xref: /aosp_15_r20/external/tensorflow/tensorflow/python/tpu/tpu_embedding_v2.py (revision b6fb3261f9314811a0f4371741dbb8839866f948)

# Copyright 2020 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.
# ==============================================================================
"""Mid level API for TPU Embeddings."""

import functools
from typing import Any, Callable, Dict, Iterable, List, Optional, Text, Tuple, Union

from absl import logging

from tensorflow.core.framework import attr_value_pb2
from tensorflow.core.protobuf.tpu import tpu_embedding_configuration_pb2
from tensorflow.python.distribute import device_util
from tensorflow.python.distribute import distribute_utils
from tensorflow.python.distribute import distribution_strategy_context
from tensorflow.python.distribute import sharded_variable
from tensorflow.python.distribute import tpu_strategy
from tensorflow.python.eager import context
from tensorflow.python.eager import def_function
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import device as tf_device
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import sparse_tensor
from tensorflow.python.framework.tensor_shape import TensorShape
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import variable_scope
from tensorflow.python.ops import variables as tf_variables
from tensorflow.python.ops.ragged import ragged_tensor
from tensorflow.python.saved_model import registration
from tensorflow.python.saved_model import save_context
from tensorflow.python.tpu import tpu
from tensorflow.python.tpu import tpu_embedding_v2_utils
from tensorflow.python.tpu.ops import tpu_ops
from tensorflow.python.trackable import autotrackable
from tensorflow.python.trackable import base
from tensorflow.python.types import internal as internal_types
from tensorflow.python.util import compat
from tensorflow.python.util import nest
from tensorflow.python.util import tf_inspect
from tensorflow.python.util.tf_export import tf_export


_HOOK_KEY = "TPUEmbedding_saveable"
_NAME_KEY = "_tpu_embedding_layer"


class TPUEmbeddingVariable(sharded_variable.ShardedVariableMixin):
  """A ShardedVariable class for TPU."""

  @property
  def _in_graph_mode(self):
    return self.variables[0]._in_graph_mode  # pylint: disable=protected-access


def _add_key_attr(op, name):
  op._set_attr(_NAME_KEY, attr_value_pb2.AttrValue(s=compat.as_bytes(name)))  # pylint: disable=protected-access


@tf_export("tpu.experimental.embedding.TPUEmbedding")
class TPUEmbedding(autotrackable.AutoTrackable):
  """The TPUEmbedding mid level API.

  NOTE: When instantiated under a TPUStrategy, this class can only be created
  once per call to `tf.tpu.experimental.initialize_tpu_system`. If you wish to
  re-initialize the embedding engine you must re-initialize the tpu as well.
  Doing this will clear any variables from TPU, so ensure you have checkpointed
  before you do this. If a further instances of the class are needed,
  set the `initialize_tpu_embedding` argument to `False`.

  This class can be used to support training large embeddings on TPU. When
  creating an instance of this class, you must specify the complete set of
  tables and features you expect to lookup in those tables. See the
  documentation of `tf.tpu.experimental.embedding.TableConfig` and
  `tf.tpu.experimental.embedding.FeatureConfig` for more details on the complete
  set of options. We will cover the basic usage here.

  NOTE: multiple `FeatureConfig` objects can use the same `TableConfig` object,
  allowing different features to share the same table:

  ```python
  table_config_one = tf.tpu.experimental.embedding.TableConfig(
      vocabulary_size=...,
      dim=...)
  table_config_two = tf.tpu.experimental.embedding.TableConfig(
      vocabulary_size=...,
      dim=...)
  feature_config = {
      'feature_one': tf.tpu.experimental.embedding.FeatureConfig(
          table=table_config_one),
      'feature_two': tf.tpu.experimental.embedding.FeatureConfig(
          table=table_config_one),
      'feature_three': tf.tpu.experimental.embedding.FeatureConfig(
          table=table_config_two)}
  ```

  There are two modes under which the `TPUEmbedding` class can used. This
  depends on if the class was created under a `TPUStrategy` scope or not.

  Under `TPUStrategy`, we allow access to the method `enqueue`, `dequeue` and
  `apply_gradients`. We will show examples below of how to use these to train
  and evaluate your model. Under CPU, we only access to the `embedding_tables`
  property which allow access to the embedding tables so that you can use them
  to run model evaluation/prediction on CPU.

  First lets look at the `TPUStrategy` mode. Initial setup looks like:

  ```python
  strategy = tf.distribute.TPUStrategy(...)
  with strategy.scope():
    embedding = tf.tpu.experimental.embedding.TPUEmbedding(
        feature_config=feature_config,
        optimizer=tf.tpu.experimental.embedding.SGD(0.1))
  ```

  When creating a distributed dataset that is to be passed to the enqueue
  operation a special input option must be specified:

  ```python
  distributed_dataset = (
      strategy.distribute_datasets_from_function(
          dataset_fn=...,
          options=tf.distribute.InputOptions(
              experimental_fetch_to_device=False))
  dataset_iterator = iter(distributed_dataset)
  ```

  Different feature inputs can have different shapes. For dense and sparse
  tensor, rank 2 and above is supported. For ragged tensor, although only rank 2
  is supported, you can specify the output shape to be rank 2 and above. The
  output shape specified in the FeatureConfig has the first priority. The input
  shape passed in build method has second priority and the input shapes
  auto detected from input feature has the lowest priority. The latter two will
  be converted to output shapes by omitting the last dimension. If the lower
  priority one has output shapes which don't match the former one. A ValueError
  will be raised. Only when the former one has undefined output shapes, the
  latter one can override.

  NOTE: All batches passed to the layer can have different input shapes. But
  these input shapes need to match with the output shapes set by either
  `FeatureConfig` or build method except for ragged tensor. Only 2D
  ragged tensor with output shape set to higher dimensions is allowed as
  long as the total number of elements matches. All subsequent calls must have
  the same input shapes. In the event that the input shapes cannot be
  automatically determined by the enqueue method, you must call
  the build method with the input shapes or provide output shapes in the
  `FeatureConfig` to initialize the layer.

  To use this API on TPU you should use a custom training loop. Below is an
  example of a training and evaluation step:

  ```python
  @tf.function
  def training_step(dataset_iterator, num_steps):
    def tpu_step(tpu_features):
      with tf.GradientTape() as tape:
        activations = embedding.dequeue()
        tape.watch(activations)
        model_output = model(activations)
        loss = ...  # some function of labels and model_output

      embedding_gradients = tape.gradient(loss, activations)
      embedding.apply_gradients(embedding_gradients)
      # Insert your model gradient and optimizer application here

    for _ in tf.range(num_steps):
      embedding_features, tpu_features = next(dataset_iterator)
      embedding.enqueue(embedding_features, training=True)
      strategy.run(tpu_step, args=(tpu_features, ))

  @tf.function
  def evalution_step(dataset_iterator, num_steps):
    def tpu_step(tpu_features):
      activations = embedding.dequeue()
      model_output = model(activations)
      # Insert your evaluation code here.

    for _ in tf.range(num_steps):
      embedding_features, tpu_features = next(dataset_iterator)
      embedding.enqueue(embedding_features, training=False)
      strategy.run(tpu_step, args=(tpu_features, ))
  ```

  NOTE: The calls to `enqueue` have `training` set to `True` when
  `embedding.apply_gradients` is used and set to `False` when
  `embedding.apply_gradients` is not present in the function. If you don't
  follow this pattern you may cause an error to be raised or the tpu may
  deadlock.

  In the above examples, we assume that the user has a dataset which returns
  a tuple where the first element of the tuple matches the structure of what
  was passed as the `feature_config` argument to the object initializer. Also we
  utilize `tf.range` to get a `tf.while_loop` in order to increase performance.

  When checkpointing your model, you should include your
  `tf.tpu.experimental.embedding.TPUEmbedding` object in the checkpoint. It is a
  trackable object and saving it will save the embedding tables and their
  optimizer slot variables:

  ```python
  checkpoint = tf.train.Checkpoint(model=model, embedding=embedding)
  checkpoint.save(...)
  ```

  On CPU, only the `embedding_table` property is usable. This will allow you to
  restore a checkpoint to the object and have access to the table variables:

  ```python
  model = model_fn(...)
  embedding = tf.tpu.experimental.embedding.TPUEmbedding(
      feature_config=feature_config,
      optimizer=tf.tpu.experimental.embedding.SGD(0.1))
  checkpoint = tf.train.Checkpoint(model=model, embedding=embedding)
  checkpoint.restore(...)

  tables = embedding.embedding_tables
  ```

  You can now use table in functions like `tf.nn.embedding_lookup` to perform
  your embedding lookup and pass to your model.

  """

  def __init__(
      self,
      feature_config: Union[tpu_embedding_v2_utils.FeatureConfig, Iterable],  # pylint:disable=g-bare-generic
      optimizer: Optional[tpu_embedding_v2_utils._Optimizer],  # pylint:disable=protected-access
      pipeline_execution_with_tensor_core: bool = False):
    """Creates the TPUEmbedding mid level API object.

    ```python
    strategy = tf.distribute.TPUStrategy(...)
    with strategy.scope():
      embedding = tf.tpu.experimental.embedding.TPUEmbedding(
          feature_config=tf.tpu.experimental.embedding.FeatureConfig(
              table=tf.tpu.experimental.embedding.TableConfig(
                  dim=...,
                  vocabulary_size=...)))
    ```

    Args:
      feature_config: A nested structure of
        `tf.tpu.experimental.embedding.FeatureConfig` configs.
      optimizer: An instance of one of `tf.tpu.experimental.embedding.SGD`,
        `tf.tpu.experimental.embedding.Adagrad` or
        `tf.tpu.experimental.embedding.Adam`. When not created under
        TPUStrategy may be set to None to avoid the creation of the optimizer
        slot variables, useful for optimizing memory consumption when exporting
        the model for serving where slot variables aren't needed.
      pipeline_execution_with_tensor_core: If True, the TPU embedding
        computations will overlap with the TensorCore computations (and hence
        will be one step old). Set to True for improved performance.

    Raises:
      ValueError: If optimizer is not one of tf.tpu.experimental.embedding.(SGD,
      Adam or Adagrad) or None when created under a TPUStrategy.
    """
    self._strategy = distribution_strategy_context.get_strategy()
    self._using_tpu = isinstance(self._strategy, (tpu_strategy.TPUStrategy,
                                                  tpu_strategy.TPUStrategyV2))
    self._pipeline_execution_with_tensor_core = (
        pipeline_execution_with_tensor_core)

    self._feature_config = feature_config
    self._output_shapes = []
    for feature in nest.flatten(feature_config):
      self._output_shapes.append(feature.output_shape)

    # The TPU embedding ops are slightly inconsistent with how they refer to
    # tables:
    # * The enqueue op takes a parallel list of tensors for input, one of those
    #   is the table id for the feature which matches the integer index of the
    #   table in the proto created by _create_config_proto().
    # * The recv_tpu_embedding_activations op emits lookups per table in the
    #   order from the config proto.
    # * The send_tpu_embedding_gradients expects input tensors to be per table
    #   in the same order as the config proto.
    # * Per optimizer load and retrieve ops are specified per table and take the
    #   table name rather than the table id.
    # Thus we must fix a common order to tables and ensure they have unique
    # names.

    # Set table order here to the order of the first occurence of the table in a
    # feature provided by the user. The order of this struct must be fixed
    # to provide the user with deterministic behavior over multiple
    # instantiations.
    self._table_config = []
    for feature in nest.flatten(feature_config):
      if feature.table not in self._table_config:
        self._table_config.append(feature.table)

    # Ensure tables have unique names. Also error check the optimizer as we
    # specifically don't do that in the TableConfig class to allow high level
    # APIs that are built on this to use strings/other classes to represent
    # optimizers (before they are passed to this class).
    table_names = []
    for i, table in enumerate(self._table_config):
      if table.optimizer is None:
        # TODO(bfontain) Should we allow some sort of optimizer merging here?
        table.optimizer = optimizer
      if ((table.optimizer is not None or self._using_tpu) and
          not isinstance(table.optimizer, tpu_embedding_v2_utils._Optimizer)):  # pylint: disable=protected-access
        raise ValueError("{} is an unsupported optimizer class. Please pass an "
                         "instance of one of the optimizer classes under "
                         "tf.tpu.experimental.embedding.".format(
                             type(table.optimizer)))
      if table.name is None:
        table.name = "table_{}".format(i)
      if table.name in table_names:
        raise ValueError("Tables must have a unique name. "
                         f"Multiple tables with name {table.name} found.")
      table_names.append(table.name)

    if self._using_tpu:
      # Extract a list of callable learning rates also in fixed order. Each
      # table in the confix proto will get a index into this list and we will
      # pass this list in the same order after evaluation to the
      # send_tpu_embedding_gradients op.
      self._dynamic_learning_rates = list({
          table.optimizer.learning_rate for table in self._table_config if
          callable(table.optimizer.learning_rate)})

      # We need to list of host devices for the load/retrieve operations.
      self._hosts = get_list_of_hosts(self._strategy)

    self._built = False
    self._verify_output_shapes_on_enqueue = True

  def build(self, per_replica_input_shapes=None, per_replica_batch_size=None):  # pylint:disable=g-bare-generic
    """Create the underlying variables and initializes the TPU for embeddings.

    This method creates the underlying variables (including slot variables). If
    created under a TPUStrategy, this will also initialize the TPU for
    embeddings.

    This function will automatically get called by enqueue, which will try to
    determine your output shapes. If this fails, you must manually
    call this method before you call enqueue.

    Args:
      per_replica_input_shapes: A nested structure of The per replica input
        shapes that matches the structure of the feature config. The input
        shapes should be the same as the input shape of the feature (except for
        ragged tensor) Note that it is fixed and the same per replica input
        shapes must be used for both training and evaluation. If you want to
        calculate this from the global input shapes, you can use
        `num_replicas_in_sync` property of your strategy object. May be set to
        None if not created under a TPUStrategy.
      per_replica_batch_size: (Deprecated) The per replica batch size that you
        intend to use. Note that is fixed and the same batch size must be used
        for both training and evaluation. If you want to calculate this from the
        global batch size, you can use `num_replicas_in_sync` property of your
        strategy object. May be set to None if not created under a TPUStrategy.

    Raises:
      ValueError: If per_replica_input_shapes is inconsistent with the output
      shapes stored in the feature config or the output shapes get from the
      input shapes are not fully defined.
      RuntimeError: If tpu embedding is already initialized on TPU.
    """
    if self._built:
      return

    if self._using_tpu:
      # If the tpu embedding is already initialized on TPU, raise runtime error.
      # Below logic is not added in `initialize_system_for_tpu_embedding`
      # because doing exception control flow in graph mode is difficult.
      if tpu_ops.is_tpu_embedding_initialized():
        raise RuntimeError(
            "TPU is already initialized for embeddings. This may be caused by "
            "using multiple TPUEmbedding instances in a TPU scope which is "
            "unsupported")
      self._get_and_update_output_shapes_from_input(per_replica_input_shapes,
                                                    per_replica_batch_size)

      self._config_proto = self._create_config_proto()

      logging.info("Initializing TPU Embedding engine.")
      tpu_embedding_v2_utils.log_tpu_embedding_configuration(self._config_proto)

      @def_function.function
      def load_config():
        tpu.initialize_system_for_tpu_embedding(self._config_proto)

      load_config()
      logging.info("Done initializing TPU Embedding engine.")

    # Create and load variables and slot variables into the TPU.
    # Note that this is a dict of dicts. Keys to the first dict are table names.
    # We would prefer to use TableConfigs, but then these variables won't be
    # properly tracked by the tracking API.
    self._variables = self._create_variables_and_slots()

    self._built = True

    # This is internally conditioned self._built and self._using_tpu
    self._load_variables()

  def _maybe_build(self,
                   output_shapes: Optional[Union[List[int], Iterable]] = None):  # pylint:disable=g-bare-generic
    if not self._built:
      # This can be called while tracing a function, so we wrap the
      # initialization code with init_scope so it runs eagerly, this means that
      # it will not be included the function graph generated by tracing so that
      # we can be sure that we only initialize the TPU for embeddings exactly
      # once.
      with ops.init_scope():
        self.build(output_shapes)

  def _get_and_update_output_shapes_from_input(
      self,
      per_replica_input_shapes: Optional[List[TensorShape]] = None,
      per_replica_batch_size: Optional[int] = None):
    """Get and update the per replica output shapes from the input."""
    per_replica_output_shapes = None
    if per_replica_batch_size and per_replica_input_shapes is None:
      logging.warning(
          "per_replica_batch_size argument will be deprecated, please specify "
          "all the input shapes using per_replica_input_shapes argument.")
      per_replica_output_shapes = self._get_output_shapes_from_batch_size(
          per_replica_batch_size)

    # Update the input shapes if provided.
    if per_replica_input_shapes is not None:
      if isinstance(per_replica_input_shapes, int):
        logging.warning(
            "Passing batch size to per_replica_input_shapes argument will be"
            " deprecated, please specify all the input shapes using"
            " per_replica_input_shapes argument.")
        per_replica_output_shapes = self._get_output_shapes_from_batch_size(
            per_replica_input_shapes)
      else:
        nest.assert_same_structure(
            nest.flatten(per_replica_input_shapes),
            nest.flatten(self._feature_config))

        # Convert the nested structure to list.
        per_replica_input_shapes = nest.flatten(per_replica_input_shapes)

        per_replica_output_shapes = self._get_output_shapes_from_input_shapes(
            per_replica_input_shapes)

    if per_replica_output_shapes is not None:

      # Check the output shapes with existing output shapes setting.
      self._check_output_shapes(per_replica_output_shapes)

      # Update the output shapes with existing output shapes setting.
      # This is necessary Because the output shapes might be missing from
      # the feature config, the usr can set it:
      #  1. calling the build method
      #  2. output shapes auto detected when calling the dequeue method for
      #     for the first time. The dequeue method will call build method
      #     with the output shapes.
      # Either these two situations will lead to an update to the existing
      # output shapes.
      self._update_output_shapes(per_replica_output_shapes)

    # Check if the output shapes are fully defined. This is required in order
    # to set them in the feature descriptor field of the tpu embedding config
    # proto.
    self._check_output_shapes_fully_defined()

  def _get_output_shapes_from_input_shapes(
      self, input_shapes: List[TensorShape]) -> List[TensorShape]:
    """Get output shapes from the flattened input shapes list."""
    output_shapes = []
    for input_shape, feature in zip(input_shapes,
                                    nest.flatten(self._feature_config)):
      if input_shape.rank is None or input_shape.rank < 1:
        raise ValueError(
            "Received input tensor of shape {}. Rank must be 1 and above"
            .format(input_shape))
      # Update the input shape with the max sequence length. Only update when
      # 1. Input feature is 2D ragged or sparse tensor.
      # 2. Output shape is not set in the feature config and the max sequence
      #    length is set.
      if (len(input_shape) == 2 and input_shape[-1] != 1 and
          not feature.output_shape and feature.max_sequence_length > 0):
        input_shape_list = input_shape.as_list()
        input_shape_list.insert(
            len(input_shape_list) - 1, feature.max_sequence_length)
        input_shape = TensorShape(input_shape_list)
      if input_shape.rank == 1:
        output_shapes.append(input_shape)
      else:
        output_shapes.append(input_shape[:-1])
    return output_shapes

  @property
  def embedding_tables(
      self
  ) -> Dict[tpu_embedding_v2_utils.TableConfig, tf_variables.Variable]:
    """Returns a dict of embedding tables, keyed by `TableConfig`.

    This property only works when the `TPUEmbedding` object is created under a
    non-TPU strategy. This is intended to be used to for CPU based lookup when
    creating a serving checkpoint.

    Returns:
      A dict of embedding tables, keyed by `TableConfig`.

    Raises:
      RuntimeError: If object was created under a `TPUStrategy`.
    """
    # We don't support returning tables on TPU due to their sharded nature and
    # the fact that when using a TPUStrategy:
    # 1. Variables are stale and are only updated when a checkpoint is made.
    # 2. Updating the variables won't affect the actual tables on the TPU.
    if self._using_tpu:
      if save_context.in_save_context():
        return {table: self._variables[table.name]["parameters"].variables[0]
                for table in self._table_config}
      raise RuntimeError("Unable to retrieve embedding tables when using a TPU "
                         "strategy. If you need access, save your model, "
                         "create this object under a CPU strategy and restore.")

    self._maybe_build(None)

    # Only return the tables and not the slot variables. On CPU this are honest
    # tf.Variables.
    return {table: self._variables[table.name]["parameters"]
            for table in self._table_config}

  def _create_config_proto(
      self
  ) -> tpu_embedding_configuration_pb2.TPUEmbeddingConfiguration:
    """Creates the TPUEmbeddingConfiguration proto.

    This proto is used to initialize the TPU embedding engine.

    Returns:
      A TPUEmbeddingConfiguration proto.
    """

    config_proto = tpu_embedding_configuration_pb2.TPUEmbeddingConfiguration()

    # Map each callable dynamic learning rate to its in index in the list.
    # The learning rate index is the index of the dynamic learning rate for this
    # table (if it exists) in the list we created at initialization. We don't
    # simply create one learning rate index per table as this has extremely bad
    # performance characteristics. The more separate optimization configurations
    # we have, the worse the performance will be.
    learning_rate_index = {r: i for i, r in enumerate(
        self._dynamic_learning_rates)}

    for table in self._table_config:
      table_descriptor = config_proto.table_descriptor.add()
      table_descriptor.name = table.name

      # For small tables, we pad to the number of hosts so that at least one
      # id will be assigned to each host.
      table_descriptor.vocabulary_size = max(table.vocabulary_size,
                                             self._strategy.extended.num_hosts)
      table_descriptor.dimension = table.dim

      parameters = table_descriptor.optimization_parameters

      # We handle the learning rate separately here and don't allow the
      # optimization class to handle this, as it doesn't know about dynamic
      # rates.
      if callable(table.optimizer.learning_rate):
        parameters.learning_rate.dynamic.tag = (
            learning_rate_index[table.optimizer.learning_rate])
      else:
        parameters.learning_rate.constant = table.optimizer.learning_rate

      # Use optimizer to handle the rest of the parameters.
      table.optimizer._set_optimization_parameters(parameters)  # pylint: disable=protected-access

    table_to_id = {table: i for i, table in enumerate(self._table_config)}

    # Set feature descriptor field in the config proto.
    for feature, output_shape in zip(
        nest.flatten(self._feature_config), self._output_shapes):
      feature_descriptor = config_proto.feature_descriptor.add()

      if feature.name:
        feature_descriptor.name = feature.name

      feature_descriptor.table_id = table_to_id[feature.table]
      # The input shape of the feature is the actual shape of the input tensor
      # except the last dimension because the last dimension will always be
      # reduced.
      feature_descriptor.input_shape.extend(output_shape.as_list())

    # Always set mode to training, we override the mode during enqueue.
    config_proto.mode = (
        tpu_embedding_configuration_pb2.TPUEmbeddingConfiguration.TRAINING)

    config_proto.num_hosts = self._strategy.extended.num_hosts
    config_proto.num_tensor_cores = self._strategy.num_replicas_in_sync

    # TODO(bfontain): Allow users to pick MOD for the host sharding.
    config_proto.sharding_strategy = (
        tpu_embedding_configuration_pb2.TPUEmbeddingConfiguration.DIV_DEFAULT)
    config_proto.pipeline_execution_with_tensor_core = (
        self._pipeline_execution_with_tensor_core)

    return config_proto

  def apply_gradients(self, gradients, name: Optional[Text] = None):
    """Applies the gradient update to the embedding tables.

    If a gradient of `None` is passed in any position of the nested structure,
    then an gradient update with a zero gradient is applied for that feature.
    For optimizers like SGD or Adagrad, this is the same as applying no update
    at all. For lazy Adam and other sparsely applied optimizers with decay,
    ensure you understand the effect of applying a zero gradient.

    ```python
    strategy = tf.distribute.TPUStrategy(...)
    with strategy.scope():
      embedding = tf.tpu.experimental.embedding.TPUEmbedding(...)

    distributed_dataset = (
        strategy.distribute_datasets_from_function(
            dataset_fn=...,
            options=tf.distribute.InputOptions(
                experimental_fetch_to_device=False))
    dataset_iterator = iter(distributed_dataset)

    @tf.function
    def training_step():
      def tpu_step(tpu_features):
        with tf.GradientTape() as tape:
          activations = embedding.dequeue()
          tape.watch(activations)

          loss = ... #  some computation involving activations

        embedding_gradients = tape.gradient(loss, activations)
        embedding.apply_gradients(embedding_gradients)

      embedding_features, tpu_features = next(dataset_iterator)
      embedding.enqueue(embedding_features, training=True)
      strategy.run(tpu_step, args=(tpu_features, ))

    training_step()
    ```

    Args:
      gradients: A nested structure of gradients, with structure matching the
        `feature_config` passed to this object.
      name: A name for the underlying op.

    Raises:
      RuntimeError: If called when object wasn't created under a `TPUStrategy`
        or if not built (either by manually calling build or calling enqueue).
      ValueError: If a non-`tf.Tensor` non-`None` gradient is passed in, or a
        `tf.Tensor` of the incorrect shape is passed in. Also if
        the size of any sequence in `gradients` does not match corresponding
        sequence in `feature_config`.
      TypeError: If the type of any sequence in `gradients` does not match
        corresponding sequence in `feature_config`.
    """
    if not self._using_tpu:
      raise RuntimeError("apply_gradients is not valid when TPUEmbedding "
                         "object is not created under a TPUStrategy.")

    if not self._built:
      raise RuntimeError("apply_gradients called on unbuilt TPUEmbedding "
                         "object. Please either call enqueue first or manually "
                         "call the build method.")

    nest.assert_same_structure(self._feature_config, gradients)
    updated_gradients = []
    for (path, gradient), feature, output_shape in zip(
        nest.flatten_with_joined_string_paths(gradients),
        nest.flatten(self._feature_config), self._output_shapes):
      full_output_shape = list(output_shape) + [feature.table.dim]
      if gradient is not None and not isinstance(gradient, ops.Tensor):
        raise ValueError(
            f"found non-tensor type: {type(gradient)} at path {path}.")
      if gradient is not None:
        if gradient.shape != full_output_shape:
          raise ValueError("Found gradient of shape {} at path {}. Expected "
                           "shape {}.".format(gradient.shape, path,
                                              full_output_shape))
      else:
        # No gradient for this feature, since we must give a gradient for all
        # features, pass in a zero tensor here. Note that this is not correct
        # for all optimizers.
        logging.warning(
            "No gradient passed for feature %s, sending zero "
            "gradient. This may not be correct behavior for certain "
            "optimizers like Adam.", path)
        gradient = array_ops.zeros(full_output_shape, dtype=dtypes.float32)
      # Some gradients can be passed with op which shape is not correctly set.
      # This ensures that the shape of the gradient is correctly set.
      updated_gradients.append(
          array_ops.reshape(gradient, shape=gradient.shape))
    op = tpu_ops.send_tpu_embedding_gradients(
        inputs=updated_gradients,
        learning_rates=[
            math_ops.cast(fn(), dtype=dtypes.float32)
            for fn in self._dynamic_learning_rates
        ],
        config=self._config_proto.SerializeToString())

    # Apply the name tag to the op.
    if name is not None:
      _add_key_attr(op, name)

  def dequeue(self, name: Optional[Text] = None):
    """Get the embedding results.

    Returns a nested structure of `tf.Tensor` objects, matching the structure of
    the `feature_config` argument to the `TPUEmbedding` class. The output shape
    of the tensors is `(*output_shape, dim)`, `dim` is the dimension of the
    corresponding `TableConfig`. For output_shape, there are three places where
    it can be set.
      1. FeatureConfig provided in the __init__ function.
      2. Per_replica_output_shapes by directly calling the build method
           after initializing the tpu embedding class.
      3. Auto detected from the shapes of the input feature.
    The priority of these places is the exact same order.

    ```python
    strategy = tf.distribute.TPUStrategy(...)
    with strategy.scope():
      embedding = tf.tpu.experimental.embedding.TPUEmbedding(...)

    distributed_dataset = (
        strategy.distribute_datasets_from_function(
            dataset_fn=...,
            options=tf.distribute.InputOptions(
                experimental_fetch_to_device=False))
    dataset_iterator = iter(distributed_dataset)

    @tf.function
    def training_step():
      def tpu_step(tpu_features):
        with tf.GradientTape() as tape:
          activations = embedding.dequeue()
          tape.watch(activations)

          loss = ... #  some computation involving activations

        embedding_gradients = tape.gradient(loss, activations)
        embedding.apply_gradients(embedding_gradients)

      embedding_features, tpu_features = next(dataset_iterator)
      embedding.enqueue(embedding_features, training=True)
      strategy.run(tpu_step, args=(tpu_features, ))

    training_step()
    ```

    Args:
      name: A name for the underlying op.

    Returns:
      A nested structure of tensors, with the same structure as `feature_config`
    passed to this instance of the `TPUEmbedding` object.

    Raises:
      RuntimeError: If called when object wasn't created under a `TPUStrategy`
        or if not built (either by manually calling build or calling enqueue).
    """
    if not self._using_tpu:
      raise RuntimeError("dequeue is not valid when TPUEmbedding object is not "
                         "created under a TPUStrategy.")

    if not self._built:
      raise RuntimeError("dequeue called on unbuilt TPUEmbedding object. "
                         "Please either call enqueue first or manually call "
                         "the build method.")

    # The activations returned by this op are per feature.
    activations = tpu_ops.recv_tpu_embedding_activations(
        num_outputs=len(self._config_proto.feature_descriptor),
        config=self._config_proto.SerializeToString())

    # Apply the name tag to the op.
    if name is not None:
      _add_key_attr(activations[0].op, name)

    # Pack the list back into the same nested structure as the features.
    return nest.pack_sequence_as(self._feature_config, activations)

  def _create_variables_and_slots(
      self
  ) -> Dict[Text, Dict[Text, tf_variables.Variable]]:
    """Create variables for TPU embeddings.

    Note under TPUStrategy this will ensure that all creations happen within a
    variable creation scope of the sharded variable creator.

    Returns:
      A dict of dicts. The outer dict is keyed by the table names and the inner
      dicts are keyed by 'parameters' and the slot variable names.
    """

    def create_variables(table):
      """Create all variables."""
      variable_shape = (table.vocabulary_size, table.dim)

      def getter(name, shape, dtype, initializer, trainable):
        del shape
        # _add_variable_with_custom_getter clears the shape sometimes, so we
        # take the global shape from outside the getter.
        initial_value = functools.partial(initializer, variable_shape,
                                          dtype=dtype)
        return tf_variables.Variable(
            name=name,
            initial_value=initial_value,
            shape=variable_shape,
            dtype=dtype,
            trainable=trainable)

      def variable_creator(name, initializer, trainable=True):
        # use add_variable_with_custom_getter here so that we take advantage of
        # the checkpoint loading to allow restore before the variables get
        # created which avoids double initialization.
        return self._add_variable_with_custom_getter(
            name=name,
            initializer=initializer,
            shape=variable_shape,
            dtype=dtypes.float32,
            getter=getter,
            trainable=trainable)

      parameters = variable_creator(table.name, table.initializer,
                                    trainable=not self._using_tpu)

      def slot_creator(name, initializer):
        return variable_creator(table.name + "/" + name,
                                initializer,
                                False)

      if table.optimizer is not None:
        slot_vars = table.optimizer._create_slots(parameters, slot_creator)  # pylint: disable=protected-access
      else:
        slot_vars = {}
      slot_vars["parameters"] = parameters
      return slot_vars

    # Store tables based on name rather than TableConfig as we can't track
    # through dicts with non-string keys, i.e. we won't be able to save.
    variables = {}
    for table in self._table_config:
      if not self._using_tpu:
        variables[table.name] = create_variables(table)
      else:
        with variable_scope.variable_creator_scope(
            make_sharded_variable_creator(self._hosts)):
          variables[table.name] = create_variables(table)

    return variables

  def _load_variables(self):
    # Only load the variables if we are:
    # 1) Using TPU
    # 2) Variables are created
    # 3) Not in save context (except if running eagerly)
    if self._using_tpu and self._built and not (
        not context.executing_eagerly() and save_context.in_save_context()):
      _load_variables_impl(self._config_proto.SerializeToString(),
                           self._hosts,
                           self._variables,
                           self._table_config)

  def _retrieve_variables(self):
    # Only retrieve the variables if we are:
    # 1) Using TPU
    # 2) Variables are created
    # 3) Not in save context (except if running eagerly)
    if self._using_tpu and self._built and not (
        not context.executing_eagerly() and save_context.in_save_context()):
      _retrieve_variables_impl(self._config_proto.SerializeToString(),
                               self._hosts,
                               self._variables,
                               self._table_config)

  # Some helper functions for the below enqueue function.
  def _add_data_for_tensor(self, tensor, weight, indices, values, weights,
                           int_zeros, float_zeros, path):
    if weight is not None:
      raise ValueError(
          "Weight specified for dense input {}, which is not allowed. "
          "Weight will always be 1 in this case.".format(path))
    # For tensors, there are no indices and no weights.
    indices.append(int_zeros)
    values.append(math_ops.cast(array_ops.reshape(tensor, [-1]), dtypes.int64))
    weights.append(float_zeros)

  def _add_data_for_sparse_tensor(self, tensor, weight, indices, values,
                                  weights, int_zeros, float_zeros, path,
                                  feature):
    sample_indices = math_ops.cast(tensor.indices, dtypes.int32)
    if tensor.shape.rank == 2:
      if not feature.output_shape and feature.max_sequence_length > 0:
        # Add one dimension to the last axis.
        sample_indices = array_ops.pad(
            sample_indices, paddings=[[0, 0], [0, 1]])
    indices.append(sample_indices)
    values.append(math_ops.cast(tensor.values, dtypes.int64))
    # If we have weights they must be a SparseTensor.
    if weight is not None:
      if not isinstance(weight, sparse_tensor.SparseTensor):
        raise ValueError("Weight for {} is type {} which does not match "
                         "type input which is SparseTensor.".format(
                             path, type(weight)))
      weights.append(math_ops.cast(weight.values, dtypes.float32))
    else:
      weights.append(float_zeros)

  def _add_data_for_ragged_tensor(self, tensor, weight, row_splits, values,
                                  weights, int_zeros, float_zeros, path,
                                  feature):
    row_splits.append(math_ops.cast(tensor.row_splits, dtypes.int32))
    values.append(math_ops.cast(tensor.values, dtypes.int64))
    # If we have weights they must be a RaggedTensor.
    if weight is not None:
      if not isinstance(weight, ragged_tensor.RaggedTensor):
        raise ValueError("Weight for {} is type {} which does not match "
                         "type input which is RaggedTensor.".format(
                             path, type(weight)))
      weights.append(math_ops.cast(weight.values, dtypes.float32))
    else:
      weights.append(float_zeros)

  def _generate_enqueue_op(
      self,
      flat_inputs: List[internal_types.NativeObject],
      flat_weights: List[Optional[internal_types.NativeObject]],
      flat_features: List[tpu_embedding_v2_utils.FeatureConfig],
      device_ordinal: int,
      mode_override: Text
  ) -> ops.Operation:
    """Outputs a the enqueue op given the inputs and weights.

    Args:
      flat_inputs: A list of input tensors.
      flat_weights: A list of input weights (or None) of the same length as
        flat_inputs.
      flat_features: A list of FeatureConfigs of the same length as flat_inputs.
      device_ordinal: The device to create the enqueue op for.
      mode_override: A tensor containing the string "train" or "inference".

    Returns:
      The enqueue op.
    """
    # Combiners are per table, list in the same order as the table order.
    combiners = [table.combiner for table in self._table_config]

    # These parallel arrays will be the inputs to the enqueue op.
    # sample_indices for sparse, row_splits for ragged.
    indices_or_row_splits = []
    values = []
    weights = []

    # We have to supply a empty/zero tensor in a list position where we don't
    # have data (e.g. indices for standard Tensor input, weight when no weight
    # is specified). We create one op here per call, so that we reduce the
    # graph size.
    int_zeros = array_ops.zeros((0,), dtype=dtypes.int32)
    float_zeros = array_ops.zeros((0,), dtype=dtypes.float32)

    # In the following loop we insert casts so that everything is either int32
    # or float32. This is because op inputs which are lists of tensors must be
    # of the same type within the list. Moreover the CPU implementations of
    # these ops cast to these types anyway, so we don't lose any data by casting
    # early.
    for inp, weight, (path, feature) in zip(
        flat_inputs, flat_weights, flat_features):
      if isinstance(inp, ops.Tensor):
        self._add_data_for_tensor(inp, weight, indices_or_row_splits, values,
                                  weights, int_zeros, float_zeros, path)
      elif isinstance(inp, sparse_tensor.SparseTensor):
        self._add_data_for_sparse_tensor(inp, weight, indices_or_row_splits,
                                         values, weights, int_zeros,
                                         float_zeros, path, feature)
      elif isinstance(inp, ragged_tensor.RaggedTensor):
        self._add_data_for_ragged_tensor(inp, weight, indices_or_row_splits,
                                         values, weights, int_zeros,
                                         float_zeros, path, feature)
      else:
        raise ValueError("Input {} is of unknown type {}. Please only pass "
                         "Tensor, SparseTensor or RaggedTensor as input to "
                         "enqueue.".format(path, type(inp)))

    return tpu_ops.enqueue_tpu_embedding_arbitrary_tensor_batch(
        sample_indices_or_row_splits=indices_or_row_splits,
        embedding_indices=values,
        aggregation_weights=weights,
        mode_override=mode_override,
        device_ordinal=device_ordinal,
        combiners=combiners)

  def _raise_error_for_incorrect_control_flow_context(self):
    """Raises an error if we are not in the TPUReplicateContext."""
    # Do not allow any XLA control flow (i.e. control flow in between a
    # TPUStrategy's run call and the call to this function), as we can't
    # extract the enqueue from the head when in XLA control flow.
    graph = ops.get_default_graph()
    in_tpu_ctx = False
    while graph is not None:
      ctx = graph._get_control_flow_context()  # pylint: disable=protected-access
      while ctx is not None:
        if isinstance(ctx, tpu.TPUReplicateContext):
          in_tpu_ctx = True
          break
        ctx = ctx.outer_context
      if in_tpu_ctx:
        break
      graph = getattr(graph, "outer_graph", None)
    if graph != ops.get_default_graph() and in_tpu_ctx:
      raise RuntimeError(
          "Current graph {} does not match graph which contains "
          "TPUReplicateContext {}. This is most likely due to the fact that "
          "enqueueing embedding data is called inside control flow or a "
          "nested function inside `strategy.run`. This is not supported "
          "because outside compilation fails to extract the enqueue ops as "
          "head of computation.".format(ops.get_default_graph(), graph))
    return in_tpu_ctx

  def _raise_error_for_non_direct_inputs(self, features):
    """Checks all tensors in features to see if they are a direct input."""

    # expand_composites here is important: as composite tensors pass through
    # tpu.replicate, they get 'flattened' into their component tensors and then
    # repacked before being passed to the tpu function. In means that it is the
    # component tensors which are produced by an op with the
    # "_tpu_input_identity" attribute.
    for path, input_tensor in nest.flatten_with_joined_string_paths(
        features, expand_composites=True):
      if input_tensor.op.type == "Placeholder":
        continue
      try:
        is_input = input_tensor.op.get_attr("_tpu_input_identity")
      except ValueError:
        is_input = False
      if not is_input:
        raise ValueError(
            "Received input tensor {} which is the output of op {} (type {}) "
            "which does not have the `_tpu_input_identity` attr. Please "
            "ensure that the inputs to this layer are taken directly from "
            "the arguments of the function called by "
            "strategy.run. Two possible causes are: dynamic batch size "
            "support or you are using a keras layer and are not passing "
            "tensors which match the dtype of the `tf.keras.Input`s."
            "If you are triggering dynamic batch size support, you can "
            "disable it by passing tf.distribute.RunOptions("
            "experimental_enable_dynamic_batch_size=False) to the options "
            "argument of strategy.run().".format(path,
                                                 input_tensor.op.name,
                                                 input_tensor.op.type))

  def _raise_error_for_inputs_not_on_cpu(self, flat_inputs, flat_paths):
    """Checks all tensors in features to see are placed on the CPU."""

    def check_device(path, device_string):
      spec = tf_device.DeviceSpec.from_string(device_string)
      if spec.device_type == "TPU":
        raise ValueError(
            "Received input tensor {} which is on a TPU input device {}. Input "
            "tensors for TPU embeddings must be placed on the CPU. Please "
            "ensure that your dataset is prefetching tensors to the host by "
            "setting the 'experimental_fetch_to_device' option of the "
            "dataset distribution function. See the documentation of the "
            "enqueue method for an example.".format(path, device_string))

    # expand_composites here is important, we need to check the device of each
    # underlying tensor.
    for input_tensor, input_path in zip(flat_inputs, flat_paths):
      if nest.is_nested_or_composite(input_tensor):
        input_tensors = nest.flatten(input_tensor, expand_composites=True)
      else:
        input_tensors = [input_tensor]
      for t in input_tensors:
        if (t.op.type == "Identity" and
            t.op.inputs[0].op.type == "TPUReplicatedInput"):
          for tensor in t.op.inputs[0].op.inputs:
            check_device(input_path, tensor.device)
        else:
          check_device(input_path, t.device)

  def enqueue(
      self,
      features,
      weights=None,
      training: bool = True,
      name: Optional[Text] = None,
      device: Optional[Text] = None):
    """Enqueues id tensors for embedding lookup.

    This function enqueues a structure of features to be looked up in the
    embedding tables. We expect that the input shapes of each of the tensors in
    features matches the output shapes set via FeatureConfig or build method
    (if any). the output shapes will be auto detected based on the input shapes
    with the max_sequence_length or output shape setting in the FeatureConfig.
    Note that the output shapes is based on per replica batch size.
    If your input dataset is batched to the global batch size and you use
    `tf.distribute.TPUStrategy`'s `experimental_distribute_dataset`
    or if you use `distribute_datasets_from_function` and batch
    to the per core batch size computed by the context passed to your input
    function, the output shapes should match automatically.

    The auto detected the output shapes:
      1. For dense tensor, if rank 2 or above, make sure the tensor has last
         dimension as 1. The output shape will be the input shape excluding
         the last dimension.
      2. For sparse tensor, make sure the tensor has rank 2 and above.
           a. If feature config has max_sequence_length equals 0 or output shape
              set (the max_sequence_length setting will be ignored), the
              output shape will be the input shape excluding the last dimension.
           b. Otherwize if the tensor is rank 2, the output shape will be input
              shape  with last dimension set as max_sequence_length. If the
              tensor is above rank 2, the output shape will be the input shape
              excluding the last dimension and the last dimension of the output
              shape will be set to max_sequence_length.
      3. For ragged tensor, make sure the tensor has rank 2.
           a. If feature config has max_sequence_length equals 0 or output shape
              set (the max_sequence_length setting will be ignored), the
              output shape will be the input shape excluding the last dimension.
           b. Otherwise, the output shape will be the input shape excluding the
              last dimension and the last dimension of the output shape will be
              set to max_sequence_length.

    ```python
    strategy = tf.distribute.TPUStrategy(...)
    with strategy.scope():
      embedding = tf.tpu.experimental.embedding.TPUEmbedding(...)

    distributed_dataset = (
        strategy.distribute_datasets_from_function(
            dataset_fn=...,
            options=tf.distribute.InputOptions(
                experimental_fetch_to_device=False))
    dataset_iterator = iter(distributed_dataset)

    @tf.function
    def training_step():
      def tpu_step(tpu_features):
        with tf.GradientTape() as tape:
          activations = embedding.dequeue()
          tape.watch(activations)

          loss = ... #  some computation involving activations

        embedding_gradients = tape.gradient(loss, activations)
        embedding.apply_gradients(embedding_gradients)

      embedding_features, tpu_features = next(dataset_iterator)
      embedding.enqueue(embedding_features, training=True)
      strategy.run(tpu_step, args=(tpu_features,))

    training_step()
    ```

    NOTE: You should specify `training=True` when using
    `embedding.apply_gradients` as above and `training=False` when not using
    `embedding.apply_gradients` (e.g. for frozen embeddings or when doing
    evaluation).

    For finer grained control, in the above example the line

    ```
      embedding.enqueue(embedding_features, training=True)
    ```

    may be replaced with

    ```
      per_core_embedding_features = self.strategy.experimental_local_results(
          embedding_features)

      def per_core_enqueue(ctx):
        core_id = ctx.replica_id_in_sync_group
        device = strategy.extended.worker_devices[core_id]
        embedding.enqueue(per_core_embedding_features[core_id],
                          device=device)

      strategy.experimental_distribute_values_from_function(
          per_core_queue_inputs)
    ```

    Args:
      features: A nested structure of `tf.Tensor`s, `tf.SparseTensor`s or
        `tf.RaggedTensor`s, with the same structure as `feature_config`. Inputs
        will be downcast to `tf.int32`. Only one type out of `tf.SparseTensor`
        or `tf.RaggedTensor` is supported per call.
      weights: If not `None`, a nested structure of `tf.Tensor`s,
        `tf.SparseTensor`s or `tf.RaggedTensor`s, matching the above, except
        that the tensors should be of float type (and they will be downcast to
        `tf.float32`). For `tf.SparseTensor`s we assume the `indices` are the
        same for the parallel entries from `features` and similarly for
        `tf.RaggedTensor`s we assume the row_splits are the same.
      training: Defaults to `True`. If `False`, enqueue the batch as inference
        batch (forward pass only). Do not call `apply_gradients` when this is
        `False` as this may lead to a deadlock.
       name: A name for the underlying op.
       device: The device name (e.g. '/task:0/device:TPU:2') where this batch
         should be enqueued. This should be set if and only if features is not a
         `tf.distribute.DistributedValues` and enqueue is not being called
         inside a TPU context (e.g. inside `TPUStrategy.run`).

    Raises:
      ValueError: When called inside a strategy.run call and input is not
        directly taken from the args of the `strategy.run` call. Also if
        the size of any sequence in `features` does not match corresponding
        sequence in `feature_config`. Similarly for `weights`, if not `None`.
        If input shapes of features is unequal or different from a previous
        call.
      RuntimeError: When called inside a strategy.run call and inside XLA
        control flow. If batch_size is not able to be determined and build was
        not called.
      TypeError: If the type of any sequence in `features` does not match
        corresponding sequence in `feature_config`. Similarly for `weights`, if
        not `None`.
    """
    if not self._using_tpu:
      raise RuntimeError("enqueue is not valid when TPUEmbedding object is not "
                         "created under a TPUStrategy.")

    in_tpu_context = self._raise_error_for_incorrect_control_flow_context()

    nest.assert_same_structure(self._feature_config, features)

    if not self._verify_output_shapes_on_enqueue:
      if not self._output_shapes or not self._built:
        raise ValueError(
            "Configured not to check output shapes on each enqueue() call; please "
            "ensure build() was called with output shapes to initialize "
            "the TPU for embeddings.")
    else:
      input_shapes = self._get_input_shapes(features, in_tpu_context)

      self._maybe_build(input_shapes)
      # If is already built, we still need to check if the output shapes matches
      # with the previous ones.
      self._check_output_shapes(
          self._get_output_shapes_from_input_shapes(input_shapes))

    flat_inputs = nest.flatten(features)
    flat_weights = [None] * len(flat_inputs)
    if weights is not None:
      nest.assert_same_structure(self._feature_config, weights)
      flat_weights = nest.flatten(weights)
    flat_features = nest.flatten_with_joined_string_paths(self._feature_config)
    flat_paths, _ = zip(*flat_features)

    self._raise_error_for_inputs_not_on_cpu(flat_inputs, flat_paths)
    # If we are in a tpu_context, automatically apply outside compilation.
    if in_tpu_context:
      self._raise_error_for_non_direct_inputs(features)

      def generate_enqueue_ops():
        """Generate enqueue ops for outside compilation."""
        # Note that we put array_ops.where_v2 rather than a python if so that
        # the op is explicitly create and the constant ops are both in the graph
        # even though we don't expect training to be a tensor (and thus generate
        # control flow automatically). This need to make it easier to re-write
        # the graph later if we need to fix which mode needs to be used.
        mode_override = array_ops.where_v2(training,
                                           constant_op.constant("train"),
                                           constant_op.constant("inference"))
        # Device ordinal is -1 here, a later rewrite will fix this once the op
        # is expanded by outside compilation.
        enqueue_op = self._generate_enqueue_op(
            flat_inputs, flat_weights, flat_features, device_ordinal=-1,
            mode_override=mode_override)

        # Apply the name tag to the op.
        if name is not None:
          _add_key_attr(enqueue_op, name)

        # Ensure that this op has outbound control flow, otherwise it won't be
        # executed.
        ops.get_default_graph().control_outputs.append(enqueue_op)

      tpu.outside_compilation(generate_enqueue_ops)

    elif device is None:
      mode_override = "train" if training else "inference"
      # We generate enqueue ops per device, so we need to gather the all
      # features for a single device in to a dict.
      # We rely here on the fact that the devices in the PerReplica value occur
      # in the same (standard) order as self._strategy.extended.worker_devices.
      enqueue_ops = []
      for replica_id in range(self._strategy.num_replicas_in_sync):
        replica_inputs = distribute_utils.select_replica(replica_id,
                                                         flat_inputs)
        replica_weights = distribute_utils.select_replica(replica_id,
                                                          flat_weights)
        tpu_device = self._strategy.extended.worker_devices[replica_id]
        # TPU devices string are like /job:worker/replica:0/task:0/device:TPU:0
        # the device ordinal is the last number
        device_ordinal = (
            tf_device.DeviceSpec.from_string(tpu_device).device_index)

        with ops.device(device_util.get_host_for_device(tpu_device)):
          enqueue_op = self._generate_enqueue_op(
              replica_inputs, replica_weights, flat_features,
              device_ordinal=device_ordinal, mode_override=mode_override)

          # Apply the name tag to the op.
          if name is not None:
            _add_key_attr(enqueue_op, name)
          enqueue_ops.append(enqueue_op)
      ops.get_default_graph().control_outputs.extend(enqueue_ops)
    else:
      mode_override = "train" if training else "inference"
      device_spec = tf_device.DeviceSpec.from_string(device)
      if device_spec.device_type != "TPU":
        raise ValueError(
            "Non-TPU device {} passed to enqueue.".format(device))

      with ops.device(device_util.get_host_for_device(device)):
        enqueue_op = self._generate_enqueue_op(
            flat_inputs, flat_weights, flat_features,
            device_ordinal=device_spec.device_index,
            mode_override=mode_override)

        # Apply the name tag to the op.
        if name is not None:
          _add_key_attr(enqueue_op, name)
        ops.get_default_graph().control_outputs.append(enqueue_op)

  def _get_input_shapes(self, tensors,
                        in_tpu_context: bool) -> List[TensorShape]:
    """Get the input shapes from the input tensor."""
    input_shapes = []
    for (path, maybe_tensor), feature in zip(
        nest.flatten_with_joined_string_paths(tensors),
        nest.flatten(self._feature_config)):
      if not in_tpu_context:
        tensor = distribute_utils.select_replica(0, maybe_tensor)
      else:
        tensor = maybe_tensor

      if isinstance(tensor, ops.Tensor):
        input_shapes.append(
            self._get_input_shape_for_tensor(tensor, feature, path))
      elif isinstance(tensor, sparse_tensor.SparseTensor):
        input_shapes.append(
            self._get_input_shape_for_sparse_tensor(tensor, feature, path))
      elif isinstance(tensor, ragged_tensor.RaggedTensor):
        input_shapes.append(
            self._get_input_shape_for_ragged_tensor(tensor, feature, path))
    return input_shapes

  def _get_input_shape_for_tensor(self, tensor, feature, path) -> TensorShape:
    """Get the input shape for the dense tensor."""
    shape = tensor.shape.as_list()
    if len(shape) < 1:
      raise ValueError("Only rank 1 and above dense tensor is supported,"
                       " find rank {} sparse tensor for input {}".format(
                           len(shape), path))
    if len(shape) > 1 and shape[-1] != 1:
      raise ValueError(
          "Rank 2 or above dense tensor should have last dimension as 1 "
          "as the last dimension will always be reduced. "
          "Instead got dense tensor as shape {}".format(shape))
    return TensorShape(shape)

  def _get_input_shape_for_sparse_tensor(self, tensor, feature,
                                         path) -> TensorShape:
    """Get the input shape for the sparse tensor."""
    shape = tensor.shape.as_list()
    # Only 2 and above rank sparse tensor is supported.
    if len(shape) < 2:
      raise ValueError("Only rank 2 and above sparse tensor is supported,"
                       " find rank {} sparse tensor for input {}".format(
                           len(shape), path))
    if not feature.output_shape and feature.max_sequence_length > 0:
      # If the max_sequence_length is set and the output shape for FeatureConfig
      # is not set, we modify the shape of the input feature. Only rank 2
      # feature output shape is modified
      if len(shape) == 2:
        # If the sparse tensor is 2D and max_sequence_length is set,
        # we need to add one dimension to the input feature.
        shape.insert(len(shape) - 1, feature.max_sequence_length)

    return TensorShape(shape)

  def _get_input_shape_for_ragged_tensor(self, tensor, feature,
                                         path) -> TensorShape:
    """Get the input shape for the ragged tensor."""
    shape = tensor.shape.as_list()
    # Only rank 2 ragged tensor is supported.
    if len(shape) != 2:
      raise ValueError("Only rank 2 ragged tensor is supported,"
                       " find rank {} ragged tensor for input {}".format(
                           len(shape), path))
    if not feature.output_shape and feature.max_sequence_length > 0:
      # If the max_sequence_length is set and the output shape for FeatureConfig
      # is not set, add the sequence length as second last dimension of
      # the ragged tensor.
      shape.insert(len(shape) - 1, feature.max_sequence_length)

    return TensorShape(shape)

  def _update_output_shapes(self, incoming_output_shapes: List[TensorShape]):
    """Update the existing output shapes based on the new output shapes.

    The existing output shapes always have higher piority than the new incoming
    output shapes.
    Args:
      incoming_output_shapes: nested structure of TensorShape to override the
        existing output shapes.
    """
    nest.assert_same_structure(self._output_shapes, incoming_output_shapes)
    updated_output_shapes = []
    for old_output_shape, incoming_output_shape in zip(self._output_shapes,
                                                       incoming_output_shapes):
      if old_output_shape:
        updated_output_shapes.append(old_output_shape)
      else:
        updated_output_shapes.append(incoming_output_shape)
    self._output_shapes = updated_output_shapes

  def _check_output_shapes(self, incoming_output_shapes: List[TensorShape]):
    """Check the incoming output shapes against the output shapes stored."""
    # The incoming output shape should have the same structure with the existing
    # output shapes.
    nest.assert_same_structure(self._output_shapes, incoming_output_shapes)

    for (path, _), old_output_shape, incoming_output_shape in zip(
        nest.flatten_with_joined_string_paths(self._feature_config),
        self._output_shapes, incoming_output_shapes):
      # First check if both shapes are not None.
      if old_output_shape and incoming_output_shape:
        # We skip the check when the incoming output shape is rank 1 or 2 and
        # rank of the old output shape is larger. This can happen for
        # (sequence) ragged tensor, we push the check down to the enqueue op.
        if (len(incoming_output_shape) == 1 or len(incoming_output_shape)
            == 2) and len(old_output_shape) > len(incoming_output_shape):
          continue
        if len(old_output_shape) != len(
            incoming_output_shape) or not self._is_tensor_shape_match(
                old_output_shape, incoming_output_shape):
          raise ValueError(
              f"Inconsistent shape founded for input feature {path}, "
              f"Output shape is set to be {old_output_shape}, "
              f"But got incoming output shape {incoming_output_shape}")

  def _check_output_shapes_fully_defined(self):
    """Check if the output shape is fully defined."""
    for (path, _), output_shape in zip(
        nest.flatten_with_joined_string_paths(self._feature_config),
        self._output_shapes):
      if not output_shape.is_fully_defined():
        raise ValueError(
            f"Input Feature {path} has output shape set as "
            f"{output_shape} which is not fully defined. "
            "Please specify the fully defined shape in either FeatureConfig "
            "or for the build method.")

  def _is_tensor_shape_match(self, shape_a: TensorShape,
                             shape_b: TensorShape) -> bool:
    """Check if shape b matches with shape a."""
    for s_a, s_b in zip(shape_a.as_list(), shape_b.as_list()):
      if s_a and s_b and s_a != s_b:
        return False
    return True

  def _get_output_shapes_from_batch_size(self, per_replica_batch_size):
    """Get the output shapes from the batch size."""
    output_shapes = []
    for feature in nest.flatten(self._feature_config):
      if not feature.output_shape and feature.max_sequence_length > 0:
        output_shapes.append(
            TensorShape([per_replica_batch_size, feature.max_sequence_length]))
      else:
        output_shapes.append(TensorShape(per_replica_batch_size))
    return output_shapes


@def_function.function
def _load_variables_impl(
    config: Text,
    hosts: List[Tuple[int, Text]],
    variables: Dict[Text, Dict[Text, tf_variables.Variable]],
    table_config: tpu_embedding_v2_utils.TableConfig):
  """Load embedding tables to onto TPU for each table and host.

  Args:
    config: A serialized TPUEmbeddingConfiguration proto.
    hosts: A list of CPU devices, on per host.
    variables: A dictionary of dictionaries of TPUEmbeddingVariables. First key
      is the table name, second key is 'parameters' or the optimizer slot name.
    table_config: A list of tf.tpu.experimental.embedding.TableConfig objects.
  """
  def select_fn(host_id):

    def select_or_zeros(x):
      if host_id >= len(x.variables):
        # In the edge case where we have more hosts than variables, due to using
        # a small number of rows, we load zeros for the later hosts. We copy
        # the shape of the first host's variables, which we assume is defined
        # because TableConfig guarantees at least one row.
        return array_ops.zeros_like(x.variables[0])
      return x.variables[host_id]

    return select_or_zeros

  for host_id, host in enumerate(hosts):
    with ops.device(host):
      host_variables = nest.map_structure(select_fn(host_id), variables)
      for table in table_config:
        table.optimizer._load()(  # pylint: disable=protected-access
            table_name=table.name,
            num_shards=len(hosts),
            shard_id=host_id,
            config=config,
            **host_variables[table.name])
        # Ensure that only the first table/first host gets a config so that we
        # don't bloat graph by attaching this large string to each op.
        # We have num tables * num hosts of these so for models with a large
        # number of tables training on a large slice, this can be an issue.
        config = None


@def_function.function
def _retrieve_variables_impl(
    config: Text,
    hosts: List[Tuple[int, Text]],
    variables: Dict[Text, Dict[Text, tf_variables.Variable]],
    table_config: tpu_embedding_v2_utils.TableConfig):
  """Retrieve embedding tables from TPU to host memory.

  Args:
    config: A serialized TPUEmbeddingConfiguration proto.
    hosts: A list of all the host CPU devices.
    variables: A dictionary of dictionaries of TPUEmbeddingVariables. First key
      is the table name, second key is 'parameters' or the optimizer slot name.
    table_config: A list of tf.tpu.experimental.embedding.TableConfig objects.
  """
  for host_id, host in enumerate(hosts):
    with ops.device(host):
      for table in table_config:
        retrieved = table.optimizer._retrieve()(  # pylint: disable=protected-access
            table_name=table.name,
            num_shards=len(hosts),
            shard_id=host_id,
            config=config)
        # When there are no slot variables (e.g with SGD) this returns a
        # single tensor rather than a tuple. In this case we put the tensor in
        # a list to make the following code easier to write.
        if not isinstance(retrieved, tuple):
          retrieved = (retrieved,)

        for i, slot in enumerate(["parameters"] +
                                 table.optimizer._slot_names()):  # pylint: disable=protected-access
          # We must assign the CPU variables the values of tensors that were
          # returned from the TPU.
          sharded_var = variables[table.name][slot]
          if host_id < len(sharded_var.variables):
            # In the edge case where we have more hosts than variables, due to
            # using a small number of rows, we skip the later hosts.
            sharded_var.variables[host_id].assign(retrieved[i])
        # Ensure that only the first table/first host gets a config so that we
        # don't bloat graph by attaching this large string to each op.
        # We have num tables * num hosts of these so for models with a large
        # number of tables training on a large slice, this can be an issue.
        config = None


def _save_callback(trackables, **unused_kwargs):
  for trackable in trackables.values():
    trackable._retrieve_variables()  # pylint: disable=protected-access
  return []


def _restore_callback(trackables, **unused_kwargs):
  for trackable in trackables.values():
    trackable._load_variables()  # pylint: disable=protected-access


registration.register_tf_checkpoint_saver(
    "TPUEmbeddingCallback",
    predicate=lambda x: isinstance(x, TPUEmbedding),
    save_fn=_save_callback,
    restore_fn=_restore_callback,
    # Set strict_predicate_restore to `False` to because the isinstance
    # predicate check does not pass after a TPUEmbedding object is loaded from
    # SavedModel.
    strict_predicate_restore=False
)


def get_list_of_hosts(strategy: tpu_strategy.TPUStrategy) -> List[Text]:
  """Returns a sorted list of CPU devices for the remote jobs.

  Args:
    strategy: A TPUStrategy object.

  Returns:
    A sort list of device strings.
  """
  list_of_hosts = []
  # Assume this is sorted by task
  for tpu_device in strategy.extended.worker_devices:
    host = device_util.get_host_for_device(tpu_device)
    if host not in list_of_hosts:
      list_of_hosts.append(host)
  assert len(list_of_hosts) == strategy.extended.num_hosts
  return list_of_hosts


def extract_variable_info(
    kwargs) -> Tuple[Text, Tuple[int, ...], dtypes.DType, Callable[[], Any]]:
  """Extracts the variable creation attributes from the kwargs.

  Args:
    kwargs: a dict of keyword arguments that were passed to a variable creator
      scope.

  Returns:
    A tuple of variable name, shape, dtype, initialization function.
  """
  if (isinstance(kwargs["initial_value"], functools.partial) and (
      "shape" in kwargs["initial_value"].keywords or
      kwargs["initial_value"].args)):
    # Sometimes shape is passed positionally, sometimes it's passed as a kwarg.
    if "shape" in kwargs["initial_value"].keywords:
      shape = kwargs["initial_value"].keywords["shape"]
    else:
      shape = kwargs["initial_value"].args[0]
    return (kwargs["name"], shape,
            kwargs["initial_value"].keywords.get("dtype", kwargs["dtype"]),
            kwargs["initial_value"].func)
  elif "shape" not in kwargs or kwargs["shape"] is None or not callable(
      kwargs["initial_value"]):
    raise ValueError(
        "Unable to extract initializer function and shape from {}. Please "
        "either pass a function that expects a shape and dtype as the "
        "initial value for your variable or functools.partial object with "
        "the shape and dtype kwargs set. This is needed so that we can "
        "initialize the shards of the ShardedVariable locally.".format(
            kwargs["initial_value"]))
  else:
    return (kwargs["name"], kwargs["shape"], kwargs["dtype"],
            kwargs["initial_value"])


def make_sharded_variable_creator(
    hosts: List[Text]) -> Callable[..., TPUEmbeddingVariable]:
  """Makes a sharded variable creator given a list of hosts.

  Args:
    hosts: a list of tensorflow devices on which to shard the tensors.

  Returns:
    A variable creator function.
  """

  def sharded_variable_creator(
      next_creator: Callable[..., tf_variables.Variable], *args, **kwargs):
    """The sharded variable creator."""
    kwargs["skip_mirrored_creator"] = True

    num_hosts = len(hosts)
    name, shape, dtype, unwrapped_initial_value = extract_variable_info(kwargs)
    initial_value = kwargs["initial_value"]
    rows = shape[0]
    cols = shape[1]
    partial_partition = rows % num_hosts
    full_rows_per_host = rows // num_hosts
    # We partition as if we were using MOD sharding: at least
    # `full_rows_per_host` rows to `num_hosts` hosts, where the first
    # `partial_partition` hosts get an additional row when the number of rows
    # is not cleanly divisible. Note that `full_rows_per_host` may be zero.
    partitions = (
        [full_rows_per_host + 1] * partial_partition
        + [full_rows_per_host] * (num_hosts - partial_partition))
    variables = []
    sharding_aware = "shard_info" in tf_inspect.getargspec(initial_value).args

    # Keep track of offset for sharding aware initializers.
    offset = 0
    kwargs["dtype"] = dtype
    for i, p in enumerate(partitions):
      if p == 0:
        # Skip variable creation for empty partitions, resulting from the edge
        # case of 'rows < num_hosts'. This is safe because both load/restore
        # can handle the missing values.
        continue
      with ops.device(hosts[i]):
        kwargs["name"] = "{}_{}".format(name, i)
        kwargs["shape"] = (p, cols)
        if sharding_aware:
          shard_info = base.ShardInfo(kwargs["shape"], (offset, 0))
          kwargs["initial_value"] = functools.partial(
              initial_value, shard_info=shard_info)
          offset += p
        else:
          kwargs["initial_value"] = functools.partial(
              unwrapped_initial_value, kwargs["shape"], dtype=dtype)
        variables.append(next_creator(*args, **kwargs))
    return TPUEmbeddingVariable(variables, name=name)
  return sharded_variable_creator