lstm.py

# 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.
# ==============================================================================
"""Long Short-Term Memory layer."""


import uuid

import tensorflow.compat.v2 as tf

from keras import activations
from keras import backend
from keras import constraints
from keras import initializers
from keras import regularizers
from keras.engine import base_layer
from keras.engine.input_spec import InputSpec
from keras.layers.rnn import gru_lstm_utils
from keras.layers.rnn import rnn_utils
from keras.layers.rnn.base_rnn import RNN
from keras.layers.rnn.dropout_rnn_cell_mixin import DropoutRNNCellMixin
from keras.utils import tf_utils

# isort: off
from tensorflow.python.platform import tf_logging as logging
from tensorflow.python.util.tf_export import keras_export

RECURRENT_DROPOUT_WARNING_MSG = (
    "RNN `implementation=2` is not supported when `recurrent_dropout` is set. "
    "Using `implementation=1`."
)


@keras_export("keras.layers.LSTMCell", v1=[])
class LSTMCell(DropoutRNNCellMixin, base_layer.BaseRandomLayer):
    """Cell class for the LSTM layer.

    See [the Keras RNN API guide](https://www.tensorflow.org/guide/keras/rnn)
    for details about the usage of RNN API.

    This class processes one step within the whole time sequence input, whereas
    `tf.keras.layer.LSTM` processes the whole sequence.

    For example:

    >>> inputs = tf.random.normal([32, 10, 8])
    >>> rnn = tf.keras.layers.RNN(tf.keras.layers.LSTMCell(4))
    >>> output = rnn(inputs)
    >>> print(output.shape)
    (32, 4)
    >>> rnn = tf.keras.layers.RNN(
    ...    tf.keras.layers.LSTMCell(4),
    ...    return_sequences=True,
    ...    return_state=True)
    >>> whole_seq_output, final_memory_state, final_carry_state = rnn(inputs)
    >>> print(whole_seq_output.shape)
    (32, 10, 4)
    >>> print(final_memory_state.shape)
    (32, 4)
    >>> print(final_carry_state.shape)
    (32, 4)

    Args:
      units: Positive integer, dimensionality of the output space.
      activation: Activation function to use. Default: hyperbolic tangent
        (`tanh`). If you pass `None`, no activation is applied (ie. "linear"
        activation: `a(x) = x`).
      recurrent_activation: Activation function to use for the recurrent step.
        Default: sigmoid (`sigmoid`). If you pass `None`, no activation is
        applied (ie. "linear" activation: `a(x) = x`).
      use_bias: Boolean, (default `True`), whether the layer uses a bias vector.
      kernel_initializer: Initializer for the `kernel` weights matrix, used for
        the linear transformation of the inputs. Default: `glorot_uniform`.
      recurrent_initializer: Initializer for the `recurrent_kernel` weights
        matrix, used for the linear transformation of the recurrent state.
        Default: `orthogonal`.
      bias_initializer: Initializer for the bias vector. Default: `zeros`.
      unit_forget_bias: Boolean (default `True`). If True, add 1 to the bias of
        the forget gate at initialization. Setting it to true will also force
        `bias_initializer="zeros"`. This is recommended in [Jozefowicz et
          al.](https://github.com/mlresearch/v37/blob/gh-pages/jozefowicz15.pdf)
      kernel_regularizer: Regularizer function applied to the `kernel` weights
        matrix. Default: `None`.
      recurrent_regularizer: Regularizer function applied to
        the `recurrent_kernel` weights matrix. Default: `None`.
      bias_regularizer: Regularizer function applied to the bias vector.
        Default: `None`.
      kernel_constraint: Constraint function applied to the `kernel` weights
        matrix. Default: `None`.
      recurrent_constraint: Constraint function applied to the
        `recurrent_kernel` weights matrix. Default: `None`.
      bias_constraint: Constraint function applied to the bias vector. Default:
        `None`.
      dropout: Float between 0 and 1. Fraction of the units to drop for the
        linear transformation of the inputs. Default: 0.
      recurrent_dropout: Float between 0 and 1. Fraction of the units to drop
        for the linear transformation of the recurrent state. Default: 0.

    Call arguments:
      inputs: A 2D tensor, with shape of `[batch, feature]`.
      states: List of 2 tensors that corresponding to the cell's units. Both of
        them have shape `[batch, units]`, the first tensor is the memory state
        from previous time step, the second tensor is the carry state from
        previous time step. For timestep 0, the initial state provided by user
        will be feed to cell.
      training: Python boolean indicating whether the layer should behave in
        training mode or in inference mode. Only relevant when `dropout` or
        `recurrent_dropout` is used.
    """

    def __init__(
        self,
        units,
        activation="tanh",
        recurrent_activation="sigmoid",
        use_bias=True,
        kernel_initializer="glorot_uniform",
        recurrent_initializer="orthogonal",
        bias_initializer="zeros",
        unit_forget_bias=True,
        kernel_regularizer=None,
        recurrent_regularizer=None,
        bias_regularizer=None,
        kernel_constraint=None,
        recurrent_constraint=None,
        bias_constraint=None,
        dropout=0.0,
        recurrent_dropout=0.0,
        **kwargs,
    ):
        if units <= 0:
            raise ValueError(
                "Received an invalid value for argument `units`, "
                f"expected a positive integer, got {units}."
            )
        # By default use cached variable under v2 mode, see b/143699808.
        if tf.compat.v1.executing_eagerly_outside_functions():
            self._enable_caching_device = kwargs.pop(
                "enable_caching_device", True
            )
        else:
            self._enable_caching_device = kwargs.pop(
                "enable_caching_device", False
            )
        super().__init__(**kwargs)
        self.units = units
        self.activation = activations.get(activation)
        self.recurrent_activation = activations.get(recurrent_activation)
        self.use_bias = use_bias

        self.kernel_initializer = initializers.get(kernel_initializer)
        self.recurrent_initializer = initializers.get(recurrent_initializer)
        self.bias_initializer = initializers.get(bias_initializer)
        self.unit_forget_bias = unit_forget_bias

        self.kernel_regularizer = regularizers.get(kernel_regularizer)
        self.recurrent_regularizer = regularizers.get(recurrent_regularizer)
        self.bias_regularizer = regularizers.get(bias_regularizer)

        self.kernel_constraint = constraints.get(kernel_constraint)
        self.recurrent_constraint = constraints.get(recurrent_constraint)
        self.bias_constraint = constraints.get(bias_constraint)

        self.dropout = min(1.0, max(0.0, dropout))
        self.recurrent_dropout = min(1.0, max(0.0, recurrent_dropout))
        implementation = kwargs.pop("implementation", 2)
        if self.recurrent_dropout != 0 and implementation != 1:
            logging.debug(RECURRENT_DROPOUT_WARNING_MSG)
            self.implementation = 1
        else:
            self.implementation = implementation
        self.state_size = [self.units, self.units]
        self.output_size = self.units

    @tf_utils.shape_type_conversion
    def build(self, input_shape):
        super().build(input_shape)
        default_caching_device = rnn_utils.caching_device(self)
        input_dim = input_shape[-1]
        self.kernel = self.add_weight(
            shape=(input_dim, self.units * 4),
            name="kernel",
            initializer=self.kernel_initializer,
            regularizer=self.kernel_regularizer,
            constraint=self.kernel_constraint,
            caching_device=default_caching_device,
        )
        self.recurrent_kernel = self.add_weight(
            shape=(self.units, self.units * 4),
            name="recurrent_kernel",
            initializer=self.recurrent_initializer,
            regularizer=self.recurrent_regularizer,
            constraint=self.recurrent_constraint,
            caching_device=default_caching_device,
        )

        if self.use_bias:
            if self.unit_forget_bias:

                def bias_initializer(_, *args, **kwargs):
                    return backend.concatenate(
                        [
                            self.bias_initializer(
                                (self.units,), *args, **kwargs
                            ),
                            initializers.get("ones")(
                                (self.units,), *args, **kwargs
                            ),
                            self.bias_initializer(
                                (self.units * 2,), *args, **kwargs
                            ),
                        ]
                    )

            else:
                bias_initializer = self.bias_initializer
            self.bias = self.add_weight(
                shape=(self.units * 4,),
                name="bias",
                initializer=bias_initializer,
                regularizer=self.bias_regularizer,
                constraint=self.bias_constraint,
                caching_device=default_caching_device,
            )
        else:
            self.bias = None
        self.built = True

    def _compute_carry_and_output(self, x, h_tm1, c_tm1):
        """Computes carry and output using split kernels."""
        x_i, x_f, x_c, x_o = x
        h_tm1_i, h_tm1_f, h_tm1_c, h_tm1_o = h_tm1
        i = self.recurrent_activation(
            x_i + backend.dot(h_tm1_i, self.recurrent_kernel[:, : self.units])
        )
        f = self.recurrent_activation(
            x_f
            + backend.dot(
                h_tm1_f, self.recurrent_kernel[:, self.units : self.units * 2]
            )
        )
        c = f * c_tm1 + i * self.activation(
            x_c
            + backend.dot(
                h_tm1_c,
                self.recurrent_kernel[:, self.units * 2 : self.units * 3],
            )
        )
        o = self.recurrent_activation(
            x_o
            + backend.dot(h_tm1_o, self.recurrent_kernel[:, self.units * 3 :])
        )
        return c, o

    def _compute_carry_and_output_fused(self, z, c_tm1):
        """Computes carry and output using fused kernels."""
        z0, z1, z2, z3 = z
        i = self.recurrent_activation(z0)
        f = self.recurrent_activation(z1)
        c = f * c_tm1 + i * self.activation(z2)
        o = self.recurrent_activation(z3)
        return c, o

    def call(self, inputs, states, training=None):
        h_tm1 = states[0]  # previous memory state
        c_tm1 = states[1]  # previous carry state

        dp_mask = self.get_dropout_mask_for_cell(inputs, training, count=4)
        rec_dp_mask = self.get_recurrent_dropout_mask_for_cell(
            h_tm1, training, count=4
        )

        if self.implementation == 1:
            if 0 < self.dropout < 1.0:
                inputs_i = inputs * dp_mask[0]
                inputs_f = inputs * dp_mask[1]
                inputs_c = inputs * dp_mask[2]
                inputs_o = inputs * dp_mask[3]
            else:
                inputs_i = inputs
                inputs_f = inputs
                inputs_c = inputs
                inputs_o = inputs
            k_i, k_f, k_c, k_o = tf.split(
                self.kernel, num_or_size_splits=4, axis=1
            )
            x_i = backend.dot(inputs_i, k_i)
            x_f = backend.dot(inputs_f, k_f)
            x_c = backend.dot(inputs_c, k_c)
            x_o = backend.dot(inputs_o, k_o)
            if self.use_bias:
                b_i, b_f, b_c, b_o = tf.split(
                    self.bias, num_or_size_splits=4, axis=0
                )
                x_i = backend.bias_add(x_i, b_i)
                x_f = backend.bias_add(x_f, b_f)
                x_c = backend.bias_add(x_c, b_c)
                x_o = backend.bias_add(x_o, b_o)

            if 0 < self.recurrent_dropout < 1.0:
                h_tm1_i = h_tm1 * rec_dp_mask[0]
                h_tm1_f = h_tm1 * rec_dp_mask[1]
                h_tm1_c = h_tm1 * rec_dp_mask[2]
                h_tm1_o = h_tm1 * rec_dp_mask[3]
            else:
                h_tm1_i = h_tm1
                h_tm1_f = h_tm1
                h_tm1_c = h_tm1
                h_tm1_o = h_tm1
            x = (x_i, x_f, x_c, x_o)
            h_tm1 = (h_tm1_i, h_tm1_f, h_tm1_c, h_tm1_o)
            c, o = self._compute_carry_and_output(x, h_tm1, c_tm1)
        else:
            if 0.0 < self.dropout < 1.0:
                inputs = inputs * dp_mask[0]
            z = backend.dot(inputs, self.kernel)
            z += backend.dot(h_tm1, self.recurrent_kernel)
            if self.use_bias:
                z = backend.bias_add(z, self.bias)

            z = tf.split(z, num_or_size_splits=4, axis=1)
            c, o = self._compute_carry_and_output_fused(z, c_tm1)

        h = o * self.activation(c)
        return h, [h, c]

    def get_config(self):
        config = {
            "units": self.units,
            "activation": activations.serialize(self.activation),
            "recurrent_activation": activations.serialize(
                self.recurrent_activation
            ),
            "use_bias": self.use_bias,
            "kernel_initializer": initializers.serialize(
                self.kernel_initializer
            ),
            "recurrent_initializer": initializers.serialize(
                self.recurrent_initializer
            ),
            "bias_initializer": initializers.serialize(self.bias_initializer),
            "unit_forget_bias": self.unit_forget_bias,
            "kernel_regularizer": regularizers.serialize(
                self.kernel_regularizer
            ),
            "recurrent_regularizer": regularizers.serialize(
                self.recurrent_regularizer
            ),
            "bias_regularizer": regularizers.serialize(self.bias_regularizer),
            "kernel_constraint": constraints.serialize(self.kernel_constraint),
            "recurrent_constraint": constraints.serialize(
                self.recurrent_constraint
            ),
            "bias_constraint": constraints.serialize(self.bias_constraint),
            "dropout": self.dropout,
            "recurrent_dropout": self.recurrent_dropout,
            "implementation": self.implementation,
        }
        config.update(rnn_utils.config_for_enable_caching_device(self))
        base_config = super().get_config()
        return dict(list(base_config.items()) + list(config.items()))

    def get_initial_state(self, inputs=None, batch_size=None, dtype=None):
        return list(
            rnn_utils.generate_zero_filled_state_for_cell(
                self, inputs, batch_size, dtype
            )
        )


@keras_export("keras.layers.LSTM", v1=[])
class LSTM(DropoutRNNCellMixin, RNN, base_layer.BaseRandomLayer):
    """Long Short-Term Memory layer - Hochreiter 1997.

    See [the Keras RNN API guide](https://www.tensorflow.org/guide/keras/rnn)
    for details about the usage of RNN API.

    Based on available runtime hardware and constraints, this layer
    will choose different implementations (cuDNN-based or pure-TensorFlow)
    to maximize the performance. If a GPU is available and all
    the arguments to the layer meet the requirement of the cuDNN kernel
    (see below for details), the layer will use a fast cuDNN implementation.

    The requirements to use the cuDNN implementation are:

    1. `activation` == `tanh`
    2. `recurrent_activation` == `sigmoid`
    3. `recurrent_dropout` == 0
    4. `unroll` is `False`
    5. `use_bias` is `True`
    6. Inputs, if use masking, are strictly right-padded.
    7. Eager execution is enabled in the outermost context.

    For example:

    >>> inputs = tf.random.normal([32, 10, 8])
    >>> lstm = tf.keras.layers.LSTM(4)
    >>> output = lstm(inputs)
    >>> print(output.shape)
    (32, 4)
    >>> lstm = tf.keras.layers.LSTM(4, return_sequences=True, return_state=True)
    >>> whole_seq_output, final_memory_state, final_carry_state = lstm(inputs)
    >>> print(whole_seq_output.shape)
    (32, 10, 4)
    >>> print(final_memory_state.shape)
    (32, 4)
    >>> print(final_carry_state.shape)
    (32, 4)

    Args:
      units: Positive integer, dimensionality of the output space.
      activation: Activation function to use.
        Default: hyperbolic tangent (`tanh`). If you pass `None`, no activation
        is applied (ie. "linear" activation: `a(x) = x`).
      recurrent_activation: Activation function to use for the recurrent step.
        Default: sigmoid (`sigmoid`). If you pass `None`, no activation is
        applied (ie. "linear" activation: `a(x) = x`).
      use_bias: Boolean (default `True`), whether the layer uses a bias vector.
      kernel_initializer: Initializer for the `kernel` weights matrix, used for
        the linear transformation of the inputs. Default: `glorot_uniform`.
      recurrent_initializer: Initializer for the `recurrent_kernel` weights
        matrix, used for the linear transformation of the recurrent state.
        Default: `orthogonal`.
      bias_initializer: Initializer for the bias vector. Default: `zeros`.
      unit_forget_bias: Boolean (default `True`). If True, add 1 to the bias of
        the forget gate at initialization. Setting it to true will also force
        `bias_initializer="zeros"`. This is recommended in [Jozefowicz et
            al.](http://www.jmlr.org/proceedings/papers/v37/jozefowicz15.pdf).
      kernel_regularizer: Regularizer function applied to the `kernel` weights
        matrix. Default: `None`.
      recurrent_regularizer: Regularizer function applied to the
        `recurrent_kernel` weights matrix. Default: `None`.
      bias_regularizer: Regularizer function applied to the bias vector.
        Default: `None`.
      activity_regularizer: Regularizer function applied to the output of the
        layer (its "activation"). Default: `None`.
      kernel_constraint: Constraint function applied to the `kernel` weights
        matrix. Default: `None`.
      recurrent_constraint: Constraint function applied to the
        `recurrent_kernel` weights matrix. Default: `None`.
      bias_constraint: Constraint function applied to the bias vector. Default:
        `None`.
      dropout: Float between 0 and 1. Fraction of the units to drop for the
        linear transformation of the inputs. Default: 0.
      recurrent_dropout: Float between 0 and 1. Fraction of the units to drop
        for the linear transformation of the recurrent state. Default: 0.
      return_sequences: Boolean. Whether to return the last output in the output
        sequence, or the full sequence. Default: `False`.
      return_state: Boolean. Whether to return the last state in addition to the
        output. Default: `False`.
      go_backwards: Boolean (default `False`). If True, process the input
        sequence backwards and return the reversed sequence.
      stateful: Boolean (default `False`). If True, the last state for each
      sample at index i in a batch will be used as initial state for the sample
        of index i in the following batch.
      time_major: The shape format of the `inputs` and `outputs` tensors.
        If True, the inputs and outputs will be in shape
        `[timesteps, batch, feature]`, whereas in the False case, it will be
        `[batch, timesteps, feature]`. Using `time_major = True` is a bit more
        efficient because it avoids transposes at the beginning and end of the
        RNN calculation. However, most TensorFlow data is batch-major, so by
        default this function accepts input and emits output in batch-major
        form.
      unroll: Boolean (default `False`). If True, the network will be unrolled,
        else a symbolic loop will be used. Unrolling can speed-up a RNN,
        although it tends to be more memory-intensive. Unrolling is only
        suitable for short sequences.

    Call arguments:
      inputs: A 3D tensor with shape `[batch, timesteps, feature]`.
      mask: Binary tensor of shape `[batch, timesteps]` indicating whether
        a given timestep should be masked (optional, defaults to `None`).
        An individual `True` entry indicates that the corresponding timestep
        should be utilized, while a `False` entry indicates that the
        corresponding timestep should be ignored.
      training: Python boolean indicating whether the layer should behave in
        training mode or in inference mode. This argument is passed to the cell
        when calling it. This is only relevant if `dropout` or
        `recurrent_dropout` is used (optional, defaults to `None`).
      initial_state: List of initial state tensors to be passed to the first
        call of the cell (optional, defaults to `None` which causes creation
        of zero-filled initial state tensors).
    """

    def __init__(
        self,
        units,
        activation="tanh",
        recurrent_activation="sigmoid",
        use_bias=True,
        kernel_initializer="glorot_uniform",
        recurrent_initializer="orthogonal",
        bias_initializer="zeros",
        unit_forget_bias=True,
        kernel_regularizer=None,
        recurrent_regularizer=None,
        bias_regularizer=None,
        activity_regularizer=None,
        kernel_constraint=None,
        recurrent_constraint=None,
        bias_constraint=None,
        dropout=0.0,
        recurrent_dropout=0.0,
        return_sequences=False,
        return_state=False,
        go_backwards=False,
        stateful=False,
        time_major=False,
        unroll=False,
        **kwargs,
    ):
        # return_runtime is a flag for testing, which shows the real backend
        # implementation chosen by grappler in graph mode.
        self.return_runtime = kwargs.pop("return_runtime", False)
        implementation = kwargs.pop("implementation", 2)
        if implementation == 0:
            logging.warning(
                "`implementation=0` has been deprecated, "
                "and now defaults to `implementation=1`."
                "Please update your layer call."
            )
        if "enable_caching_device" in kwargs:
            cell_kwargs = {
                "enable_caching_device": kwargs.pop("enable_caching_device")
            }
        else:
            cell_kwargs = {}
        cell = LSTMCell(
            units,
            activation=activation,
            recurrent_activation=recurrent_activation,
            use_bias=use_bias,
            kernel_initializer=kernel_initializer,
            recurrent_initializer=recurrent_initializer,
            unit_forget_bias=unit_forget_bias,
            bias_initializer=bias_initializer,
            kernel_regularizer=kernel_regularizer,
            recurrent_regularizer=recurrent_regularizer,
            bias_regularizer=bias_regularizer,
            kernel_constraint=kernel_constraint,
            recurrent_constraint=recurrent_constraint,
            bias_constraint=bias_constraint,
            dropout=dropout,
            recurrent_dropout=recurrent_dropout,
            implementation=implementation,
            dtype=kwargs.get("dtype"),
            trainable=kwargs.get("trainable", True),
            **cell_kwargs,
        )
        super().__init__(
            cell,
            return_sequences=return_sequences,
            return_state=return_state,
            go_backwards=go_backwards,
            stateful=stateful,
            time_major=time_major,
            unroll=unroll,
            **kwargs,
        )
        self.activity_regularizer = regularizers.get(activity_regularizer)
        self.input_spec = [InputSpec(ndim=3)]
        self.state_spec = [
            InputSpec(shape=(None, dim)) for dim in (self.units, self.units)
        ]
        self._could_use_gpu_kernel = (
            self.activation in (activations.tanh, tf.tanh)
            and self.recurrent_activation in (activations.sigmoid, tf.sigmoid)
            and recurrent_dropout == 0
            and not unroll
            and use_bias
            and tf.compat.v1.executing_eagerly_outside_functions()
        )
        if tf.config.list_logical_devices("GPU"):
            # Only show the message when there is GPU available, user will not
            # care about the cuDNN if there isn't any GPU.
            if self._could_use_gpu_kernel:
                logging.debug(gru_lstm_utils.CUDNN_AVAILABLE_MSG % self.name)
            else:
                logging.warning(
                    gru_lstm_utils.CUDNN_NOT_AVAILABLE_MSG % self.name
                )

        if gru_lstm_utils.use_new_gru_lstm_impl():
            self._defun_wrapper = gru_lstm_utils.DefunWrapper(
                time_major, go_backwards, "lstm"
            )

    def call(self, inputs, mask=None, training=None, initial_state=None):
        # The input should be dense, padded with zeros. If a ragged input is fed
        # into the layer, it is padded and the row lengths are used for masking.
        inputs, row_lengths = backend.convert_inputs_if_ragged(inputs)
        is_ragged_input = row_lengths is not None
        self._validate_args_if_ragged(is_ragged_input, mask)

        # LSTM does not support constants. Ignore it during process.
        inputs, initial_state, _ = self._process_inputs(
            inputs, initial_state, None
        )

        if isinstance(mask, list):
            mask = mask[0]

        input_shape = backend.int_shape(inputs)
        timesteps = input_shape[0] if self.time_major else input_shape[1]

        if not self._could_use_gpu_kernel:
            # Fall back to use the normal LSTM.
            kwargs = {"training": training}
            self._maybe_reset_cell_dropout_mask(self.cell)

            def step(inputs, states):
                return self.cell(inputs, states, **kwargs)

            last_output, outputs, states = backend.rnn(
                step,
                inputs,
                initial_state,
                constants=None,
                go_backwards=self.go_backwards,
                mask=mask,
                unroll=self.unroll,
                input_length=row_lengths
                if row_lengths is not None
                else timesteps,
                time_major=self.time_major,
                zero_output_for_mask=self.zero_output_for_mask,
                return_all_outputs=self.return_sequences,
            )
            runtime = gru_lstm_utils.runtime(gru_lstm_utils.RUNTIME_UNKNOWN)
        else:
            # Use the new defun approach for backend implementation swap.
            # Note that different implementations need to have same function
            # signature, eg, the tensor parameters need to have same shape and
            # dtypes. Since the cuDNN has an extra set of bias, those bias will
            # be passed to both normal and cuDNN implementations.
            self.reset_dropout_mask()
            dropout_mask = self.get_dropout_mask_for_cell(
                inputs, training, count=4
            )
            if dropout_mask is not None:
                inputs = inputs * dropout_mask[0]
            if gru_lstm_utils.use_new_gru_lstm_impl():
                lstm_kwargs = {
                    "inputs": inputs,
                    "init_h": gru_lstm_utils.read_variable_value(
                        initial_state[0]
                    ),
                    "init_c": gru_lstm_utils.read_variable_value(
                        initial_state[1]
                    ),
                    "kernel": gru_lstm_utils.read_variable_value(
                        self.cell.kernel
                    ),
                    "recurrent_kernel": gru_lstm_utils.read_variable_value(
                        self.cell.recurrent_kernel
                    ),
                    "bias": gru_lstm_utils.read_variable_value(self.cell.bias),
                    "mask": mask,
                    "time_major": self.time_major,
                    "go_backwards": self.go_backwards,
                    "sequence_lengths": row_lengths,
                    "zero_output_for_mask": self.zero_output_for_mask,
                }
                (
                    last_output,
                    outputs,
                    new_h,
                    new_c,
                    runtime,
                ) = self._defun_wrapper.defun_layer(**lstm_kwargs)
            else:
                gpu_lstm_kwargs = {
                    "inputs": inputs,
                    "init_h": gru_lstm_utils.read_variable_value(
                        initial_state[0]
                    ),
                    "init_c": gru_lstm_utils.read_variable_value(
                        initial_state[1]
                    ),
                    "kernel": gru_lstm_utils.read_variable_value(
                        self.cell.kernel
                    ),
                    "recurrent_kernel": gru_lstm_utils.read_variable_value(
                        self.cell.recurrent_kernel
                    ),
                    "bias": gru_lstm_utils.read_variable_value(self.cell.bias),
                    "mask": mask,
                    "time_major": self.time_major,
                    "go_backwards": self.go_backwards,
                    "sequence_lengths": row_lengths,
                    "return_sequences": self.return_sequences,
                }
                normal_lstm_kwargs = gpu_lstm_kwargs.copy()
                normal_lstm_kwargs.update(
                    {
                        "zero_output_for_mask": self.zero_output_for_mask,
                    }
                )

                if tf.executing_eagerly():
                    device_type = gru_lstm_utils.get_context_device_type()
                    can_use_gpu = (
                        # Either user specified GPU or unspecified but GPU is
                        # available.
                        (
                            device_type == gru_lstm_utils.GPU_DEVICE_NAME
                            or (
                                device_type is None
                                and tf.config.list_logical_devices("GPU")
                            )
                        )
                        and (
                            mask is None
                            or gru_lstm_utils.is_cudnn_supported_inputs(
                                mask, self.time_major
                            )
                        )
                    )
                    # Under eager context, check the device placement and prefer
                    # the GPU implementation when GPU is available.
                    if can_use_gpu:
                        last_output, outputs, new_h, new_c, runtime = gpu_lstm(
                            **gpu_lstm_kwargs
                        )
                    else:
                        (
                            last_output,
                            outputs,
                            new_h,
                            new_c,
                            runtime,
                        ) = standard_lstm(**normal_lstm_kwargs)
                else:
                    (
                        last_output,
                        outputs,
                        new_h,
                        new_c,
                        runtime,
                    ) = lstm_with_backend_selection(**normal_lstm_kwargs)

            states = [new_h, new_c]

        if self.stateful:
            updates = [
                tf.compat.v1.assign(
                    self_state, tf.cast(state, self_state.dtype)
                )
                for self_state, state in zip(self.states, states)
            ]
            self.add_update(updates)

        if self.return_sequences:
            output = backend.maybe_convert_to_ragged(
                is_ragged_input,
                outputs,
                row_lengths,
                go_backwards=self.go_backwards,
            )
        else:
            output = last_output

        if self.return_state:
            return [output] + list(states)
        elif self.return_runtime:
            return output, runtime
        else:
            return output

    @property
    def units(self):
        return self.cell.units

    @property
    def activation(self):
        return self.cell.activation

    @property
    def recurrent_activation(self):
        return self.cell.recurrent_activation

    @property
    def use_bias(self):
        return self.cell.use_bias

    @property
    def kernel_initializer(self):
        return self.cell.kernel_initializer

    @property
    def recurrent_initializer(self):
        return self.cell.recurrent_initializer

    @property
    def bias_initializer(self):
        return self.cell.bias_initializer

    @property
    def unit_forget_bias(self):
        return self.cell.unit_forget_bias

    @property
    def kernel_regularizer(self):
        return self.cell.kernel_regularizer

    @property
    def recurrent_regularizer(self):
        return self.cell.recurrent_regularizer

    @property
    def bias_regularizer(self):
        return self.cell.bias_regularizer

    @property
    def kernel_constraint(self):
        return self.cell.kernel_constraint

    @property
    def recurrent_constraint(self):
        return self.cell.recurrent_constraint

    @property
    def bias_constraint(self):
        return self.cell.bias_constraint

    @property
    def dropout(self):
        return self.cell.dropout

    @property
    def recurrent_dropout(self):
        return self.cell.recurrent_dropout

    @property
    def implementation(self):
        return self.cell.implementation

    def get_config(self):
        config = {
            "units": self.units,
            "activation": activations.serialize(self.activation),
            "recurrent_activation": activations.serialize(
                self.recurrent_activation
            ),
            "use_bias": self.use_bias,
            "kernel_initializer": initializers.serialize(
                self.kernel_initializer
            ),
            "recurrent_initializer": initializers.serialize(
                self.recurrent_initializer
            ),
            "bias_initializer": initializers.serialize(self.bias_initializer),
            "unit_forget_bias": self.unit_forget_bias,
            "kernel_regularizer": regularizers.serialize(
                self.kernel_regularizer
            ),
            "recurrent_regularizer": regularizers.serialize(
                self.recurrent_regularizer
            ),
            "bias_regularizer": regularizers.serialize(self.bias_regularizer),
            "activity_regularizer": regularizers.serialize(
                self.activity_regularizer
            ),
            "kernel_constraint": constraints.serialize(self.kernel_constraint),
            "recurrent_constraint": constraints.serialize(
                self.recurrent_constraint
            ),
            "bias_constraint": constraints.serialize(self.bias_constraint),
            "dropout": self.dropout,
            "recurrent_dropout": self.recurrent_dropout,
            "implementation": self.implementation,
        }
        config.update(rnn_utils.config_for_enable_caching_device(self.cell))
        base_config = super().get_config()
        del base_config["cell"]
        return dict(list(base_config.items()) + list(config.items()))

    @classmethod
    def from_config(cls, config):
        if "implementation" in config and config["implementation"] == 0:
            config["implementation"] = 1
        return cls(**config)


def standard_lstm(
    inputs,
    init_h,
    init_c,
    kernel,
    recurrent_kernel,
    bias,
    mask,
    time_major,
    go_backwards,
    sequence_lengths,
    zero_output_for_mask,
    return_sequences,
):
    """LSTM with standard kernel implementation.

    This implementation can be run on all types for hardware.

    This implementation lifts out all the layer weights and make them function
    parameters. It has same number of tensor input params as the cuDNN
    counterpart. The RNN step logic has been simplified, eg dropout and mask is
    removed since cuDNN implementation does not support that.

    Note that the first half of the bias tensor should be ignored by this impl.
    The cuDNN impl need an extra set of input gate bias. In order to make the
    both function take same shape of parameter, that extra set of bias is also
    feed
    here.

    Args:
      inputs: input tensor of LSTM layer.
      init_h: initial state tensor for the cell output.
      init_c: initial state tensor for the cell hidden state.
      kernel: weights for cell kernel.
      recurrent_kernel: weights for cell recurrent kernel.
      bias: weights for cell kernel bias and recurrent bias. Only recurrent bias
        is used in this case.
      mask: Boolean tensor for mask out the steps within sequence.
        An individual `True` entry indicates that the corresponding timestep
        should be utilized, while a `False` entry indicates that the
        corresponding timestep should be ignored.
      time_major: boolean, whether the inputs are in the format of
        [time, batch, feature] or [batch, time, feature].
      go_backwards: Boolean (default False). If True, process the input sequence
        backwards and return the reversed sequence.
      sequence_lengths: The lengths of all sequences coming from a variable
        length input, such as ragged tensors. If the input has a fixed timestep
        size, this should be None.
      zero_output_for_mask: Boolean, whether to output zero for masked timestep.
      return_sequences: Boolean. If True, return the recurrent outputs for all
        timesteps in the sequence. If False, only return the output for the
        last timestep (which consumes less memory).

    Returns:
      last_output: output tensor for the last timestep, which has shape
        [batch, units].
      outputs:
        - If `return_sequences=True`: output tensor for all timesteps,
          which has shape [batch, time, units].
        - Else, a tensor equal to `last_output` with shape [batch, 1, units]
      state_0: the cell output, which has same shape as init_h.
      state_1: the cell hidden state, which has same shape as init_c.
      runtime: constant string tensor which indicate real runtime hardware. This
        value is for testing purpose and should be used by user.
    """
    input_shape = backend.int_shape(inputs)
    timesteps = input_shape[0] if time_major else input_shape[1]

    def step(cell_inputs, cell_states):
        """Step function that will be used by Keras RNN backend."""
        h_tm1 = cell_states[0]  # previous memory state
        c_tm1 = cell_states[1]  # previous carry state

        z = backend.dot(cell_inputs, kernel)
        z += backend.dot(h_tm1, recurrent_kernel)
        z = backend.bias_add(z, bias)

        z0, z1, z2, z3 = tf.split(z, 4, axis=1)

        i = tf.sigmoid(z0)
        f = tf.sigmoid(z1)
        c = f * c_tm1 + i * tf.tanh(z2)
        o = tf.sigmoid(z3)

        h = o * tf.tanh(c)
        return h, [h, c]

    last_output, outputs, new_states = backend.rnn(
        step,
        inputs,
        [init_h, init_c],
        constants=None,
        unroll=False,
        time_major=time_major,
        mask=mask,
        go_backwards=go_backwards,
        input_length=(
            sequence_lengths if sequence_lengths is not None else timesteps
        ),
        zero_output_for_mask=zero_output_for_mask,
        return_all_outputs=return_sequences,
    )
    return (
        last_output,
        outputs,
        new_states[0],
        new_states[1],
        gru_lstm_utils.runtime(gru_lstm_utils.RUNTIME_CPU),
    )


def gpu_lstm(
    inputs,
    init_h,
    init_c,
    kernel,
    recurrent_kernel,
    bias,
    mask,
    time_major,
    go_backwards,
    sequence_lengths,
    return_sequences,
):
    """LSTM with either cuDNN or ROCm implementation which is only available for
    GPU.

    Note that currently only right padded data is supported, or the result will
    be polluted by the unmasked data which should be filtered.

    Args:
      inputs: Input tensor of LSTM layer.
      init_h: Initial state tensor for the cell output.
      init_c: Initial state tensor for the cell hidden state.
      kernel: Weights for cell kernel.
      recurrent_kernel: Weights for cell recurrent kernel.
      bias: Weights for cell kernel bias and recurrent bias. Only recurrent bias
        is used in this case.
      mask: Boolean tensor for mask out the steps within sequence. An individual
        `True` entry indicates that the corresponding timestep should be
        utilized, while a `False` entry indicates that the corresponding
        timestep should be ignored.
      time_major: Boolean, whether the inputs are in the format of [time, batch,
        feature] or [batch, time, feature].
      go_backwards: Boolean (default False). If True, process the input sequence
        backwards and return the reversed sequence.
      sequence_lengths: The lengths of all sequences coming from a variable
        length input, such as ragged tensors. If the input has a fixed timestep
        size, this should be None.
      return_sequences: Boolean. If True, return the recurrent outputs for all
        timesteps in the sequence. If False, only return the output for the
        last timestep, matching the CPU function output format.

    Returns:
      last_output: Output tensor for the last timestep, which has shape
        [batch, units].
      outputs:
        - If `return_sequences=True`: output tensor for all timesteps,
          which has shape [batch, time, units].
        - Else, a tensor equal to `last_output` with shape [batch, 1, units]
      state_0: The cell output, which has same shape as init_h.
      state_1: The cell hidden state, which has same shape as init_c.
      runtime: Constant string tensor which indicate real runtime hardware. This
        value is for testing purpose and should not be used by user.
    """
    if mask is not None:
        sequence_lengths = gru_lstm_utils.calculate_sequence_by_mask(
            mask, time_major
        )

    if not time_major and sequence_lengths is None:
        inputs = tf.transpose(inputs, perm=(1, 0, 2))
        seq_axis, batch_axis = (0, 1)
    else:
        seq_axis, batch_axis = (0, 1) if time_major else (1, 0)
    # For init_h and init_c, cuDNN expects one more dim of num_layers before or
    # after batch dim for time major or batch major inputs respectively
    init_h = tf.expand_dims(init_h, axis=seq_axis)
    init_c = tf.expand_dims(init_c, axis=seq_axis)

    weights = tf.split(kernel, 4, axis=1)
    weights += tf.split(recurrent_kernel, 4, axis=1)
    # cuDNN has an extra set of bias for inputs, we disable them (setting to 0),
    # so that mathematically it is same as the canonical LSTM implementation.
    full_bias = tf.concat((tf.zeros_like(bias), bias), 0)

    if tf.sysconfig.get_build_info()["is_rocm_build"]:
        # ROCm MIOpen's weight sequence for LSTM is different from both
        # canonical and Cudnn format
        # MIOpen: [i, f, o, c] Cudnn/Canonical: [i, f, c, o]
        # i is input gate weights.
        # f is forget gate weights.
        # o is output gate weights.
        # c is cell gate weights.
        weights = [weights[x] for x in (0, 1, 3, 2, 4, 5, 7, 6)]
        # full_bias is a tensor of shape (8*n,)
        full_bias = tf.split(full_bias, 8, axis=0)
        full_bias = [full_bias[x] for x in (0, 1, 3, 2, 4, 5, 7, 6)]

    params = gru_lstm_utils.canonical_to_params(
        weights=weights,
        biases=tf.split(full_bias, 8),
        shape=tf.constant([-1]),
        transpose_weights=True,
    )

    if sequence_lengths is not None:
        if go_backwards:
            # Three reversals are required. E.g.,
            # normal input = [1, 2, 3, 0, 0]  # where 0 need to be masked
            # reversed_input_to_cudnn = [3, 2, 1, 0, 0]
            # output_from_cudnn = [6, 5, 4, 0, 0]
            # expected_output = [0, 0, 6, 5 ,4]
            inputs = tf.reverse_sequence(
                inputs,
                sequence_lengths,
                seq_axis=seq_axis,
                batch_axis=batch_axis,
            )
        outputs, h, c, _, _ = tf.raw_ops.CudnnRNNV3(
            input=inputs,
            input_h=init_h,
            input_c=init_c,
            params=params,
            is_training=True,
            rnn_mode="lstm",
            sequence_lengths=sequence_lengths,
            time_major=time_major,
        )
        if go_backwards:
            outputs = tf.reverse_sequence(
                outputs,
                sequence_lengths,
                seq_axis=seq_axis,
                batch_axis=batch_axis,
            )
            outputs = tf.reverse(outputs, axis=[seq_axis])
    else:
        # # Fill the array with shape [batch] with value of max timesteps.
        # sequence_length = array_ops.fill([array_ops.shape(inputs)[1]],
        #                                  array_ops.shape(inputs)[0])
        if go_backwards:
            # Reverse axis 0 since the input is already convert to time major.
            inputs = tf.reverse(inputs, axis=[0])
        outputs, h, c, _ = tf.raw_ops.CudnnRNN(
            input=inputs,
            input_h=init_h,
            input_c=init_c,
            params=params,
            is_training=True,
            rnn_mode="lstm",
        )

    last_output = outputs[-1]
    if not time_major and sequence_lengths is None and return_sequences:
        outputs = tf.transpose(outputs, perm=[1, 0, 2])
    h = tf.squeeze(h, axis=seq_axis)
    c = tf.squeeze(c, axis=seq_axis)

    # In the case of variable length input, the cudnn kernel will fill zeros for
    # the output, whereas the default keras behavior is to bring over the
    # previous output for t-1, so that in the return_sequence=False case, user
    # can quickly get the final effect output instead just 0s at the last
    # timestep.  In order to mimic the default keras behavior, we copy the final
    # h state as the last_output, since it is numerically same as the output.
    if sequence_lengths is not None:
        last_output = h

    # Match CPU return format
    if not return_sequences:
        outputs = tf.expand_dims(last_output, axis=0 if time_major else 1)

    return (
        last_output,
        outputs,
        h,
        c,
        gru_lstm_utils.runtime(gru_lstm_utils.RUNTIME_GPU),
    )


def lstm_with_backend_selection(
    inputs,
    init_h,
    init_c,
    kernel,
    recurrent_kernel,
    bias,
    mask,
    time_major,
    go_backwards,
    sequence_lengths,
    zero_output_for_mask,
    return_sequences,
):
    """Call the LSTM with optimized backend kernel selection.

    Under the hood, this function will create two TF function, one with the most
    generic kernel and can run on all device condition, and the second one with
    cuDNN specific kernel, which can only run on GPU.

    The first function will be called with normal_lstm_params, while the second
    function is not called, but only registered in the graph. The Grappler will
    do the proper graph rewrite and swap the optimized TF function based on the
    device placement.

    Args:
      inputs: Input tensor of LSTM layer.
      init_h: Initial state tensor for the cell output.
      init_c: Initial state tensor for the cell hidden state.
      kernel: Weights for cell kernel.
      recurrent_kernel: Weights for cell recurrent kernel.
      bias: Weights for cell kernel bias and recurrent bias. Only recurrent bias
        is used in this case.
      mask: Boolean tensor for mask out the steps within sequence.
        An individual `True` entry indicates that the corresponding timestep
        should be utilized, while a `False` entry indicates that the
        corresponding timestep should be ignored.
      time_major: Boolean, whether the inputs are in the format of
        [time, batch, feature] or [batch, time, feature].
      go_backwards: Boolean (default False). If True, process the input sequence
        backwards and return the reversed sequence.
      sequence_lengths: The lengths of all sequences coming from a variable
        length input, such as ragged tensors. If the input has a fixed timestep
        size, this should be None.
      zero_output_for_mask: Boolean, whether to output zero for masked timestep.
      return_sequences: Boolean. If True, return the recurrent outputs for all
        timesteps in the sequence. If False, only return the output for the
        last timestep (which consumes less memory).

    Returns:
      List of output tensors, same as standard_lstm.
    """
    params = {
        "inputs": inputs,
        "init_h": init_h,
        "init_c": init_c,
        "kernel": kernel,
        "recurrent_kernel": recurrent_kernel,
        "bias": bias,
        "mask": mask,
        "time_major": time_major,
        "go_backwards": go_backwards,
        "sequence_lengths": sequence_lengths,
        "zero_output_for_mask": zero_output_for_mask,
        "return_sequences": return_sequences,
    }

    def gpu_lstm_with_fallback(
        inputs,
        init_h,
        init_c,
        kernel,
        recurrent_kernel,
        bias,
        mask,
        time_major,
        go_backwards,
        sequence_lengths,
        zero_output_for_mask,
        return_sequences,
    ):
        """Use cuDNN kernel when mask is none or strictly right padded."""
        if mask is None:
            return gpu_lstm(
                inputs=inputs,
                init_h=init_h,
                init_c=init_c,
                kernel=kernel,
                recurrent_kernel=recurrent_kernel,
                bias=bias,
                mask=mask,
                time_major=time_major,
                go_backwards=go_backwards,
                sequence_lengths=sequence_lengths,
                return_sequences=return_sequences,
            )

        def cudnn_lstm_fn():
            return gpu_lstm(
                inputs=inputs,
                init_h=init_h,
                init_c=init_c,
                kernel=kernel,
                recurrent_kernel=recurrent_kernel,
                bias=bias,
                mask=mask,
                time_major=time_major,
                go_backwards=go_backwards,
                sequence_lengths=sequence_lengths,
                return_sequences=return_sequences,
            )

        def stardard_lstm_fn():
            return standard_lstm(
                inputs=inputs,
                init_h=init_h,
                init_c=init_c,
                kernel=kernel,
                recurrent_kernel=recurrent_kernel,
                bias=bias,
                mask=mask,
                time_major=time_major,
                go_backwards=go_backwards,
                sequence_lengths=sequence_lengths,
                zero_output_for_mask=zero_output_for_mask,
                return_sequences=return_sequences,
            )

        return tf.cond(
            gru_lstm_utils.is_cudnn_supported_inputs(mask, time_major),
            true_fn=cudnn_lstm_fn,
            false_fn=stardard_lstm_fn,
        )

    if gru_lstm_utils.use_new_gru_lstm_impl():
        # Chooses the implementation dynamically based on the running device.
        (
            last_output,
            outputs,
            new_h,
            new_c,
            runtime,
        ) = tf.__internal__.execute_fn_for_device(
            {
                gru_lstm_utils.CPU_DEVICE_NAME: lambda: standard_lstm(**params),
                gru_lstm_utils.GPU_DEVICE_NAME: lambda: gpu_lstm_with_fallback(
                    **params
                ),
            },
            lambda: standard_lstm(**params),
        )
    else:
        # Each time a `tf.function` is called, we will give it a unique
        # identifiable API name, so that Grappler won't get confused when it
        # sees multiple LSTM layers added into same graph, and it will be able
        # to pair up the different implementations across them.
        api_name = "lstm_" + str(uuid.uuid4())
        supportive_attribute = {
            "time_major": time_major,
            "go_backwards": go_backwards,
        }
        defun_standard_lstm = gru_lstm_utils.generate_defun_backend(
            api_name,
            gru_lstm_utils.CPU_DEVICE_NAME,
            standard_lstm,
            supportive_attribute,
        )
        defun_gpu_lstm = gru_lstm_utils.generate_defun_backend(
            api_name,
            gru_lstm_utils.GPU_DEVICE_NAME,
            gpu_lstm_with_fallback,
            supportive_attribute,
        )

        # Call the normal LSTM impl and register the cuDNN impl function. The
        # grappler will kick in during session execution to optimize the graph.
        last_output, outputs, new_h, new_c, runtime = defun_standard_lstm(
            **params
        )
        gru_lstm_utils.function_register(defun_gpu_lstm, **params)

    return last_output, outputs, new_h, new_c, runtime