tensor,
                                    [batch_size * seq_length, width])
  output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
  return output_tensor
Example #29
Source File: run_squad.py    From uai-sdk with Apache License 2.0 5 votes vote down vote up 
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
                 use_one_hot_embeddings):
  """Creates a classification model."""
  model = modeling.BertModel(
      config=bert_config,
      is_training=is_training,
      input_ids=input_ids,
      input_mask=input_mask,
      token_type_ids=segment_ids,
      use_one_hot_embeddings=use_one_hot_embeddings)

  final_hidden = model.get_sequence_output()

  final_hidden_shape = modeling.get_shape_list(final_hidden, expected_rank=3)
  batch_size = final_hidden_shape[0]
  seq_length = final_hidden_shape[1]
  hidden_size = final_hidden_shape[2]

  output_weights = tf.get_variable(
      "cls/squad/output_weights", [2, hidden_size],
      initializer=tf.truncated_normal_initializer(stddev=0.02))

  output_bias = tf.get_variable(
      "cls/squad/output_bias", [2], initializer=tf.zeros_initializer())

  final_hidden_matrix = tf.reshape(final_hidden,
                                   [batch_size * seq_length, hidden_size])
  logits = tf.matmul(final_hidden_matrix, output_weights, transpose_b=True)
  logits = tf.nn.bias_add(logits, output_bias)

  logits = tf.reshape(logits, [batch_size, seq_length, 2])
  logits = tf.transpose(logits, [2, 0, 1])

  unstacked_logits = tf.unstack(logits, axis=0)

  (start_logits, end_logits) = (unstacked_logits[0], unstacked_logits[1])

  return (start_logits, end_logits)
Example #30
Source File: run_pretraining.py    From bert-as-language-model with Apache License 2.0 5 votes vote down vote up 
def gather_indexes(sequence_tensor, positions):
  """Gathers the vectors at the specific positions over a minibatch."""
  sequence_shape = modeling.get_shape_list(sequence_tensor, expected_rank=3)
  batch_size = sequence_shape[0]
  seq_length = sequence_shape[1]
  width = sequence_shape[2]

  flat_offsets = tf.reshape(
      tf.range(0, batch_size, dtype=tf.int32) * seq_length, [-1, 1])
  flat_positions = tf.reshape(positions + flat_offsets, [-1])
  flat_sequence_tensor = tf.reshape(sequence_tensor,
                                    [batch_size * seq_length, width])
  output_tensor = tf.gather(flat_sequence_tensor, flat_positions)
  return output_tensor