masked_softmax.py

# Copyright 2022 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.

"""Keras-based softmax layer with optional masking."""
# pylint: disable=g-classes-have-attributes

import tensorflow as tf


def _large_compatible_negative(tensor_type):
  """Large negative number as Tensor.

  This function is necessary because the standard value for epsilon
  in this module (-1e9) cannot be represented using `tf.float16`.

  Args:
    tensor_type: A dtype to determine the type.

  Returns:
    A large negative number.
  """
  if tensor_type == tf.float16:
    return tf.float16.min
  return -1e9


@tf.keras.utils.register_keras_serializable(package='Text')
class MaskedSoftmax(tf.keras.layers.Layer):
  """Performs a softmax with optional masking on a tensor.

  Args:
    mask_expansion_axes: Any axes that should be padded on the mask tensor.
    normalization_axes: On which axes the softmax should perform.
  """

  def __init__(self,
               mask_expansion_axes=None,
               normalization_axes=None,
               **kwargs):
    self._mask_expansion_axes = mask_expansion_axes
    if normalization_axes is None:
      self._normalization_axes = (-1,)
    else:
      self._normalization_axes = normalization_axes
    super().__init__(**kwargs)

  def call(self, scores, mask=None):

    if mask is not None:
      for _ in range(len(scores.shape) - len(mask.shape)):
        mask = tf.expand_dims(mask, axis=self._mask_expansion_axes)

      # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
      # masked positions, this operation will create a tensor which is 0.0 for
      # positions we want to attend and -1.e9 for masked positions.
      adder = (1.0 - tf.cast(mask, scores.dtype)) * _large_compatible_negative(
          scores.dtype)
      # Since we are adding it to the raw scores before the softmax, this is
      # effectively the same as removing these entirely.
      scores += adder

    if len(self._normalization_axes) == 1:
      return tf.nn.softmax(scores, axis=self._normalization_axes[0])
    else:
      return tf.math.exp(scores - tf.math.reduce_logsumexp(
          scores, axis=self._normalization_axes, keepdims=True))

  def get_config(self):
    config = {
        'mask_expansion_axes': self._mask_expansion_axes,
        'normalization_axes': self._normalization_axes
    }
    base_config = super().get_config()
    return dict(list(base_config.items()) + list(config.items()))