#TF-NLP #Funnel-Transformer Initial check-in of the Funnel Transformer framework.

The current Funnel-Transformer contains only the encoder part and uses 1D-MaxPool for query pooling.

PiperOrigin-RevId: 392961143

official/nlp/modeling/networks/funnel_transformer.py

+# Copyright 2021 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.
+
+"""Funnel Transformer network."""
+# pylint: disable=g-classes-have-attributes
+from typing import Union, Collection
+from absl import logging
+import tensorflow as tf
+
+from official.nlp import keras_nlp
+
+
+def _pool_and_concat(data, unpool_length: int, stride: int,
+                     axes: Union[Collection[int], int]):
+  """Pools the data along a given axis with stride.
+
+  It also skips first unpool_length elements.
+
+  Args:
+    data: Tensor to be pooled.
+    unpool_length: Leading elements to be skipped.
+    stride: Stride for the given axis.
+    axes: Axes to pool the Tensor.
+
+  Returns:
+    Pooled and concatenated Tensor.
+  """
+  # Wraps the axes as a list.
+  if isinstance(axes, int):
+    axes = [axes]
+
+  for axis in axes:
+    # Skips first `unpool_length` tokens.
+    unpool_tensor_shape = [slice(None)] * axis + [slice(None, unpool_length)]
+    unpool_tensor = data[unpool_tensor_shape]
+    # Pools the second half.
+    pool_tensor_shape = [slice(None)] * axis + [
+        slice(unpool_length, None, stride)
+    ]
+    pool_tensor = data[pool_tensor_shape]
+    data = tf.concat((unpool_tensor, pool_tensor), axis=axis)
+  return data
+
+
+@tf.keras.utils.register_keras_serializable(package='Text')
+class FunnelTransformerEncoder(tf.keras.layers.Layer):
+  """Funnel Transformer-based encoder network.
+
+  Funnel Transformer Implementation of https://arxiv.org/abs/2006.03236.
+  This implementation utilizes the base framework with Bert
+  (https://arxiv.org/abs/1810.04805).
+  Its output is compatible with `BertEncoder`.
+
+  Args:
+    vocab_size: The size of the token vocabulary.
+    hidden_size: The size of the transformer hidden layers.
+    num_layers: The number of transformer layers.
+    num_attention_heads: The number of attention heads for each transformer. The
+      hidden size must be divisible by the number of attention heads.
+    max_sequence_length: The maximum sequence length that this encoder can
+      consume. If None, max_sequence_length uses the value from sequence length.
+      This determines the variable shape for positional embeddings.
+    type_vocab_size: The number of types that the 'type_ids' input can take.
+    inner_dim: The output dimension of the first Dense layer in a two-layer
+      feedforward network for each transformer.
+    inner_activation: The activation for the first Dense layer in a two-layer
+      feedforward network for each transformer.
+    output_dropout: Dropout probability for the post-attention and output
+      dropout.
+    attention_dropout: The dropout rate to use for the attention layers within
+      the transformer layers.
+    pool_stride: Pooling stride to compress the sequence length.
+    unpool_length: Leading n tokens to be skipped from pooling.
+    initializer: The initialzer to use for all weights in this encoder.
+    output_range: The sequence output range, [0, output_range), by slicing the
+      target sequence of the last transformer layer. `None` means the entire
+      target sequence will attend to the source sequence, which yields the full
+      output.
+    embedding_width: The width of the word embeddings. If the embedding width is
+      not equal to hidden size, embedding parameters will be factorized into two
+      matrices in the shape of ['vocab_size', 'embedding_width'] and
+      ['embedding_width', 'hidden_size'] ('embedding_width' is usually much
+      smaller than 'hidden_size').
+    embedding_layer: An optional Layer instance which will be called to generate
+      embeddings for the input word IDs.
+    norm_first: Whether to normalize inputs to attention and intermediate dense
+      layers. If set False, output of attention and intermediate dense layers is
+      normalized.
+  """
+
+  def __init__(
+      self,
+      vocab_size,
+      hidden_size=768,
+      num_layers=12,
+      num_attention_heads=12,
+      max_sequence_length=512,
+      type_vocab_size=16,
+      inner_dim=3072,
+      inner_activation=lambda x: tf.keras.activations.gelu(x, approximate=True),
+      output_dropout=0.1,
+      attention_dropout=0.1,
+      pool_stride=2,
+      unpool_length=0,
+      initializer=tf.keras.initializers.TruncatedNormal(stddev=0.02),
+      output_range=None,
+      embedding_width=None,
+      embedding_layer=None,
+      norm_first=False,
+      **kwargs):
+    super().__init__(**kwargs)
+    activation = tf.keras.activations.get(inner_activation)
+    initializer = tf.keras.initializers.get(initializer)
+
+    if embedding_width is None:
+      embedding_width = hidden_size
+
+    if embedding_layer is None:
+      self._embedding_layer = keras_nlp.layers.OnDeviceEmbedding(
+          vocab_size=vocab_size,
+          embedding_width=embedding_width,
+          initializer=initializer,
+          name='word_embeddings')
+    else:
+      self._embedding_layer = embedding_layer
+
+    self._position_embedding_layer = keras_nlp.layers.PositionEmbedding(
+        initializer=initializer,
+        max_length=max_sequence_length,
+        name='position_embedding')
+
+    self._type_embedding_layer = keras_nlp.layers.OnDeviceEmbedding(
+        vocab_size=type_vocab_size,
+        embedding_width=embedding_width,
+        initializer=initializer,
+        use_one_hot=True,
+        name='type_embeddings')
+
+    self._embedding_norm_layer = tf.keras.layers.LayerNormalization(
+        name='embeddings/layer_norm', axis=-1, epsilon=1e-12, dtype=tf.float32)
+
+    self._embedding_dropout = tf.keras.layers.Dropout(
+        rate=output_dropout, name='embedding_dropout')
+
+    # We project the 'embedding' output to 'hidden_size' if it is not already
+    # 'hidden_size'.
+    self._embedding_projection = None
+    if embedding_width != hidden_size:
+      self._embedding_projection = tf.keras.layers.experimental.EinsumDense(
+          '...x,xy->...y',
+          output_shape=hidden_size,
+          bias_axes='y',
+          kernel_initializer=initializer,
+          name='embedding_projection')
+
+    self._transformer_layers = []
+    self._attention_mask_layer = keras_nlp.layers.SelfAttentionMask(
+        name='self_attention_mask')
+    for i in range(num_layers):
+      layer = keras_nlp.layers.TransformerEncoderBlock(
+          num_attention_heads=num_attention_heads,
+          inner_dim=inner_dim,
+          inner_activation=inner_activation,
+          output_dropout=output_dropout,
+          attention_dropout=attention_dropout,
+          norm_first=norm_first,
+          output_range=output_range if i == num_layers - 1 else None,
+          kernel_initializer=initializer,
+          name='transformer/layer_%d' % i)
+      self._transformer_layers.append(layer)
+
+    self._pooler_layer = tf.keras.layers.Dense(
+        units=hidden_size,
+        activation='tanh',
+        kernel_initializer=initializer,
+        name='pooler_transform')
+    self._att_input_pool_layer = tf.keras.layers.MaxPooling1D(
+        pool_size=pool_stride,
+        strides=pool_stride,
+        padding='same',
+        name='att_input_pool_layer')
+    self._pool_stride = pool_stride
+    self._unpool_length = unpool_length
+
+    self._config = {
+        'vocab_size': vocab_size,
+        'hidden_size': hidden_size,
+        'num_layers': num_layers,
+        'num_attention_heads': num_attention_heads,
+        'max_sequence_length': max_sequence_length,
+        'type_vocab_size': type_vocab_size,
+        'inner_dim': inner_dim,
+        'inner_activation': tf.keras.activations.serialize(activation),
+        'output_dropout': output_dropout,
+        'attention_dropout': attention_dropout,
+        'initializer': tf.keras.initializers.serialize(initializer),
+        'output_range': output_range,
+        'embedding_width': embedding_width,
+        'embedding_layer': embedding_layer,
+        'norm_first': norm_first,
+        'pool_stride': pool_stride,
+        'unpool_length': unpool_length,
+    }
+
+  def call(self, inputs):
+    # inputs are [word_ids, mask, type_ids]
+    if isinstance(inputs, (list, tuple)):
+      logging.warning('List inputs to  %s are discouraged.', self.__class__)
+      if len(inputs) == 3:
+        word_ids, mask, type_ids = inputs
+      else:
+        raise ValueError('Unexpected inputs to %s with length at %d.' %
+                         (self.__class__, len(inputs)))
+    elif isinstance(inputs, dict):
+      word_ids = inputs.get('input_word_ids')
+      mask = inputs.get('input_mask')
+      type_ids = inputs.get('input_type_ids')
+    else:
+      raise ValueError('Unexpected inputs type to %s.' % self.__class__)
+
+    word_embeddings = self._embedding_layer(word_ids)
+    # absolute position embeddings
+    position_embeddings = self._position_embedding_layer(word_embeddings)
+    type_embeddings = self._type_embedding_layer(type_ids)
+
+    embeddings = tf.keras.layers.add(
+        [word_embeddings, position_embeddings, type_embeddings])
+    embeddings = self._embedding_norm_layer(embeddings)
+    embeddings = self._embedding_dropout(embeddings)
+
+    if self._embedding_projection is not None:
+      embeddings = self._embedding_projection(embeddings)
+
+    attention_mask = self._attention_mask_layer(embeddings, mask)
+
+    encoder_outputs = []
+    x = embeddings
+    # TODO(b/195972228): attention_mask can be co-generated with pooling.
+    attention_mask = _pool_and_concat(
+        attention_mask,
+        unpool_length=self._unpool_length,
+        stride=self._pool_stride,
+        axes=[1])
+    for layer in self._transformer_layers:
+      # Pools layer for compressing the query length.
+      pooled_inputs = self._att_input_pool_layer(x[:, self._unpool_length:, :])
+      query_inputs = tf.concat(
+          values=(tf.cast(
+              x[:, :self._unpool_length, :],
+              dtype=pooled_inputs.dtype), pooled_inputs),
+          axis=1)
+      x = layer([query_inputs, x, attention_mask])
+      # Pools the corresponding attention_mask.
+      attention_mask = _pool_and_concat(
+          attention_mask,
+          unpool_length=self._unpool_length,
+          stride=self._pool_stride,
+          axes=[1, 2])
+      encoder_outputs.append(x)
+
+    last_encoder_output = encoder_outputs[-1]
+    first_token_tensor = last_encoder_output[:, 0, :]
+    pooled_output = self._pooler_layer(first_token_tensor)
+
+    return dict(
+        sequence_output=encoder_outputs[-1],
+        pooled_output=pooled_output,
+        encoder_outputs=encoder_outputs)
+
+  def get_embedding_table(self):
+    return self._embedding_layer.embeddings
+
+  def get_embedding_layer(self):
+    return self._embedding_layer
+
+  def get_config(self):
+    return dict(self._config)
+
+  @property
+  def transformer_layers(self):
+    """List of Transformer layers in the encoder."""
+    return self._transformer_layers
+
+  @property
+  def pooler_layer(self):
+    """The pooler dense layer after the transformer layers."""
+    return self._pooler_layer
+
+  @classmethod
+  def from_config(cls, config, custom_objects=None):
+    if 'embedding_layer' in config and config['embedding_layer'] is not None:
+      warn_string = (
+          'You are reloading a model that was saved with a '
+          'potentially-shared embedding layer object. If you contine to '
+          'train this model, the embedding layer will no longer be shared. '
+          'To work around this, load the model outside of the Keras API.')
+      print('WARNING: ' + warn_string)
+      logging.warn(warn_string)
+
+    return cls(**config)

official/nlp/modeling/networks/funnel_transformer_test.py

+# Copyright 2021 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.
+
+"""Tests for transformer-based bert encoder network."""
+
+from absl.testing import parameterized
+import numpy as np
+import tensorflow as tf
+
+from official.nlp.modeling.networks import funnel_transformer
+
+
+class SingleLayerModel(tf.keras.Model):
+
+  def __init__(self, layer):
+    super().__init__()
+    self.layer = layer
+
+  def call(self, inputs):
+    return self.layer(inputs)
+
+
+class FunnelTransformerEncoderTest(parameterized.TestCase, tf.test.TestCase):
+
+  def tearDown(self):
+    super(FunnelTransformerEncoderTest, self).tearDown()
+    tf.keras.mixed_precision.set_global_policy("float32")
+
+  @parameterized.named_parameters(("mix", "mixed_float16", tf.float16),
+                                  ("float32", "float32", tf.float32))
+  def test_network_creation(self, policy, pooled_dtype):
+    tf.keras.mixed_precision.set_global_policy(policy)
+
+    hidden_size = 32
+    sequence_length = 21
+    pool_stride = 2
+    num_layers = 3
+    # Create a small FunnelTransformerEncoder for testing.
+    test_network = funnel_transformer.FunnelTransformerEncoder(
+        vocab_size=100,
+        hidden_size=hidden_size,
+        num_attention_heads=2,
+        num_layers=num_layers,
+        pool_stride=pool_stride,
+        unpool_length=0)
+    # Create the inputs (note that the first dimension is implicit).
+    word_ids = tf.keras.Input(shape=(sequence_length,), dtype=tf.int32)
+    mask = tf.keras.Input(shape=(sequence_length,), dtype=tf.int32)
+    type_ids = tf.keras.Input(shape=(sequence_length,), dtype=tf.int32)
+    dict_outputs = test_network([word_ids, mask, type_ids])
+    data = dict_outputs["sequence_output"]
+    pooled = dict_outputs["pooled_output"]
+
+    self.assertIsInstance(test_network.transformer_layers, list)
+    self.assertLen(test_network.transformer_layers, num_layers)
+    self.assertIsInstance(test_network.pooler_layer, tf.keras.layers.Dense)
+
+    # Stride=2 compresses sequence length to half the size at each layer.
+    # This configuration gives each layer of seq length: 21->11->6->3.
+    expected_data_shape = [None, 3, hidden_size]
+    expected_pooled_shape = [None, hidden_size]
+
+    self.assertAllEqual(expected_data_shape, data.shape.as_list())
+    self.assertAllEqual(expected_pooled_shape, pooled.shape.as_list())
+
+    # The default output dtype is float32.
+    # If float_dtype is set to float16, the data output is float32 (from a layer
+    # norm) and pool output should be float16.
+    self.assertAllEqual(tf.float32, data.dtype)
+    self.assertAllEqual(pooled_dtype, pooled.dtype)
+
+  @parameterized.named_parameters(
+      ("large_stride_no_unpool", 3, 0),
+      ("no_stride_no_unpool", 1, 0),
+      ("large_stride_with_unpool", 3, 1),
+      ("large_stride_with_large_unpool", 5, 10),
+      ("no_stride_with_unpool", 1, 1),
+  )
+  def test_all_encoder_outputs_network_creation(self, pool_stride,
+                                                unpool_length):
+    hidden_size = 32
+    sequence_length = 21
+    num_layers = 3
+    # Create a small FunnelTransformerEncoder for testing.
+    test_network = funnel_transformer.FunnelTransformerEncoder(
+        vocab_size=100,
+        hidden_size=hidden_size,
+        num_attention_heads=2,
+        num_layers=num_layers,
+        pool_stride=pool_stride,
+        unpool_length=unpool_length)
+    # Create the inputs (note that the first dimension is implicit).
+    word_ids = tf.keras.Input(shape=(sequence_length,), dtype=tf.int32)
+    mask = tf.keras.Input(shape=(sequence_length,), dtype=tf.int32)
+    type_ids = tf.keras.Input(shape=(sequence_length,), dtype=tf.int32)
+    dict_outputs = test_network([word_ids, mask, type_ids])
+    all_encoder_outputs = dict_outputs["encoder_outputs"]
+    pooled = dict_outputs["pooled_output"]
+
+    expected_data_shape = [None, sequence_length, hidden_size]
+    expected_pooled_shape = [None, hidden_size]
+    self.assertLen(all_encoder_outputs, num_layers)
+    for data in all_encoder_outputs:
+      expected_data_shape[1] = unpool_length + (expected_data_shape[1] +
+                                                pool_stride - 1 -
+                                                unpool_length) // pool_stride
+      self.assertAllEqual(expected_data_shape, data.shape.as_list())
+    self.assertAllEqual(expected_pooled_shape, pooled.shape.as_list())
+
+    # The default output dtype is float32.
+    self.assertAllEqual(tf.float32, all_encoder_outputs[-1].dtype)
+    self.assertAllEqual(tf.float32, pooled.dtype)
+
+  @parameterized.named_parameters(
+      ("all_sequence", None, 3, 0),
+      ("output_range", 1, 1, 0),
+      ("all_sequence_wit_unpool", None, 4, 1),
+      ("output_range_with_unpool", 1, 1, 1),
+      ("output_range_with_large_unpool", 1, 1, 2),
+  )
+  def test_network_invocation(self, output_range, out_seq_len, unpool_length):
+    hidden_size = 32
+    sequence_length = 21
+    vocab_size = 57
+    num_types = 7
+    pool_stride = 2
+    # Create a small FunnelTransformerEncoder for testing.
+    test_network = funnel_transformer.FunnelTransformerEncoder(
+        vocab_size=vocab_size,
+        hidden_size=hidden_size,
+        num_attention_heads=2,
+        num_layers=3,
+        type_vocab_size=num_types,
+        output_range=output_range,
+        pool_stride=pool_stride,
+        unpool_length=unpool_length)
+    # Create the inputs (note that the first dimension is implicit).
+    word_ids = tf.keras.Input(shape=(sequence_length,), dtype=tf.int32)
+    mask = tf.keras.Input(shape=(sequence_length,), dtype=tf.int32)
+    type_ids = tf.keras.Input(shape=(sequence_length,), dtype=tf.int32)
+    dict_outputs = test_network([word_ids, mask, type_ids])
+    data = dict_outputs["sequence_output"]
+    pooled = dict_outputs["pooled_output"]
+
+    # Create a model based off of this network:
+    model = tf.keras.Model([word_ids, mask, type_ids], [data, pooled])
+
+    # Invoke the model. We can't validate the output data here (the model is too
+    # complex) but this will catch structural runtime errors.
+    batch_size = 3
+    word_id_data = np.random.randint(
+        vocab_size, size=(batch_size, sequence_length))
+    mask_data = np.random.randint(2, size=(batch_size, sequence_length))
+    type_id_data = np.random.randint(
+        num_types, size=(batch_size, sequence_length))
+    outputs = model.predict([word_id_data, mask_data, type_id_data])
+    self.assertEqual(outputs[0].shape[1], out_seq_len)  # output_range
+
+    # Creates a FunnelTransformerEncoder with max_sequence_length !=
+    # sequence_length
+    max_sequence_length = 128
+    test_network = funnel_transformer.FunnelTransformerEncoder(
+        vocab_size=vocab_size,
+        hidden_size=hidden_size,
+        max_sequence_length=max_sequence_length,
+        num_attention_heads=2,
+        num_layers=3,
+        type_vocab_size=num_types,
+        pool_stride=pool_stride)
+    dict_outputs = test_network([word_ids, mask, type_ids])
+    data = dict_outputs["sequence_output"]
+    pooled = dict_outputs["pooled_output"]
+    model = tf.keras.Model([word_ids, mask, type_ids], [data, pooled])
+    outputs = model.predict([word_id_data, mask_data, type_id_data])
+    self.assertEqual(outputs[0].shape[1], 3)
+
+    # Creates a FunnelTransformerEncoder with embedding_width != hidden_size
+    test_network = funnel_transformer.FunnelTransformerEncoder(
+        vocab_size=vocab_size,
+        hidden_size=hidden_size,
+        max_sequence_length=max_sequence_length,
+        num_attention_heads=2,
+        num_layers=3,
+        type_vocab_size=num_types,
+        embedding_width=16,
+        pool_stride=pool_stride)
+    dict_outputs = test_network([word_ids, mask, type_ids])
+    data = dict_outputs["sequence_output"]
+    pooled = dict_outputs["pooled_output"]
+    model = tf.keras.Model([word_ids, mask, type_ids], [data, pooled])
+    outputs = model.predict([word_id_data, mask_data, type_id_data])
+    self.assertEqual(outputs[0].shape[-1], hidden_size)
+    self.assertTrue(hasattr(test_network, "_embedding_projection"))
+
+  def test_serialize_deserialize(self):
+    # Create a network object that sets all of its config options.
+    kwargs = dict(
+        vocab_size=100,
+        hidden_size=32,
+        num_layers=3,
+        num_attention_heads=2,
+        max_sequence_length=21,
+        type_vocab_size=12,
+        inner_dim=1223,
+        inner_activation="relu",
+        output_dropout=0.05,
+        attention_dropout=0.22,
+        initializer="glorot_uniform",
+        output_range=-1,
+        embedding_width=16,
+        embedding_layer=None,
+        norm_first=False,
+        pool_stride=2,
+        unpool_length=0)
+    network = funnel_transformer.FunnelTransformerEncoder(**kwargs)
+    expected_config = dict(kwargs)
+    expected_config["inner_activation"] = tf.keras.activations.serialize(
+        tf.keras.activations.get(expected_config["inner_activation"]))
+    expected_config["initializer"] = tf.keras.initializers.serialize(
+        tf.keras.initializers.get(expected_config["initializer"]))
+    self.assertEqual(network.get_config(), expected_config)
+    # Create another network object from the first object's config.
+    new_network = funnel_transformer.FunnelTransformerEncoder.from_config(
+        network.get_config())
+
+    # If the serialization was successful, the new config should match the old.
+    self.assertAllEqual(network.get_config(), new_network.get_config())
+
+    # Tests model saving/loading.
+    model_path = self.get_temp_dir() + "/model"
+    network_wrapper = SingleLayerModel(network)
+    # One forward-path to ensure input_shape.
+    batch_size = 3
+    sequence_length = 21
+    vocab_size = 100
+    num_types = 12
+    word_id_data = np.random.randint(
+        vocab_size, size=(batch_size, sequence_length))
+    mask_data = np.random.randint(2, size=(batch_size, sequence_length))
+    type_id_data = np.random.randint(
+        num_types, size=(batch_size, sequence_length))
+
+    _ = network_wrapper.predict([word_id_data, mask_data, type_id_data])
+    network_wrapper.save(model_path)
+    _ = tf.keras.models.load_model(model_path)
+
+
+if __name__ == "__main__":
+  tf.test.main()