official/projects/qat/nlp/modeling/layers/transformer_encoder_block.py
# Copyright 2024 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 TransformerEncoder block layer."""
from typing import Optional
from absl import logging
import tensorflow as tf, tf_keras
import tensorflow_model_optimization as tfmot
from official.projects.qat.nlp.modeling.layers.multi_head_attention import MultiHeadAttentionQuantized
from official.projects.qat.nlp.quantization import configs
from official.projects.qat.nlp.quantization import wrappers
def _quantized_multi_head_attention(*args, **kwargs):
layer = MultiHeadAttentionQuantized(*args, **kwargs)
return wrappers.MultiHeadAttentionQuantizeWrapper(
layer, configs.DefaultMultiHeadAttentionQuantizeConfig())
def _quantized_einsum_dense(*args, **kwargs):
layer = tf_keras.layers.EinsumDense(*args, **kwargs)
return tfmot.quantization.keras.QuantizeWrapperV2(
layer, configs.DefaultEinsumDenseQuantizeConfig())
def _output_quantize(layer):
return tfmot.quantization.keras.QuantizeWrapperV2(
layer, configs.Default8BitOutputQuantizeConfig())
class TransformerEncoderBlockQuantized(tf_keras.layers.Layer):
"""TransformerEncoderBlock layer.
This layer implements the Transformer Encoder from
"Attention Is All You Need". (https://arxiv.org/abs/1706.03762),
which combines a `tf_keras.layers.MultiHeadAttention` layer with a
two-layer feedforward network.
References:
[Attention Is All You Need](https://arxiv.org/abs/1706.03762)
[BERT: Pre-training of Deep Bidirectional Transformers for Language
Understanding](https://arxiv.org/abs/1810.04805)
"""
def __init__(self,
num_attention_heads,
inner_dim,
inner_activation,
output_range=None,
kernel_initializer="glorot_uniform",
bias_initializer="zeros",
kernel_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
kernel_constraint=None,
bias_constraint=None,
use_bias=True,
norm_first=False,
norm_epsilon=1e-12,
output_dropout=0.0,
attention_dropout=0.0,
inner_dropout=0.0,
attention_initializer=None,
attention_axes=None,
**kwargs):
"""Initializes `TransformerEncoderBlock`.
Args:
num_attention_heads: Number of attention heads.
inner_dim: The output dimension of the first Dense layer in a two-layer
feedforward network.
inner_activation: The activation for the first Dense layer in a two-layer
feedforward network.
output_range: the sequence output range, [0, output_range) for slicing the
target sequence. `None` means the target sequence is not sliced.
kernel_initializer: Initializer for dense layer kernels.
bias_initializer: Initializer for dense layer biases.
kernel_regularizer: Regularizer for dense layer kernels.
bias_regularizer: Regularizer for dense layer biases.
activity_regularizer: Regularizer for dense layer activity.
kernel_constraint: Constraint for dense layer kernels.
bias_constraint: Constraint for dense layer kernels.
use_bias: Whether to enable use_bias in attention layer. If set False,
use_bias in attention layer is disabled.
norm_first: Whether to normalize inputs to attention and intermediate
dense layers. If set False, output of attention and intermediate dense
layers is normalized.
norm_epsilon: Epsilon value to initialize normalization layers.
output_dropout: Dropout probability for the post-attention and output
dropout.
attention_dropout: Dropout probability for within the attention layer.
inner_dropout: Dropout probability for the first Dense layer in a
two-layer feedforward network.
attention_initializer: Initializer for kernels of attention layers. If set
`None`, attention layers use kernel_initializer as initializer for
kernel.
attention_axes: axes over which the attention is applied. `None` means
attention over all axes, but batch, heads, and features.
**kwargs: keyword arguments/
"""
super().__init__(**kwargs)
if output_range is not None:
logging.warning("`output_range` is available as an argument for `call()`."
"The `output_range` as __init__ argument is deprecated.")
self._num_heads = num_attention_heads
self._inner_dim = inner_dim
self._inner_activation = inner_activation
self._attention_dropout = attention_dropout
self._attention_dropout_rate = attention_dropout
self._output_dropout = output_dropout
self._output_dropout_rate = output_dropout
self._output_range = output_range
self._kernel_initializer = tf_keras.initializers.get(kernel_initializer)
self._bias_initializer = tf_keras.initializers.get(bias_initializer)
self._kernel_regularizer = tf_keras.regularizers.get(kernel_regularizer)
self._bias_regularizer = tf_keras.regularizers.get(bias_regularizer)
self._activity_regularizer = tf_keras.regularizers.get(activity_regularizer)
self._kernel_constraint = tf_keras.constraints.get(kernel_constraint)
self._bias_constraint = tf_keras.constraints.get(bias_constraint)
self._use_bias = use_bias
self._norm_first = norm_first
self._norm_epsilon = norm_epsilon
self._inner_dropout = inner_dropout
if attention_initializer:
self._attention_initializer = tf_keras.initializers.get(
attention_initializer)
else:
self._attention_initializer = self._kernel_initializer
self._attention_axes = attention_axes
def build(self, input_shape):
if isinstance(input_shape, tf.TensorShape):
input_tensor_shape = input_shape
elif isinstance(input_shape, (list, tuple)):
input_tensor_shape = tf.TensorShape(input_shape[0])
else:
raise ValueError(
"The type of input shape argument is not supported, got: %s" %
type(input_shape))
if len(input_tensor_shape.as_list()) != 3:
raise ValueError("TransformerEncoderBlock expects a three-dimensional "
"input of shape [batch, sequence, width].")
hidden_size = input_tensor_shape[-1]
if hidden_size % self._num_heads != 0:
raise ValueError(
"The input size (%d) is not a multiple of the number of attention "
"heads (%d)" % (hidden_size, self._num_heads))
self._attention_head_size = int(hidden_size // self._num_heads)
common_kwargs = dict(
bias_initializer=self._bias_initializer,
kernel_regularizer=self._kernel_regularizer,
bias_regularizer=self._bias_regularizer,
activity_regularizer=self._activity_regularizer,
kernel_constraint=self._kernel_constraint,
bias_constraint=self._bias_constraint)
self._attention_layer = _quantized_multi_head_attention(
num_heads=self._num_heads,
key_dim=self._attention_head_size,
dropout=self._attention_dropout,
use_bias=self._use_bias,
kernel_initializer=self._attention_initializer,
attention_axes=self._attention_axes,
name="self_attention",
**common_kwargs)
self._attention_dropout = tf_keras.layers.Dropout(rate=self._output_dropout)
# Use float32 in layernorm for numeric stability.
# It is probably safe in mixed_float16, but we haven't validated this yet.
self._attention_layer_norm = _output_quantize(
tf_keras.layers.LayerNormalization(
name="self_attention_layer_norm",
axis=-1,
epsilon=self._norm_epsilon,
dtype=tf.float32))
self._intermediate_dense = _quantized_einsum_dense(
"abc,cd->abd",
output_shape=(None, self._inner_dim),
bias_axes="d",
kernel_initializer=self._kernel_initializer,
name="intermediate",
**common_kwargs)
policy = tf_keras.mixed_precision.global_policy()
if policy.name == "mixed_bfloat16":
# bfloat16 causes BERT with the LAMB optimizer to not converge
# as well, so we use float32.
# TODO(b/154538392): Investigate this.
policy = tf.float32
self._intermediate_activation_layer = _output_quantize(
tf_keras.layers.Activation(
self._inner_activation, dtype=policy))
self._inner_dropout_layer = tf_keras.layers.Dropout(
rate=self._inner_dropout)
self._output_dense = _quantized_einsum_dense(
"abc,cd->abd",
output_shape=(None, hidden_size),
bias_axes="d",
name="output",
kernel_initializer=self._kernel_initializer,
**common_kwargs)
self._output_dropout = tf_keras.layers.Dropout(rate=self._output_dropout)
# Use float32 in layernorm for numeric stability.
self._output_layer_norm = _output_quantize(
tf_keras.layers.LayerNormalization(
name="output_layer_norm",
axis=-1,
epsilon=self._norm_epsilon,
dtype=tf.float32))
self._add = _output_quantize(tf_keras.layers.Add())
self._output_add = tf_keras.layers.Add()
super().build(input_shape)
def get_config(self):
config = {
"num_attention_heads":
self._num_heads,
"inner_dim":
self._inner_dim,
"inner_activation":
self._inner_activation,
"output_dropout":
self._output_dropout_rate,
"attention_dropout":
self._attention_dropout_rate,
"output_range":
self._output_range,
"kernel_initializer":
tf_keras.initializers.serialize(self._kernel_initializer),
"bias_initializer":
tf_keras.initializers.serialize(self._bias_initializer),
"kernel_regularizer":
tf_keras.regularizers.serialize(self._kernel_regularizer),
"bias_regularizer":
tf_keras.regularizers.serialize(self._bias_regularizer),
"activity_regularizer":
tf_keras.regularizers.serialize(self._activity_regularizer),
"kernel_constraint":
tf_keras.constraints.serialize(self._kernel_constraint),
"bias_constraint":
tf_keras.constraints.serialize(self._bias_constraint),
"use_bias":
self._use_bias,
"norm_first":
self._norm_first,
"norm_epsilon":
self._norm_epsilon,
"inner_dropout":
self._inner_dropout,
"attention_initializer":
tf_keras.initializers.serialize(self._attention_initializer),
"attention_axes": self._attention_axes,
}
base_config = super().get_config()
return dict(list(base_config.items()) + list(config.items()))
def call(self, inputs, output_range: Optional[tf.Tensor] = None):
"""Transformer self-attention encoder block call.
Args:
inputs: a single tensor or a list of tensors. `input tensor` as the single
sequence of embeddings. [`input tensor`, `attention mask`] to have the
additional attention mask. [`query tensor`, `key value tensor`,
`attention mask`] to have separate input streams for the query, and
key/value to the multi-head attention.
output_range: the sequence output range, [0, output_range) for slicing the
target sequence. `None` means the target sequence is not sliced. If you
would like to have no change to the model training, it is better to only
set the `output_range` for serving.
Returns:
An ouput tensor with the same dimensions as input/query tensor.
"""
if isinstance(inputs, (list, tuple)):
if len(inputs) == 2:
input_tensor, attention_mask = inputs
key_value = None
elif len(inputs) == 3:
input_tensor, key_value, attention_mask = inputs
else:
raise ValueError("Unexpected inputs to %s with length at %d" %
(self.__class__, len(inputs)))
else:
input_tensor, key_value, attention_mask = (inputs, None, None)
if output_range is None:
output_range = self._output_range
if output_range:
if self._norm_first:
source_tensor = input_tensor[:, 0:output_range, :]
input_tensor = self._attention_layer_norm(input_tensor)
if key_value is not None:
key_value = self._attention_layer_norm(key_value)
target_tensor = input_tensor[:, 0:output_range, :]
if attention_mask is not None:
attention_mask = attention_mask[:, 0:output_range, :]
else:
if self._norm_first:
source_tensor = input_tensor
input_tensor = self._attention_layer_norm(input_tensor)
if key_value is not None:
key_value = self._attention_layer_norm(key_value)
target_tensor = input_tensor
if key_value is None:
key_value = input_tensor
attention_output = self._attention_layer(
query=target_tensor, value=key_value, attention_mask=attention_mask)
attention_output = self._attention_dropout(attention_output)
if self._norm_first:
attention_output = self._add([source_tensor, attention_output])
else:
attention_output = self._attention_layer_norm(
self._add([target_tensor, attention_output]))
if self._norm_first:
source_attention_output = attention_output
attention_output = self._output_layer_norm(attention_output)
inner_output = self._intermediate_dense(attention_output)
inner_output = self._intermediate_activation_layer(inner_output)
inner_output = self._inner_dropout_layer(inner_output)
layer_output = self._output_dense(inner_output)
layer_output = self._output_dropout(layer_output)
if self._norm_first:
return self._output_add([source_attention_output, layer_output])
# During mixed precision training, layer norm output is always fp32 for now.
# Casts fp32 for the subsequent add.
layer_output = tf.cast(layer_output, tf.float32)
return self._output_layer_norm(
self._output_add([layer_output, attention_output]))