modeling_tf_utils.py

# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
# Copyright (c) 2018, NVIDIA CORPORATION.  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.
"""TF general model utils."""
import functools
import logging
import os

import h5py
import numpy as np
import tensorflow as tf
from tensorflow.python.keras.saving import hdf5_format

from .configuration_utils import PretrainedConfig
from .file_utils import DUMMY_INPUTS, TF2_WEIGHTS_NAME, WEIGHTS_NAME, cached_path, hf_bucket_url, is_remote_url
from .modeling_tf_pytorch_utils import load_pytorch_checkpoint_in_tf2_model


logger = logging.getLogger(__name__)


class TFModelUtilsMixin:
    """
    A few utilities for `tf.keras.Model`s, to be used as a mixin.
    """

    def num_parameters(self, only_trainable: bool = False) -> int:
        """
        Get number of (optionally, trainable) parameters in the model.
        """
        if only_trainable:
            return int(sum(np.prod(w.shape.as_list()) for w in self.trainable_variables))
        else:
            return self.count_params()


def keras_serializable(cls):
    """
    Decorate a Keras Layer class to support Keras serialization.

    This is done by:
    1. adding a `transformers_config` dict to the Keras config dictionary in `get_config` (called by Keras at
       serialization time
    2. wrapping `__init__` to accept that `transformers_config` dict (passed by Keras at deserialization time) and
       convert it to a config object for the actual layer initializer
    3. registering the class as a custom object in Keras (if the Tensorflow version supports this), so that it does
       not need to be supplied in `custom_objects` in the call to `tf.keras.models.load_model`

    :param cls: a tf.keras.layers.Layers subclass that accepts a `config` argument to its initializer (typically a
                `TF*MainLayer` class in this project)
    :return: the same class object, with modifications for Keras deserialization.
    """
    initializer = cls.__init__

    config_class = getattr(cls, "config_class", None)
    if config_class is None:
        raise AttributeError("Must set `config_class` to use @keras_serializable")

    @functools.wraps(initializer)
    def wrapped_init(self, *args, **kwargs):
        transformers_config = kwargs.pop("transformers_config", None)
        config = args[0] if args and isinstance(args[0], PretrainedConfig) else kwargs.get("config", None)
        if config is not None and transformers_config is not None:
            raise ValueError("Must pass either `config` or `transformers_config`, not both")
        elif config is not None:
            # normal layer construction, call with unchanged args (config is already in there)
            initializer(self, *args, **kwargs)
        elif transformers_config is not None:
            # Keras deserialization, convert dict to config
            config = config_class.from_dict(transformers_config)
            initializer(self, config, *args, **kwargs)
        else:
            raise ValueError("Must pass either `config` (PretrainedConfig) or `transformers_config` (dict)")
        self._transformers_config = config
        self._kwargs = kwargs

    cls.__init__ = wrapped_init

    if not hasattr(cls, "get_config"):
        raise TypeError("Only use @keras_serializable on tf.keras.layers.Layer subclasses")
    if hasattr(cls.get_config, "_is_default"):

        def get_config(self):
            cfg = super(cls, self).get_config()
            cfg["transformers_config"] = self._transformers_config.to_dict()
            cfg.update(self._kwargs)
            return cfg

        cls.get_config = get_config

    cls._keras_serializable = True
    if hasattr(tf.keras.utils, "register_keras_serializable"):
        cls = tf.keras.utils.register_keras_serializable()(cls)
    return cls


class TFQuestionAnsweringLoss:
    def compute_loss(self, labels, logits):
        loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(
            from_logits=True, reduction=tf.keras.losses.Reduction.NONE
        )
        start_loss = loss_fn(labels["start_position"], logits[0])
        end_loss = loss_fn(labels["end_position"], logits[1])

        return (start_loss + end_loss) / 2.0


class TFTokenClassificationLoss:
    def compute_loss(self, labels, logits):
        loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(
            from_logits=True, reduction=tf.keras.losses.Reduction.NONE
        )
        active_loss = tf.reshape(labels, (-1,)) != -1
        reduced_logits = tf.boolean_mask(tf.reshape(logits, (-1, shape_list(logits)[2])), active_loss)
        labels = tf.boolean_mask(tf.reshape(labels, (-1,)), active_loss)

        return loss_fn(labels, reduced_logits)


class TFSequenceClassificationLoss:
    def compute_loss(self, labels, logits):
        if shape_list(logits)[1] == 1:
            loss_fn = tf.keras.losses.MeanSquaredError(reduction=tf.keras.losses.Reduction.NONE)
        else:
            loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(
                from_logits=True, reduction=tf.keras.losses.Reduction.NONE
            )

        return loss_fn(labels, logits)


TFMultipleChoiceLoss = TFSequenceClassificationLoss


class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin):
    r""" Base class for all TF models.

        :class:`~transformers.TFPreTrainedModel` takes care of storing the configuration of the models and handles methods for loading/downloading/saving models
        as well as a few methods common to all models to (i) resize the input embeddings and (ii) prune heads in the self-attention heads.

        Class attributes (overridden by derived classes):
            - ``config_class``: a class derived from :class:`~transformers.PretrainedConfig` to use as configuration class for this model architecture.
            - ``load_tf_weights``: a python ``method`` for loading a TensorFlow checkpoint in a PyTorch model, taking as arguments:

                - ``model``: an instance of the relevant subclass of :class:`~transformers.PreTrainedModel`,
                - ``config``: an instance of the relevant subclass of :class:`~transformers.PretrainedConfig`,
                - ``path``: a path (string) to the TensorFlow checkpoint.

            - ``base_model_prefix``: a string indicating the attribute associated to the base model in derived classes of the same architecture adding modules on top of the base model.
    """
    config_class = None
    base_model_prefix = ""

    @property
    def dummy_inputs(self):
        """ Dummy inputs to build the network.

        Returns:
            tf.Tensor with dummy inputs
        """
        return {"input_ids": tf.constant(DUMMY_INPUTS)}

    def __init__(self, config, *inputs, **kwargs):
        super().__init__(*inputs, **kwargs)
        if not isinstance(config, PretrainedConfig):
            raise ValueError(
                "Parameter config in `{}(config)` should be an instance of class `PretrainedConfig`. "
                "To create a model from a pretrained model use "
                "`model = {}.from_pretrained(PRETRAINED_MODEL_NAME)`".format(
                    self.__class__.__name__, self.__class__.__name__
                )
            )
        # Save config in model
        self.config = config

    def get_input_embeddings(self):
        """
        Returns the model's input embeddings.

        Returns:
            :obj:`tf.keras.layers.Layer`:
                A torch module mapping vocabulary to hidden states.
        """
        base_model = getattr(self, self.base_model_prefix, self)
        if base_model is not self:
            return base_model.get_input_embeddings()
        else:
            raise NotImplementedError

    def set_input_embeddings(self, value):
        """
        Set model's input embeddings

        Args:
            value (:obj:`tf.keras.layers.Layer`):
                A module mapping vocabulary to hidden states.
        """
        base_model = getattr(self, self.base_model_prefix, self)
        if base_model is not self:
            base_model.set_input_embeddings(value)
        else:
            raise NotImplementedError

    def get_output_embeddings(self):
        """
        Returns the model's output embeddings.

        Returns:
            :obj:`tf.keras.layers.Layer`:
                A torch module mapping hidden states to vocabulary.
        """
        return None  # Overwrite for models with output embeddings

    def resize_token_embeddings(self, new_num_tokens=None):
        """ Resize input token embeddings matrix of the model if new_num_tokens != config.vocab_size.
        Take care of tying weights embeddings afterwards if the model class has a `tie_weights()` method.

        Arguments:

            new_num_tokens: (`optional`) int:
                New number of tokens in the embedding matrix. Increasing the size will add newly initialized vectors at the end. Reducing the size will remove vectors from the end.
                If not provided or None: does nothing and just returns a pointer to the input tokens ``tf.Variable`` Module of the model.

        Return: ``tf.Variable``
            Pointer to the input tokens Embeddings Module of the model
        """
        model_embeds = self._resize_token_embeddings(new_num_tokens)
        if new_num_tokens is None:
            return model_embeds

        return model_embeds

    def _resize_token_embeddings(self, new_num_tokens):
        # get_input_embeddings and set_input_embeddings need to be implemented in base layer.
        base_model = getattr(self, self.base_model_prefix, self)
        old_embeddings = base_model.get_input_embeddings()
        new_embeddings = self._get_resized_embeddings(old_embeddings, new_num_tokens)
        base_model.set_input_embeddings(new_embeddings)
        # Update base model and current model config
        self.config.vocab_size = new_num_tokens
        base_model.vocab_size = new_num_tokens
        return base_model.get_input_embeddings()

    def _get_word_embeddings(self, embeddings):
        if hasattr(embeddings, "word_embeddings"):
            # TFBertEmbeddings, TFAlbertEmbeddings, TFElectraEmbeddings
            return embeddings.word_embeddings
        elif hasattr(embeddings, "weight"):
            # TFSharedEmbeddings
            return embeddings.weight
        else:
            raise ValueError("word embedding is not defined.")

    def _get_resized_embeddings(self, old_embeddings, new_num_tokens=None):
        """ Build a resized Embedding Variable from a provided token Embedding Module.
            Increasing the size will add newly initialized vectors at the end
            Reducing the size will remove vectors from the end.

        Args:
            new_num_tokens: (`optional`) int
                New number of tokens in the embedding matrix.
                Increasing the size will add newly initialized vectors at the end
                Reducing the size will remove vectors from the end
                If not provided or None: return the provided token Embedding Module.
        Return: ``tf.Variable``
            Pointer to the resized word Embedding Module or the old Embedding Module if new_num_tokens is None
        """
        word_embeddings = self._get_word_embeddings(old_embeddings)
        if new_num_tokens is None:
            return word_embeddings
        old_num_tokens, old_embedding_dim = word_embeddings.shape
        if old_num_tokens == new_num_tokens:
            return word_embeddings

        # initialize new embeddings
        # todo: initializer range is not always passed in config.
        init_range = getattr(self.config, "initializer_range", 0.02)
        new_embeddings = self.add_weight(
            "weight",
            shape=[new_num_tokens, old_embedding_dim],
            initializer=get_initializer(init_range),
            dtype=tf.float32,
        )
        init_weights = new_embeddings.numpy()

        # Copy token embeddings from the previous weights
        num_tokens_to_copy = min(old_num_tokens, new_num_tokens)
        init_weights[:num_tokens_to_copy] = word_embeddings[:num_tokens_to_copy, :]
        new_embeddings.assign(init_weights)

        return new_embeddings

    def prune_heads(self, heads_to_prune):
        """ Prunes heads of the base model.

            Arguments:

                heads_to_prune: dict with keys being selected layer indices (`int`) and associated values being the list of heads to prune in said layer (list of `int`).
        """
        raise NotImplementedError

    def save_pretrained(self, save_directory):
        """ Save a model and its configuration file to a directory, so that it
            can be re-loaded using the :func:`~transformers.PreTrainedModel.from_pretrained` class method.
        """
        assert os.path.isdir(
            save_directory
        ), "Saving path should be a directory where the model and configuration can be saved"

        # Save configuration file
        self.config.save_pretrained(save_directory)

        # If we save using the predefined names, we can load using `from_pretrained`
        output_model_file = os.path.join(save_directory, TF2_WEIGHTS_NAME)
        self.save_weights(output_model_file)
        logger.info("Model weights saved in {}".format(output_model_file))

    @classmethod
    def from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs):
        r"""Instantiate a pretrained TF 2.0 model from a pre-trained model configuration.

        The warning ``Weights from XXX not initialized from pretrained model`` means that the weights of XXX do not come pre-trained with the rest of the model.
        It is up to you to train those weights with a downstream fine-tuning task.

        The warning ``Weights from XXX not used in YYY`` means that the layer XXX is not used by YYY, therefore those weights are discarded.

        Parameters:
            pretrained_model_name_or_path: either:
                - a string with the `shortcut name` of a pre-trained model to load from cache or download, e.g.: ``bert-base-uncased``.
                - a string with the `identifier name` of a pre-trained model that was user-uploaded to our S3, e.g.: ``dbmdz/bert-base-german-cased``.
                - a path to a `directory` containing model weights saved using :func:`~transformers.PreTrainedModel.save_pretrained`, e.g.: ``./my_model_directory/``.
                - a path or url to a `PyTorch state_dict save file` (e.g. `./pt_model/pytorch_model.bin`). In this case, ``from_pt`` should be set to True and a configuration object should be provided as ``config`` argument. This loading path is slower than converting the PyTorch checkpoint in a TensorFlow model using the provided conversion scripts and loading the TensorFlow model afterwards.

            model_args: (`optional`) Sequence of positional arguments:
                All remaning positional arguments will be passed to the underlying model's ``__init__`` method

            config: (`optional`) one of:
                    - an instance of a class derived from :class:`~transformers.PretrainedConfig`, or
                    - a string valid as input to :func:`~transformers.PretrainedConfig.from_pretrained()`

                Configuration for the model to use instead of an automatically loaded configuation. Configuration can be automatically loaded when:
                    - the model is a model provided by the library (loaded with the ``shortcut-name`` string of a pretrained model), or
                    - the model was saved using :func:`~transformers.PreTrainedModel.save_pretrained` and is reloaded by suppling the save directory.
                    - the model is loaded by suppling a local directory as ``pretrained_model_name_or_path`` and a configuration JSON file named `config.json` is found in the directory.

            from_pt: (`optional`) boolean, default False:
                Load the model weights from a PyTorch state_dict save file (see docstring of pretrained_model_name_or_path argument).

            cache_dir: (`optional`) string:
                Path to a directory in which a downloaded pre-trained model
                configuration should be cached if the standard cache should not be used.

            force_download: (`optional`) boolean, default False:
                Force to (re-)download the model weights and configuration files and override the cached versions if they exists.

            resume_download: (`optional`) boolean, default False:
                Do not delete incompletely recieved file. Attempt to resume the download if such a file exists.

            proxies: (`optional`) dict, default None:
                A dictionary of proxy servers to use by protocol or endpoint, e.g.: {'http': 'foo.bar:3128', 'http://hostname': 'foo.bar:4012'}.
                The proxies are used on each request.

            output_loading_info: (`optional`) boolean:
                Set to ``True`` to also return a dictionnary containing missing keys, unexpected keys and error messages.

            kwargs: (`optional`) Remaining dictionary of keyword arguments:
                Can be used to update the configuration object (after it being loaded) and initiate the model. (e.g. ``output_attention=True``). Behave differently depending on whether a `config` is provided or automatically loaded:

                - If a configuration is provided with ``config``, ``**kwargs`` will be directly passed to the underlying model's ``__init__`` method (we assume all relevant updates to the configuration have already been done)
                - If a configuration is not provided, ``kwargs`` will be first passed to the configuration class initialization function (:func:`~transformers.PretrainedConfig.from_pretrained`). Each key of ``kwargs`` that corresponds to a configuration attribute will be used to override said attribute with the supplied ``kwargs`` value. Remaining keys that do not correspond to any configuration attribute will be passed to the underlying model's ``__init__`` function.

        Examples::

            # For example purposes. Not runnable.
            model = BertModel.from_pretrained('bert-base-uncased')    # Download model and configuration from S3 and cache.
            model = BertModel.from_pretrained('./test/saved_model/')  # E.g. model was saved using `save_pretrained('./test/saved_model/')`
            model = BertModel.from_pretrained('bert-base-uncased', output_attention=True)  # Update configuration during loading
            assert model.config.output_attention == True
            # Loading from a TF checkpoint file instead of a PyTorch model (slower)
            config = BertConfig.from_json_file('./tf_model/my_tf_model_config.json')
            model = BertModel.from_pretrained('./tf_model/my_tf_checkpoint.ckpt.index', from_pt=True, config=config)

        """
        config = kwargs.pop("config", None)
        cache_dir = kwargs.pop("cache_dir", None)
        from_pt = kwargs.pop("from_pt", False)
        force_download = kwargs.pop("force_download", False)
        resume_download = kwargs.pop("resume_download", False)
        proxies = kwargs.pop("proxies", None)
        output_loading_info = kwargs.pop("output_loading_info", False)
        local_files_only = kwargs.pop("local_files_only", False)
        use_cdn = kwargs.pop("use_cdn", True)

        # Load config if we don't provide a configuration
        if not isinstance(config, PretrainedConfig):
            config_path = config if config is not None else pretrained_model_name_or_path
            config, model_kwargs = cls.config_class.from_pretrained(
                config_path,
                *model_args,
                cache_dir=cache_dir,
                return_unused_kwargs=True,
                force_download=force_download,
                resume_download=resume_download,
                proxies=proxies,
                local_files_only=local_files_only,
                **kwargs,
            )
        else:
            model_kwargs = kwargs

        # Load model
        if pretrained_model_name_or_path is not None:
            if os.path.isdir(pretrained_model_name_or_path):
                if os.path.isfile(os.path.join(pretrained_model_name_or_path, TF2_WEIGHTS_NAME)):
                    # Load from a TF 2.0 checkpoint
                    archive_file = os.path.join(pretrained_model_name_or_path, TF2_WEIGHTS_NAME)
                elif from_pt and os.path.isfile(os.path.join(pretrained_model_name_or_path, WEIGHTS_NAME)):
                    # Load from a PyTorch checkpoint
                    archive_file = os.path.join(pretrained_model_name_or_path, WEIGHTS_NAME)
                else:
                    raise EnvironmentError(
                        "Error no file named {} found in directory {} or `from_pt` set to False".format(
                            [WEIGHTS_NAME, TF2_WEIGHTS_NAME], pretrained_model_name_or_path
                        )
                    )
            elif os.path.isfile(pretrained_model_name_or_path) or is_remote_url(pretrained_model_name_or_path):
                archive_file = pretrained_model_name_or_path
            elif os.path.isfile(pretrained_model_name_or_path + ".index"):
                archive_file = pretrained_model_name_or_path + ".index"
            else:
                archive_file = hf_bucket_url(
                    pretrained_model_name_or_path,
                    filename=(WEIGHTS_NAME if from_pt else TF2_WEIGHTS_NAME),
                    use_cdn=use_cdn,
                )

            try:
                # Load from URL or cache if already cached
                resolved_archive_file = cached_path(
                    archive_file,
                    cache_dir=cache_dir,
                    force_download=force_download,
                    proxies=proxies,
                    resume_download=resume_download,
                    local_files_only=local_files_only,
                )
                if resolved_archive_file is None:
                    raise EnvironmentError
            except EnvironmentError:
                msg = (
                    f"Can't load weights for '{pretrained_model_name_or_path}'. Make sure that:\n\n"
                    f"- '{pretrained_model_name_or_path}' is a correct model identifier listed on 'https://huggingface.co/models'\n\n"
                    f"- or '{pretrained_model_name_or_path}' is the correct path to a directory containing a file named one of {TF2_WEIGHTS_NAME}, {WEIGHTS_NAME}.\n\n"
                )
                raise EnvironmentError(msg)
            if resolved_archive_file == archive_file:
                logger.info("loading weights file {}".format(archive_file))
            else:
                logger.info("loading weights file {} from cache at {}".format(archive_file, resolved_archive_file))
        else:
            resolved_archive_file = None

        # Instantiate model.
        model = cls(config, *model_args, **model_kwargs)

        if from_pt:
            # Load from a PyTorch checkpoint
            return load_pytorch_checkpoint_in_tf2_model(model, resolved_archive_file, allow_missing_keys=True)

        model(model.dummy_inputs, training=False)  # build the network with dummy inputs

        assert os.path.isfile(resolved_archive_file), "Error retrieving file {}".format(resolved_archive_file)
        # 'by_name' allow us to do transfer learning by skipping/adding layers
        # see https://github.com/tensorflow/tensorflow/blob/00fad90125b18b80fe054de1055770cfb8fe4ba3/tensorflow/python/keras/engine/network.py#L1339-L1357
        try:
            model.load_weights(resolved_archive_file, by_name=True)
        except OSError:
            raise OSError(
                "Unable to load weights from h5 file. "
                "If you tried to load a TF 2.0 model from a PyTorch checkpoint, please set from_pt=True. "
            )

        model(model.dummy_inputs, training=False)  # Make sure restore ops are run

        # Check if the models are the same to output loading informations
        with h5py.File(resolved_archive_file, "r") as f:
            if "layer_names" not in f.attrs and "model_weights" in f:
                f = f["model_weights"]
            hdf5_layer_names = set(hdf5_format.load_attributes_from_hdf5_group(f, "layer_names"))
        model_layer_names = set(layer.name for layer in model.layers)
        missing_keys = list(model_layer_names - hdf5_layer_names)
        unexpected_keys = list(hdf5_layer_names - model_layer_names)
        error_msgs = []

        if len(missing_keys) > 0:
            logger.info(
                "Layers of {} not initialized from pretrained model: {}".format(model.__class__.__name__, missing_keys)
            )
        if len(unexpected_keys) > 0:
            logger.info(
                "Layers from pretrained model not used in {}: {}".format(model.__class__.__name__, unexpected_keys)
            )
        if len(error_msgs) > 0:
            raise RuntimeError(
                "Error(s) in loading weights for {}:\n\t{}".format(model.__class__.__name__, "\n\t".join(error_msgs))
            )
        if output_loading_info:
            loading_info = {"missing_keys": missing_keys, "unexpected_keys": unexpected_keys, "error_msgs": error_msgs}
            return model, loading_info

        return model

    def prepare_inputs_for_generation(self, inputs, **kwargs):
        return {"inputs": inputs}

    def _use_cache(self, outputs, use_cache):
        """During generation, decide whether to pass the `past` variable to the next forward pass."""
        if len(outputs) <= 1 or use_cache is False:
            return False
        if hasattr(self.config, "mem_len") and self.config.mem_len == 0:
            return False
        return True

    def generate(
        self,
        input_ids=None,
        max_length=None,
        min_length=None,
        do_sample=None,
        early_stopping=None,
        num_beams=None,
        temperature=None,
        top_k=None,
        top_p=None,
        repetition_penalty=None,
        bad_words_ids=None,
        bos_token_id=None,
        pad_token_id=None,
        eos_token_id=None,
        length_penalty=None,
        no_repeat_ngram_size=None,
        num_return_sequences=None,
        attention_mask=None,
        decoder_start_token_id=None,
        use_cache=None,
    ):
        r""" Generates sequences for models with a LM head. The method currently supports greedy or penalized greedy decoding, sampling with top-k or nucleus sampling
        and beam-search.

        Adapted in part from `Facebook's XLM beam search code`_.

        .. _`Facebook's XLM beam search code`:
           https://github.com/facebookresearch/XLM/blob/9e6f6814d17be4fe5b15f2e6c43eb2b2d76daeb4/src/model/transformer.py#L529


        Parameters:

            input_ids: (`optional`) `tf.Tensor` of `dtype=tf.int32` of shape `(batch_size, sequence_length)`
                The sequence used as a prompt for the generation. If `None` the method initializes
                it as an empty `tf.Tensor` of shape `(1,)`.

            max_length: (`optional`) int
                The max length of the sequence to be generated.  Between 1 and infinity. Default to 20.

            min_length: (`optional`) int
                The min length of the sequence to be generated.  Between 0 and infinity. Default to 0.
            do_sample: (`optional`) bool
                If set to `False` greedy decoding is used. Otherwise sampling is used. Defaults to `False` as defined in `configuration_utils.PretrainedConfig`.

            early_stopping: (`optional`) bool
                if set to `True` beam search is stopped when at least `num_beams` sentences finished per batch. Defaults to `False` as defined in `configuration_utils.PretrainedConfig`.

            num_beams: (`optional`) int
                Number of beams for beam search. Must be between 1 and infinity. 1 means no beam search. Default to 1.

            temperature: (`optional`) float
                The value used to module the next token probabilities. Must be strictely positive. Default to 1.0.

            top_k: (`optional`) int
                The number of highest probability vocabulary tokens to keep for top-k-filtering. Between 1 and infinity. Default to 50.

            top_p: (`optional`) float
                The cumulative probability of parameter highest probability vocabulary tokens to keep for nucleus sampling. Must be between 0 and 1. Default to 1.

            repetition_penalty: (`optional`) float
                The parameter for repetition penalty. Between 1.0 and infinity. 1.0 means no penalty. Default to 1.0.

            bos_token_id: (`optional`) int
                Beginning of sentence token if no prompt is provided. Default to specicic model bos_token_id or None if it does not exist.

            pad_token_id: (`optional`) int
                Pad token. Defaults to pad_token_id as defined in the models config.

            eos_token_id: (`optional`) int
                EOS token. Defaults to eos_token_id as defined in the models config.

            length_penalty: (`optional`) float
                Exponential penalty to the length. Default to 1.

            no_repeat_ngram_size: (`optional`) int
                If set to int > 0, all ngrams of size `no_repeat_ngram_size` can only occur once.

            bad_words_ids: (`optional`) list of lists of int
                `bad_words_ids` contains tokens that are not allowed to be generated. In order to get the tokens of the words that should not appear in the generated text, use `tokenizer.encode(bad_word, add_prefix_space=True)`.

            num_return_sequences: (`optional`) int
                The number of independently computed returned sequences for each element in the batch. Default to 1.

            attention_mask (`optional`) obj: `tf.Tensor` with `dtype=tf.int32` of same shape as `input_ids`
                Mask to avoid performing attention on padding token indices.
                Mask values selected in ``[0, 1]``:
                ``1`` for tokens that are NOT MASKED, ``0`` for MASKED tokens.
                Defaults to `None`.

                `What are attention masks? <../glossary.html#attention-mask>`__

            decoder_start_token_id=None: (`optional`) int
                If an encoder-decoder model starts decoding with a different token than BOS.
                Defaults to `None` and is changed to `BOS` later.

            use_cache: (`optional`) bool
                If `use_cache` is True, past key values are used to speed up decoding if applicable to model. Defaults to `True`.

        Return:

            output: `tf.Tensor` of `dtype=tf.int32` shape `(batch_size * num_return_sequences, sequence_length)`
                sequence_length is either equal to max_length or shorter if all batches finished early due to the `eos_token_id`

        Examples::

            tokenizer = AutoTokenizer.from_pretrained('distilgpt2')   # Initialize tokenizer
            model = TFAutoModelWithLMHead.from_pretrained('distilgpt2')    # Download model and configuration from S3 and cache.
            outputs = model.generate(max_length=40)  # do greedy decoding
            print('Generated: {}'.format(tokenizer.decode(outputs[0], skip_special_tokens=True)))

            tokenizer = AutoTokenizer.from_pretrained('openai-gpt')   # Initialize tokenizer
            model = TFAutoModelWithLMHead.from_pretrained('openai-gpt')    # Download model and configuration from S3 and cache.
            input_context = 'The dog'
            input_ids = tokenizer.encode(input_context, return_tensors='tf')  # encode input context
            outputs = model.generate(input_ids=input_ids, num_beams=5, num_return_sequences=3, temperature=1.5)  # generate 3 independent sequences using beam search decoding (5 beams) with sampling from initial context 'The dog'
            for i in range(3): #  3 output sequences were generated
                print('Generated {}: {}'.format(i, tokenizer.decode(outputs[i], skip_special_tokens=True)))

            tokenizer = AutoTokenizer.from_pretrained('distilgpt2')   # Initialize tokenizer
            model = TFAutoModelWithLMHead.from_pretrained('distilgpt2')    # Download model and configuration from S3 and cache.
            input_context = 'The dog'
            input_ids = tokenizer.encode(input_context, return_tensors='tf')  # encode input context
            outputs = model.generate(input_ids=input_ids, max_length=40, temperature=0.7, num_return_sequences=3)  # 3 generate sequences using by sampling
            for i in range(3): #  3 output sequences were generated
                print('Generated {}: {}'.format(i, tokenizer.decode(outputs[i], skip_special_tokens=True)))

            tokenizer = AutoTokenizer.from_pretrained('ctrl')   # Initialize tokenizer
            model = TFAutoModelWithLMHead.from_pretrained('ctrl')    # Download model and configuration from S3 and cache.
            input_context = 'Legal My neighbor is'  # "Legal" is one of the control codes for ctrl
            input_ids = tokenizer.encode(input_context, return_tensors='tf')  # encode input context
            outputs = model.generate(input_ids=input_ids, max_length=50, temperature=0.7, repetition_penalty=1.2)  # generate sequences
            print('Generated: {}'.format(tokenizer.decode(outputs[0], skip_special_tokens=True)))

            tokenizer = AutoTokenizer.from_pretrained('gpt2')   # Initialize tokenizer
            model = TFAutoModelWithLMHead.from_pretrained('gpt2')    # Download model and configuration from S3 and cache.
            input_context = 'My cute dog'  # "Legal" is one of the control codes for ctrl
            bad_words_ids = [tokenizer.encode(bad_word, add_prefix_space=True) for bad_word in ['idiot', 'stupid', 'shut up']]
            input_ids = tokenizer.encode(input_context, return_tensors='tf')  # encode input context
            outputs = model.generate(input_ids=input_ids, max_length=100, do_sample=True, bad_words_ids=bad_words_ids)  # generate sequences without allowing bad_words to be generated
        """

        # We cannot generate if the model does not have a LM head
        if self.get_output_embeddings() is None:
            raise AttributeError(
                "You tried to generate sequences with a model that does not have a LM Head."
                "Please use another model class (e.g. `TFOpenAIGPTLMHeadModel`, `TFXLNetLMHeadModel`, `TFGPT2LMHeadModel`, `TFCTRLLMHeadModel`, `TFT5ForConditionalGeneration`, `TFTransfoXLLMHeadModel`)"
            )

        max_length = max_length if max_length is not None else self.config.max_length
        min_length = min_length if min_length is not None else self.config.min_length
        do_sample = do_sample if do_sample is not None else self.config.do_sample
        early_stopping = early_stopping if early_stopping is not None else self.config.early_stopping
        use_cache = use_cache if use_cache is not None else self.config.use_cache
        num_beams = num_beams if num_beams is not None else self.config.num_beams
        temperature = temperature if temperature is not None else self.config.temperature
        top_k = top_k if top_k is not None else self.config.top_k
        top_p = top_p if top_p is not None else self.config.top_p
        repetition_penalty = repetition_penalty if repetition_penalty is not None else self.config.repetition_penalty
        bos_token_id = bos_token_id if bos_token_id is not None else self.config.bos_token_id
        pad_token_id = pad_token_id if pad_token_id is not None else self.config.pad_token_id
        eos_token_id = eos_token_id if eos_token_id is not None else self.config.eos_token_id
        length_penalty = length_penalty if length_penalty is not None else self.config.length_penalty
        no_repeat_ngram_size = (
            no_repeat_ngram_size if no_repeat_ngram_size is not None else self.config.no_repeat_ngram_size
        )
        bad_words_ids = bad_words_ids if bad_words_ids is not None else self.config.bad_words_ids
        num_return_sequences = (
            num_return_sequences if num_return_sequences is not None else self.config.num_return_sequences
        )
        decoder_start_token_id = (
            decoder_start_token_id if decoder_start_token_id is not None else self.config.decoder_start_token_id
        )

        if input_ids is not None:
            batch_size = shape_list(input_ids)[0]  # overriden by the input batch_size
        else:
            batch_size = 1

        assert isinstance(max_length, int) and max_length > 0, "`max_length` should be a strictely positive integer."
        assert isinstance(min_length, int) and min_length >= 0, "`min_length` should be a positive integer."
        assert isinstance(do_sample, bool), "`do_sample` should be a boolean."
        assert isinstance(early_stopping, bool), "`early_stopping` should be a boolean."
        assert isinstance(use_cache, bool), "`use_cache` should be a boolean."
        assert isinstance(num_beams, int) and num_beams > 0, "`num_beams` should be a strictely positive integer."
        assert temperature > 0, "`temperature` should be strictely positive."
        assert isinstance(top_k, int) and top_k >= 0, "`top_k` should be a positive integer."
        assert 0 <= top_p <= 1, "`top_p` should be between 0 and 1."
        assert repetition_penalty >= 1.0, "`repetition_penalty` should be >= 1."
        assert input_ids is not None or (
            isinstance(bos_token_id, int) and bos_token_id >= 0
        ), "If input_ids is not defined, `bos_token_id` should be a positive integer."
        assert pad_token_id is None or (
            isinstance(pad_token_id, int) and (pad_token_id >= 0)
        ), "`pad_token_id` should be a positive integer."
        assert (eos_token_id is None) or (
            isinstance(eos_token_id, int) and (eos_token_id >= 0)
        ), "`eos_token_id` should be a positive integer."
        assert length_penalty > 0, "`length_penalty` should be strictely positive."
        assert (
            isinstance(num_return_sequences, int) and num_return_sequences > 0
        ), "`num_return_sequences` should be a strictely positive integer."
        assert (
            bad_words_ids is None or isinstance(bad_words_ids, list) and isinstance(bad_words_ids[0], list)
        ), "`bad_words_ids` is either `None` or a list of lists of tokens that should not be generated"

        if input_ids is None:
            assert isinstance(bos_token_id, int) and bos_token_id >= 0, (
                "you should either supply a context to complete as `input_ids` input "
                "or a `bos_token_id` (integer >= 0) as a first token to start the generation."
            )
            input_ids = tf.fill((batch_size, 1), bos_token_id)
        else:
            assert len(shape_list(input_ids)) == 2, "Input prompt should be of shape (batch_size, sequence length)."

        # not allow to duplicate outputs when greedy decoding
        if do_sample is False:
            if num_beams == 1:
                # no_beam_search greedy generation conditions
                assert (
                    num_return_sequences == 1
                ), "Greedy decoding will always produce the same output for num_beams == 1 and num_return_sequences > 1. Please set num_return_sequences = 1"

            else:
                # beam_search greedy generation conditions
                assert (
                    num_beams >= num_return_sequences
                ), "Greedy beam search decoding cannot return more sequences than it has beams. Please set num_beams >= num_return_sequences"

        # create attention mask if necessary
        # TODO (PVP): this should later be handled by the forward fn() in each model in the future see PR 3140
        if (attention_mask is None) and (pad_token_id is not None) and (pad_token_id in input_ids.numpy()):
            attention_mask = tf.cast(tf.math.not_equal(input_ids, pad_token_id), dtype=tf.int32)
        elif attention_mask is None:
            attention_mask = tf.ones_like(input_ids)

        if pad_token_id is None and eos_token_id is not None:
            logger.warning(
                "Setting `pad_token_id` to {} (first `eos_token_id`) to generate sequence".format(eos_token_id)
            )
            pad_token_id = eos_token_id

        # current position and vocab size
        cur_len = shape_list(input_ids)[1]
        vocab_size = self.config.vocab_size

        # set effective batch size and effective batch multiplier according to do_sample
        if do_sample:
            effective_batch_size = batch_size * num_return_sequences
            effective_batch_mult = num_return_sequences
        else:
            effective_batch_size = batch_size
            effective_batch_mult = 1

        if self.config.is_encoder_decoder:
            if decoder_start_token_id is None:
                decoder_start_token_id = bos_token_id

            assert (
                decoder_start_token_id is not None
            ), "decoder_start_token_id or bos_token_id has to be defined for encoder-decoder generation"
            assert hasattr(self, "get_encoder"), "{} should have a 'get_encoder' function defined".format(self)
            assert callable(self.get_encoder), "{} should be a method".format(self.get_encoder)

            # get encoder and store encoder outputs
            encoder = self.get_encoder()

            encoder_outputs = encoder(input_ids, attention_mask=attention_mask)

        # Expand input ids if num_beams > 1 or num_return_sequences > 1
        if num_return_sequences > 1 or num_beams > 1:
            input_ids_len = shape_list(input_ids)[-1]
            input_ids = tf.broadcast_to(
                tf.expand_dims(input_ids, 1), (batch_size, effective_batch_mult * num_beams, input_ids_len)
            )
            attention_mask = tf.broadcast_to(
                tf.expand_dims(attention_mask, 1), (batch_size, effective_batch_mult * num_beams, input_ids_len)
            )
            input_ids = tf.reshape(
                input_ids, (effective_batch_size * num_beams, input_ids_len)
            )  # shape: (batch_size * num_return_sequences * num_beams, cur_len)
            attention_mask = tf.reshape(
                attention_mask, (effective_batch_size * num_beams, input_ids_len)
            )  # shape: (batch_size * num_return_sequences * num_beams, cur_len)

        if self.config.is_encoder_decoder:

            # create empty decoder_input_ids
            input_ids = tf.ones((effective_batch_size * num_beams, 1), dtype=tf.int32,) * decoder_start_token_id
            cur_len = 1

            assert (
                batch_size == encoder_outputs[0].shape[0]
            ), f"expected encoder_outputs[0] to have 1st dimension bs={batch_size}, got {encoder_outputs[0].shape[0]} "

            # expand batch_idx to assign correct encoder output for expanded input_ids (due to num_beams > 1 and num_return_sequences > 1)
            expanded_batch_idxs = tf.reshape(
                tf.repeat(tf.expand_dims(tf.range(batch_size), -1), repeats=num_beams * effective_batch_mult, axis=1),
                shape=(-1,),
            )
            # expand encoder_outputs
            encoder_outputs = (tf.gather(encoder_outputs[0], expanded_batch_idxs, axis=0), *encoder_outputs[1:])

        else:
            encoder_outputs = None
            cur_len = shape_list(input_ids)[-1]

        if num_beams > 1:
            output = self._generate_beam_search(
                input_ids,
                cur_len=cur_len,
                max_length=max_length,
                min_length=min_length,
                do_sample=do_sample,
                early_stopping=early_stopping,
                temperature=temperature,
                top_k=top_k,
                top_p=top_p,
                repetition_penalty=repetition_penalty,
                no_repeat_ngram_size=no_repeat_ngram_size,
                bad_words_ids=bad_words_ids,
                bos_token_id=bos_token_id,
                pad_token_id=pad_token_id,
                eos_token_id=eos_token_id,
                decoder_start_token_id=decoder_start_token_id,
                batch_size=effective_batch_size,
                num_return_sequences=num_return_sequences,
                length_penalty=length_penalty,
                num_beams=num_beams,
                vocab_size=vocab_size,
                encoder_outputs=encoder_outputs,
                attention_mask=attention_mask,
                use_cache=use_cache,
            )
        else:
            output = self._generate_no_beam_search(
                input_ids,
                cur_len=cur_len,
                max_length=max_length,
                min_length=min_length,
                do_sample=do_sample,
                temperature=temperature,
                top_k=top_k,
                top_p=top_p,
                repetition_penalty=repetition_penalty,
                no_repeat_ngram_size=no_repeat_ngram_size,
                bad_words_ids=bad_words_ids,
                bos_token_id=bos_token_id,
                pad_token_id=pad_token_id,
                eos_token_id=eos_token_id,
                decoder_start_token_id=decoder_start_token_id,
                batch_size=effective_batch_size,
                vocab_size=vocab_size,
                encoder_outputs=encoder_outputs,
                attention_mask=attention_mask,
                use_cache=use_cache,
            )

        return output

    def _generate_no_beam_search(
        self,
        input_ids,
        cur_len,
        max_length,
        min_length,
        do_sample,
        temperature,
        top_k,
        top_p,
        repetition_penalty,
        no_repeat_ngram_size,
        bad_words_ids,
        bos_token_id,
        pad_token_id,
        eos_token_id,
        decoder_start_token_id,
        batch_size,
        vocab_size,
        encoder_outputs,
        attention_mask,
        use_cache,
    ):
        """ Generate sequences for each example without beam search (num_beams == 1).
            All returned sequence are generated independantly.
        """

        # length of generated sentences / unfinished sentences
        unfinished_sents = tf.ones_like(input_ids[:, 0])
        sent_lengths = tf.ones_like(input_ids[:, 0]) * max_length

        past = encoder_outputs  # defined for encoder-decoder models, None for decoder-only models

        while cur_len < max_length:
            model_inputs = self.prepare_inputs_for_generation(
                input_ids, past=past, attention_mask=attention_mask, use_cache=use_cache
            )
            outputs = self(**model_inputs)
            next_token_logits = outputs[0][:, -1, :]

            # if model has past, then set the past variable to speed up decoding
            if self._use_cache(outputs, use_cache):
                past = outputs[1]

            # repetition penalty from CTRL paper (https://arxiv.org/abs/1909.05858)
            if repetition_penalty != 1.0:
                next_token_logits_penalties = _create_next_token_logits_penalties(
                    input_ids, next_token_logits, repetition_penalty
                )
                next_token_logits = tf.math.multiply(next_token_logits, next_token_logits_penalties)

            if no_repeat_ngram_size > 0:
                # calculate a list of banned tokens to prevent repetitively generating the same ngrams
                # from fairseq: https://github.com/pytorch/fairseq/blob/a07cb6f40480928c9e0548b737aadd36ee66ac76/fairseq/sequence_generator.py#L345
                banned_tokens = calc_banned_ngram_tokens(input_ids, batch_size, no_repeat_ngram_size, cur_len)
                # create banned_tokens boolean mask
                banned_tokens_indices_mask = []
                for banned_tokens_slice in banned_tokens:
                    banned_tokens_indices_mask.append(
                        [True if token in banned_tokens_slice else False for token in range(vocab_size)]
                    )

                next_token_logits = set_tensor_by_indices_to_value(
                    next_token_logits, tf.convert_to_tensor(banned_tokens_indices_mask, dtype=tf.bool), -float("inf")
                )

            if bad_words_ids is not None:
                # calculate a list of banned tokens according to bad words
                banned_tokens = calc_banned_bad_words_ids(input_ids, bad_words_ids)

                banned_tokens_indices_mask = []
                for banned_tokens_slice in banned_tokens:
                    banned_tokens_indices_mask.append(
                        [True if token in banned_tokens_slice else False for token in range(vocab_size)]
                    )

                next_token_logits = set_tensor_by_indices_to_value(
                    next_token_logits, tf.convert_to_tensor(banned_tokens_indices_mask, dtype=tf.bool), -float("inf")
                )

            # set eos token prob to zero if min_length is not reached
            if eos_token_id is not None and cur_len < min_length:
                # create eos_token_id boolean mask
                is_token_logit_eos_token = tf.convert_to_tensor(
                    [True if token is eos_token_id else False for token in range(vocab_size)], dtype=tf.bool
                )
                eos_token_indices_mask = tf.broadcast_to(is_token_logit_eos_token, [batch_size, vocab_size])

                next_token_logits = set_tensor_by_indices_to_value(
                    next_token_logits, eos_token_indices_mask, -float("inf")
                )

            if do_sample:
                # Temperature (higher temperature => more likely to sample low probability tokens)
                if temperature != 1.0:
                    next_token_logits = next_token_logits / temperature
                # Top-p/top-k filtering
                next_token_logits = tf_top_k_top_p_filtering(next_token_logits, top_k=top_k, top_p=top_p)
                # Sample
                next_token = tf.squeeze(
                    tf.random.categorical(next_token_logits, dtype=tf.int32, num_samples=1), axis=1
                )
            else:
                # Greedy decoding
                next_token = tf.math.argmax(next_token_logits, axis=-1, output_type=tf.int32)

            # update generations and finished sentences
            if eos_token_id is not None:
                # pad finished sentences if eos_token_id exist
                tokens_to_add = next_token * unfinished_sents + (pad_token_id) * (1 - unfinished_sents)
            else:
                tokens_to_add = next_token

            # add token and increase length by one
            input_ids = tf.concat([input_ids, tf.expand_dims(tokens_to_add, -1)], 1)
            cur_len = cur_len + 1

            if eos_token_id is not None:
                eos_in_sents = tokens_to_add == eos_token_id
                # if sentence is unfinished and the token to add is eos, sent_lengths is filled with current length
                is_sents_unfinished_and_token_to_add_is_eos = tf.math.multiply(
                    unfinished_sents, tf.cast(eos_in_sents, tf.int32)
                )
                sent_lengths = (
                    sent_lengths * (1 - is_sents_unfinished_and_token_to_add_is_eos)
                    + cur_len * is_sents_unfinished_and_token_to_add_is_eos
                )

                # unfinished_sents is set to zero if eos in sentence
                unfinished_sents -= is_sents_unfinished_and_token_to_add_is_eos

            # stop when there is a </s> in each sentence, or if we exceed the maximul length
            if tf.math.reduce_max(unfinished_sents) == 0:
                break

            # extend attention_mask for new generated input if only decoder
            if self.config.is_encoder_decoder is False:
                attention_mask = tf.concat(
                    [attention_mask, tf.ones((shape_list(attention_mask)[0], 1), dtype=tf.int32)], axis=-1
                )

        # if there are different sentences lengths in the batch, some batches have to be padded
        min_sent_length = tf.math.reduce_min(sent_lengths)
        max_sent_length = tf.math.reduce_max(sent_lengths)
        if min_sent_length != max_sent_length:
            assert pad_token_id is not None, "`Pad_token_id` has to be defined if batches have different lengths"
            # finished sents are filled with pad_token
            padding = tf.ones([batch_size, max_sent_length.numpy()], dtype=tf.int32) * pad_token_id

            # create length masks for tf.where operation
            broad_casted_sent_lengths = tf.broadcast_to(
                tf.expand_dims(sent_lengths, -1), [batch_size, max_sent_length]
            )
            broad_casted_range = tf.transpose(
                tf.broadcast_to(tf.expand_dims(tf.range(max_sent_length), -1), [max_sent_length, batch_size])
            )

            decoded = tf.where(broad_casted_range < broad_casted_sent_lengths, input_ids, padding)
        else:
            decoded = input_ids

        return decoded

    def _generate_beam_search(
        self,
        input_ids,
        cur_len,
        max_length,
        min_length,
        do_sample,
        early_stopping,
        temperature,
        top_k,
        top_p,
        repetition_penalty,
        no_repeat_ngram_size,
        bad_words_ids,
        bos_token_id,
        pad_token_id,
        decoder_start_token_id,
        eos_token_id,
        batch_size,
        num_return_sequences,
        length_penalty,
        num_beams,
        vocab_size,
        encoder_outputs,
        attention_mask,
        use_cache,
    ):
        """ Generate sequences for each example with beam search.
        """

        # generated hypotheses
        generated_hyps = [
            BeamHypotheses(num_beams, max_length, length_penalty, early_stopping=early_stopping)
            for _ in range(batch_size)
        ]

        # for greedy decoding it is made sure that only tokens of the first beam are considered to avoid sampling the exact same tokens three times
        if do_sample is False:
            beam_scores_begin = tf.zeros((batch_size, 1), dtype=tf.float32)
            beam_scores_end = tf.ones((batch_size, num_beams - 1), dtype=tf.float32) * (-1e9)
            beam_scores = tf.concat([beam_scores_begin, beam_scores_end], -1)
        else:
            beam_scores = tf.zeros((batch_size, num_beams), dtype=tf.float32)

        beam_scores = tf.reshape(beam_scores, (batch_size * num_beams,))

        # cache compute states
        past = encoder_outputs

        # done sentences
        done = [False for _ in range(batch_size)]

        while cur_len < max_length:
            model_inputs = self.prepare_inputs_for_generation(
                input_ids, past=past, attention_mask=attention_mask, use_cache=use_cache
            )
            outputs = self(**model_inputs)  # (batch_size * num_beams, cur_len, vocab_size)
            next_token_logits = outputs[0][:, -1, :]  # (batch_size * num_beams, vocab_size)

            # if model has past, then set the past variable to speed up decoding
            if self._use_cache(outputs, use_cache):
                past = outputs[1]

            # repetition penalty (from CTRL paper https://arxiv.org/abs/1909.05858)
            if repetition_penalty != 1.0:
                next_token_logits_penalties = _create_next_token_logits_penalties(
                    input_ids, next_token_logits, repetition_penalty
                )
                next_token_logits = tf.math.multiply(next_token_logits, next_token_logits_penalties)

            # Temperature (higher temperature => more likely to sample low probability tokens)
            if temperature != 1.0:
                next_token_logits = next_token_logits / temperature

            #             calculate log softmax score
            scores = tf.nn.log_softmax(next_token_logits, axis=-1)  # (batch_size * num_beams, vocab_size)

            # set eos token prob to zero if min_length is not reached
            if eos_token_id is not None and cur_len < min_length:
                # create eos_token_id boolean mask
                num_batch_hypotheses = batch_size * num_beams

                is_token_logit_eos_token = tf.convert_to_tensor(
                    [True if token is eos_token_id else False for token in range(vocab_size)], dtype=tf.bool
                )
                eos_token_indices_mask = tf.broadcast_to(is_token_logit_eos_token, [num_batch_hypotheses, vocab_size])

                scores = set_tensor_by_indices_to_value(scores, eos_token_indices_mask, -float("inf"))

            if no_repeat_ngram_size > 0:
                # calculate a list of banned tokens to prevent repetitively generating the same ngrams
                # from fairseq: https://github.com/pytorch/fairseq/blob/a07cb6f40480928c9e0548b737aadd36ee66ac76/fairseq/sequence_generator.py#L345
                num_batch_hypotheses = batch_size * num_beams
                banned_tokens = calc_banned_ngram_tokens(
                    input_ids, num_batch_hypotheses, no_repeat_ngram_size, cur_len
                )
                # create banned_tokens boolean mask
                banned_tokens_indices_mask = []
                for banned_tokens_slice in banned_tokens:
                    banned_tokens_indices_mask.append(
                        [True if token in banned_tokens_slice else False for token in range(vocab_size)]
                    )

                scores = set_tensor_by_indices_to_value(
                    scores, tf.convert_to_tensor(banned_tokens_indices_mask, dtype=tf.bool), -float("inf")
                )

            if bad_words_ids is not None:
                # calculate a list of banned tokens according to bad words
                banned_tokens = calc_banned_bad_words_ids(input_ids, bad_words_ids)

                banned_tokens_indices_mask = []
                for banned_tokens_slice in banned_tokens:
                    banned_tokens_indices_mask.append(
                        [True if token in banned_tokens_slice else False for token in range(vocab_size)]
                    )

                scores = set_tensor_by_indices_to_value(
                    scores, tf.convert_to_tensor(banned_tokens_indices_mask, dtype=tf.bool), -float("inf")
                )

            assert shape_list(scores) == [batch_size * num_beams, vocab_size]

            if do_sample:
                _scores = scores + tf.broadcast_to(
                    beam_scores[:, None], (batch_size * num_beams, vocab_size)
                )  # (batch_size * num_beams, vocab_size)

                # Top-p/top-k filtering
                _scores = tf_top_k_top_p_filtering(
                    _scores, top_k=top_k, top_p=top_p, min_tokens_to_keep=2
                )  # (batch_size * num_beams, vocab_size)
                # Sample 2 next tokens for each beam (so we have some spare tokens and match output of greedy beam search)
                _scores = tf.reshape(_scores, (batch_size, num_beams * vocab_size))

                next_tokens = sample_without_replacement(
                    _scores, num_samples=2 * num_beams
                )  # (batch_size, 2 * num_beams)
                # Compute next scores
                next_scores = tf.gather(_scores, next_tokens, batch_dims=1)  # (batch_size, 2 * num_beams)

                # sort the sampled vector to make sure that the first num_beams samples are the best
                next_scores_indices = tf.argsort(next_scores, direction="DESCENDING", axis=1)
                next_scores = tf.gather(next_scores, next_scores_indices, batch_dims=1)  # (batch_size, num_beams * 2)
                next_tokens = tf.gather(next_tokens, next_scores_indices, batch_dims=1)  # (batch_size, num_beams * 2)
            else:
                # Add the log prob of the new beams to the log prob of the beginning of the sequence (sum of logs == log of the product)
                next_scores = scores + tf.broadcast_to(
                    beam_scores[:, None], (batch_size * num_beams, vocab_size)
                )  # (batch_size * num_beams, vocab_size)

                # re-organize to group the beam together (we are keeping top hypothesis accross beams)
                next_scores = tf.reshape(
                    next_scores, (batch_size, num_beams * vocab_size)
                )  # (batch_size, num_beams * vocab_size)

                next_scores, next_tokens = tf.math.top_k(next_scores, k=2 * num_beams, sorted=True)

            assert shape_list(next_scores) == shape_list(next_tokens) == [batch_size, 2 * num_beams]

            # next batch beam content
            next_batch_beam = []

            # for each sentence
            for batch_idx in range(batch_size):

                # if we are done with this sentence
                if done[batch_idx]:
                    assert (
                        len(generated_hyps[batch_idx]) >= num_beams
                    ), "Batch can only be done if at least {} beams have been generated".format(num_beams)
                    assert (
                        eos_token_id is not None and pad_token_id is not None
                    ), "generated beams >= num_beams -> eos_token_id and pad_token have to be defined"
                    next_batch_beam.extend([(0, pad_token_id, 0)] * num_beams)  # pad the batch
                    continue

                # next sentence beam content
                next_sent_beam = []

                # next tokens for this sentence
                for beam_token_rank, (beam_token_id, beam_token_score) in enumerate(
                    zip(next_tokens[batch_idx], next_scores[batch_idx])
                ):
                    # get beam and token IDs
                    beam_id = beam_token_id // vocab_size
                    token_id = beam_token_id % vocab_size

                    effective_beam_id = batch_idx * num_beams + beam_id
                    # add to generated hypotheses if end of sentence or last iteration
                    if (eos_token_id is not None) and (token_id.numpy() == eos_token_id):
                        # if beam_token does not belong to top num_beams tokens, it should not be added
                        is_beam_token_worse_than_top_num_beams = beam_token_rank >= num_beams
                        if is_beam_token_worse_than_top_num_beams:
                            continue
                        generated_hyps[batch_idx].add(
                            tf.identity(input_ids[effective_beam_id]), beam_token_score.numpy()
                        )
                    else:
                        # add next predicted token if it is not eos_token
                        next_sent_beam.append((beam_token_score, token_id, effective_beam_id))

                    # the beam for next step is full
                    if len(next_sent_beam) == num_beams:
                        break

                # Check if we are done so that we can save a pad step if all(done)
                done[batch_idx] = done[batch_idx] or generated_hyps[batch_idx].is_done(
                    tf.reduce_max(next_scores[batch_idx]).numpy(), cur_len
                )

                # update next beam content
                assert len(next_sent_beam) == num_beams, "Beam should always be full"
                next_batch_beam.extend(next_sent_beam)
                assert len(next_batch_beam) == num_beams * (batch_idx + 1)

            # stop when we are done with each sentence
            if all(done):
                break

            # sanity check / prepare next batch
            assert len(next_batch_beam) == batch_size * num_beams
            beam_scores = tf.convert_to_tensor([x[0] for x in next_batch_beam], dtype=tf.float32)
            beam_tokens = tf.convert_to_tensor([x[1] for x in next_batch_beam], dtype=tf.int32)
            beam_idx = tf.convert_to_tensor([x[2] for x in next_batch_beam], dtype=tf.int32)

            # re-order batch and update current length
            input_ids = tf.stack([tf.identity(input_ids[x, :]) for x in beam_idx])
            input_ids = tf.concat([input_ids, tf.expand_dims(beam_tokens, 1)], axis=-1)
            cur_len = cur_len + 1

            # re-order internal states
            if past is not None:
                past = self._reorder_cache(past, beam_idx)

            # extend attention_mask for new generated input if only decoder
            if self.config.is_encoder_decoder is False:
                attention_mask = tf.concat(
                    [attention_mask, tf.ones((shape_list(attention_mask)[0], 1), dtype=tf.int32)], axis=-1
                )

        # finalize all open beam hypotheses and end to generated hypotheses
        for batch_idx in range(batch_size):
            # Add all open beam hypothesis to generated_hyps
            if done[batch_idx]:
                continue
            # test that beam scores match previously calculated scores if not eos and batch_idx not done
            if eos_token_id is not None and all(
                (token_id % vocab_size).numpy().item() != eos_token_id for token_id in next_tokens[batch_idx]
            ):
                assert tf.reduce_all(
                    next_scores[batch_idx, :num_beams] == tf.reshape(beam_scores, (batch_size, num_beams))[batch_idx]
                ), "If batch_idx is not done, final next scores: {} have to equal to accumulated beam_scores: {}".format(
                    next_scores[:, :num_beams][batch_idx], tf.reshape(beam_scores, (batch_size, num_beams))[batch_idx]
                )

            # need to add best num_beams hypotheses to generated hyps
            for beam_id in range(num_beams):
                effective_beam_id = batch_idx * num_beams + beam_id
                final_score = beam_scores[effective_beam_id].numpy().item()
                final_tokens = input_ids[effective_beam_id]
                generated_hyps[batch_idx].add(final_tokens, final_score)

        # depending on whether greedy generation is wanted or not define different output_batch_size and output_num_return_sequences_per_batch
        output_batch_size = batch_size if do_sample else batch_size * num_return_sequences
        output_num_return_sequences_per_batch = 1 if do_sample else num_return_sequences

        # select the best hypotheses
        sent_lengths_list = []
        best = []

        # retrieve best hypotheses
        for i, hypotheses in enumerate(generated_hyps):
            sorted_hyps = sorted(hypotheses.beams, key=lambda x: x[0])
            for j in range(output_num_return_sequences_per_batch):
                best_hyp = sorted_hyps.pop()[1]
                sent_lengths_list.append(len(best_hyp))
                best.append(best_hyp)
        assert output_batch_size == len(best), "Output batch size {} must match output beam hypotheses {}".format(
            output_batch_size, len(best)
        )

        sent_lengths = tf.convert_to_tensor(sent_lengths_list, dtype=tf.int32)

        # shorter batches are filled with pad_token
        if tf.reduce_min(sent_lengths).numpy() != tf.reduce_max(sent_lengths).numpy():
            assert pad_token_id is not None, "`Pad_token_id` has to be defined"
            sent_max_len = min(tf.reduce_max(sent_lengths).numpy() + 1, max_length)
            decoded_list = []

            # fill with hypothesis and eos_token_id if necessary
            for i, hypo in enumerate(best):
                assert sent_lengths[i] == shape_list(hypo)[0]
                # if sent_length is max_len do not pad
                if sent_lengths[i] == sent_max_len:
                    decoded_slice = hypo
                else:
                    # else pad to sent_max_len
                    num_pad_tokens = sent_max_len - sent_lengths[i]
                    padding = pad_token_id * tf.ones((num_pad_tokens,), dtype=tf.int32)
                    decoded_slice = tf.concat([hypo, padding], axis=-1)

                    # finish sentence with EOS token
                    if sent_lengths[i] < max_length:
                        decoded_slice = tf.where(
                            tf.range(sent_max_len, dtype=tf.int32) == sent_lengths[i],
                            eos_token_id * tf.ones((sent_max_len,), dtype=tf.int32),
                            decoded_slice,
                        )
                # add to list
                decoded_list.append(decoded_slice)

            decoded = tf.stack(decoded_list)
        else:
            # none of the hypotheses have an eos_token
            assert (len(hypo) == max_length for hypo in best)
            decoded = tf.stack(best)

        return decoded

    @staticmethod
    def _reorder_cache(past, beam_idx):
        return tuple(tf.gather(layer_past, beam_idx, axis=1) for layer_past in past)


def _create_next_token_logits_penalties(input_ids, logits, repetition_penalty):
    # create logit penalties for already seen input_ids
    token_penalties = np.ones(shape_list(logits))
    prev_input_ids = [np.unique(input_id) for input_id in input_ids.numpy()]
    for i, prev_input_id in enumerate(prev_input_ids):
        logit_penalized = logits[i].numpy()[prev_input_id]
        logit_penalties = np.zeros(logit_penalized.shape)
        # if previous logit score is < 0 then multiply repetition penalty else divide
        logit_penalties[logit_penalized < 0] = repetition_penalty
        logit_penalties[logit_penalized > 0] = 1 / repetition_penalty
        np.put(token_penalties[i], prev_input_id, logit_penalties)
    return tf.convert_to_tensor(token_penalties, dtype=tf.float32)


def calc_banned_ngram_tokens(prev_input_ids, num_hypos, no_repeat_ngram_size, cur_len):
    # Copied from fairseq for no_repeat_ngram in beam_search"""
    if cur_len + 1 < no_repeat_ngram_size:
        # return no banned tokens if we haven't generated no_repeat_ngram_size tokens yet
        return [[] for _ in range(num_hypos)]
    generated_ngrams = [{} for _ in range(num_hypos)]
    for idx in range(num_hypos):
        gen_tokens = prev_input_ids[idx].numpy().tolist()
        generated_ngram = generated_ngrams[idx]
        for ngram in zip(*[gen_tokens[i:] for i in range(no_repeat_ngram_size)]):
            prev_ngram_tuple = tuple(ngram[:-1])
            generated_ngram[prev_ngram_tuple] = generated_ngram.get(prev_ngram_tuple, []) + [ngram[-1]]

    def _get_generated_ngrams(hypo_idx):
        # Before decoding the next token, prevent decoding of ngrams that have already appeared
        start_idx = cur_len + 1 - no_repeat_ngram_size
        ngram_idx = tuple(prev_input_ids[hypo_idx, start_idx:cur_len].numpy().tolist())
        return generated_ngrams[hypo_idx].get(ngram_idx, [])

    banned_tokens = [_get_generated_ngrams(hypo_idx) for hypo_idx in range(num_hypos)]
    return banned_tokens


def calc_banned_bad_words_ids(prev_input_ids, bad_words_ids):
    banned_tokens = []

    def _tokens_match(prev_tokens, tokens):
        if len(tokens) == 0:
            # if bad word tokens is just one token always ban it
            return True
        if len(tokens) > len(prev_input_ids):
            # if bad word tokens are longer then prev input_ids they can't be equal
            return False

        if prev_tokens[-len(tokens) :] == tokens:
            # if tokens match
            return True
        else:
            return False

    for prev_input_ids_slice in prev_input_ids:
        banned_tokens_slice = []

        for banned_token_seq in bad_words_ids:
            assert len(banned_token_seq) > 0, "Banned words token sequences {} cannot have an empty list".format(
                bad_words_ids
            )

            if _tokens_match(prev_input_ids_slice.numpy().tolist(), banned_token_seq[:-1]) is False:
                # if tokens do not match continue
                continue

            banned_tokens_slice.append(banned_token_seq[-1])

        banned_tokens.append(banned_tokens_slice)

    return banned_tokens


def tf_top_k_top_p_filtering(logits, top_k=0, top_p=1.0, filter_value=-float("Inf"), min_tokens_to_keep=1):
    """ Filter a distribution of logits using top-k and/or nucleus (top-p) filtering
        Args:
            logits: logits distribution shape (batch size, vocabulary size)
            if top_k > 0: keep only top k tokens with highest probability (top-k filtering).
            if top_p < 1.0: keep the top tokens with cumulative probability >= top_p (nucleus filtering).
                Nucleus filtering is described in Holtzman et al. (http://arxiv.org/abs/1904.09751)
            Make sure we keep at least min_tokens_to_keep per batch example in the output
        From: https://gist.github.com/thomwolf/1a5a29f6962089e871b94cbd09daf317
    """
    logits_shape = shape_list(logits)

    if top_k > 0:
        top_k = min(max(top_k, min_tokens_to_keep), logits_shape[-1])  # Safety check
        # Remove all tokens with a probability less than the last token of the top-k
        indices_to_remove = logits < tf.math.top_k(logits, k=top_k)[0][..., -1, None]
        logits = set_tensor_by_indices_to_value(logits, indices_to_remove, filter_value)

    if top_p < 1.0:
        sorted_indices = tf.argsort(logits, direction="DESCENDING")
        sorted_logits = tf.gather(
            logits, sorted_indices, axis=-1, batch_dims=1
        )  # expects logits to be of dim (batch_size, vocab_size)

        cumulative_probs = tf.math.cumsum(tf.nn.softmax(sorted_logits, axis=-1), axis=-1)

        # Remove tokens with cumulative probability above the threshold (token with 0 are kept)
        sorted_indices_to_remove = cumulative_probs > top_p

        if min_tokens_to_keep > 1:
            # Keep at least min_tokens_to_keep (set to min_tokens_to_keep-1 because we add the first one below)
            sorted_indices_to_remove = tf.concat(
                [
                    tf.zeros_like(sorted_indices_to_remove[:, :min_tokens_to_keep]),
                    sorted_indices_to_remove[:, min_tokens_to_keep:],
                ],
                -1,
            )

        # Shift the indices to the right to keep also the first token above the threshold
        sorted_indices_to_remove = tf.roll(sorted_indices_to_remove, 1, axis=-1)
        sorted_indices_to_remove = tf.concat(
            [tf.zeros_like(sorted_indices_to_remove[:, :1]), sorted_indices_to_remove[:, 1:]], -1,
        )
        # scatter sorted tensors to original indexing
        indices_to_remove = scatter_values_on_batch_indices(sorted_indices_to_remove, sorted_indices)
        logits = set_tensor_by_indices_to_value(logits, indices_to_remove, filter_value)
    return logits


def scatter_values_on_batch_indices(values, batch_indices):
    shape = shape_list(batch_indices)
    # broadcast batch dim to shape
    broad_casted_batch_dims = tf.reshape(tf.broadcast_to(tf.expand_dims(tf.range(shape[0]), axis=-1), shape), [1, -1])
    # transform batch_indices to pair_indices
    pair_indices = tf.transpose(tf.concat([broad_casted_batch_dims, tf.reshape(batch_indices, [1, -1])], 0))
    # scatter values to pair indices
    return tf.scatter_nd(pair_indices, tf.reshape(values, [-1]), shape)


def set_tensor_by_indices_to_value(tensor, indices, value):
    # create value_tensor since tensor value assignment is not possible in TF
    value_tensor = tf.zeros_like(tensor) + value
    return tf.where(indices, value_tensor, tensor)


class BeamHypotheses(object):
    def __init__(self, num_beams, max_length, length_penalty, early_stopping):
        """
        Initialize n-best list of hypotheses.
        """
        self.max_length = max_length - 1  # ignoring bos_token
        self.length_penalty = length_penalty
        self.early_stopping = early_stopping
        self.num_beams = num_beams
        self.beams = []
        self.worst_score = 1e9

    def __len__(self):
        """
        Number of hypotheses in the list.
        """
        return len(self.beams)

    def add(self, hyp, sum_logprobs):
        """
        Add a new hypothesis to the list.
        """
        score = sum_logprobs / len(hyp) ** self.length_penalty
        if len(self) < self.num_beams or score > self.worst_score:
            self.beams.append((score, hyp))
            if len(self) > self.num_beams:
                sorted_scores = sorted([(s, idx) for idx, (s, _) in enumerate(self.beams)])
                del self.beams[sorted_scores[0][1]]
                self.worst_score = sorted_scores[1][0]
            else:
                self.worst_score = min(score, self.worst_score)

    def is_done(self, best_sum_logprobs, cur_len):
        """
        If there are enough hypotheses and that none of the hypotheses being generated
        can become better than the worst one in the heap, then we are done with this sentence.
        """

        if len(self) < self.num_beams:
            return False
        elif self.early_stopping:
            return True
        else:
            cur_score = best_sum_logprobs / cur_len ** self.length_penalty
            ret = self.worst_score >= cur_score
            return ret


class TFConv1D(tf.keras.layers.Layer):
    def __init__(self, nf, nx, initializer_range=0.02, **kwargs):
        """ TFConv1D layer as defined by Radford et al. for OpenAI GPT (and also used in GPT-2)
            Basically works like a Linear layer but the weights are transposed
        """
        super().__init__(**kwargs)
        self.nf = nf
        self.nx = nx
        self.initializer_range = initializer_range

    def build(self, input_shape):
        self.weight = self.add_weight(
            "weight", shape=[self.nx, self.nf], initializer=get_initializer(self.initializer_range)
        )
        self.bias = self.add_weight("bias", shape=[1, self.nf], initializer=tf.zeros_initializer())

    def call(self, x):
        bz, sl = shape_list(x)[:2]

        x = tf.reshape(x, [-1, self.nx])
        x = tf.matmul(x, self.weight) + self.bias

        x = tf.reshape(x, [bz, sl, self.nf])

        return x


class TFSharedEmbeddings(tf.keras.layers.Layer):
    """Construct shared token embeddings.
    """

    def __init__(self, vocab_size, hidden_size, initializer_range=None, **kwargs):
        super().__init__(**kwargs)
        self.vocab_size = vocab_size
        self.hidden_size = hidden_size
        self.initializer_range = hidden_size ** -0.5 if initializer_range is None else initializer_range

    def build(self, input_shape):
        """Build shared token embedding layer
        Shared weights logic adapted from
            https://github.com/tensorflow/models/blob/a009f4fb9d2fc4949e32192a944688925ef78659/official/transformer/v2/embedding_layer.py#L24
        """
        self.weight = self.add_weight(
            "weight", shape=[self.vocab_size, self.hidden_size], initializer=get_initializer(self.initializer_range)
        )
        super().build(input_shape)

    def get_config(self):
        config = {
            "vocab_size": self.vocab_size,
            "hidden_size": self.hidden_size,
            "initializer_range": self.initializer_range,
        }
        base_config = super().get_config()

        return dict(list(base_config.items()) + list(config.items()))

    def call(self, inputs, mode="embedding"):
        """Get token embeddings of inputs.
        Args:
            inputs: list of three int64 tensors with shape [batch_size, length]: (input_ids, position_ids, token_type_ids)
            mode: string, a valid value is one of "embedding" and "linear".
        Returns:
            outputs: (1) If mode == "embedding", output embedding tensor, float32 with
                shape [batch_size, length, embedding_size]; (2) mode == "linear", output
                linear tensor, float32 with shape [batch_size, length, vocab_size].
        Raises:
            ValueError: if mode is not valid.

        Shared weights logic adapted from
            https://github.com/tensorflow/models/blob/a009f4fb9d2fc4949e32192a944688925ef78659/official/transformer/v2/embedding_layer.py#L24
        """
        if mode == "embedding":
            return self._embedding(inputs)
        elif mode == "linear":
            return self._linear(inputs)
        else:
            raise ValueError("mode {} is not valid.".format(mode))

    def _embedding(self, input_ids):
        """Applies embedding based on inputs tensor."""
        return tf.gather(self.weight, input_ids)

    def _linear(self, inputs):
        """Computes logits by running inputs through a linear layer.
            Args:
                inputs: A float32 tensor with shape [..., hidden_size]
            Returns:
                float32 tensor with shape [..., vocab_size].
        """
        first_dims = shape_list(inputs)[:-1]

        x = tf.reshape(inputs, [-1, self.hidden_size])
        logits = tf.matmul(x, self.weight, transpose_b=True)

        return tf.reshape(logits, first_dims + [self.vocab_size])


class TFSequenceSummary(tf.keras.layers.Layer):
    r""" Compute a single vector summary of a sequence hidden states according to various possibilities:
        Args of the config class:
            summary_type:
                - 'last' => [default] take the last token hidden state (like XLNet)
                - 'first' => take the first token hidden state (like Bert)
                - 'mean' => take the mean of all tokens hidden states
                - 'cls_index' => supply a Tensor of classification token position (GPT/GPT-2)
                - 'attn' => Not implemented now, use multi-head attention
            summary_use_proj: Add a projection after the vector extraction
            summary_proj_to_labels: If True, the projection outputs to config.num_labels classes (otherwise to hidden_size). Default: False.
            summary_activation: 'tanh' => add a tanh activation to the output, Other => no activation. Default
            summary_first_dropout: Add a dropout before the projection and activation
            summary_last_dropout: Add a dropout after the projection and activation
    """

    def __init__(self, config, initializer_range=0.02, **kwargs):
        super().__init__(**kwargs)

        self.summary_type = config.summary_type if hasattr(config, "summary_use_proj") else "last"
        if self.summary_type == "attn":
            # We should use a standard multi-head attention module with absolute positional embedding for that.
            # Cf. https://github.com/zihangdai/xlnet/blob/master/modeling.py#L253-L276
            # We can probably just use the multi-head attention module of PyTorch >=1.1.0
            raise NotImplementedError

        self.has_summary = hasattr(config, "summary_use_proj") and config.summary_use_proj
        if self.has_summary:
            if hasattr(config, "summary_proj_to_labels") and config.summary_proj_to_labels and config.num_labels > 0:
                num_classes = config.num_labels
            else:
                num_classes = config.hidden_size
            self.summary = tf.keras.layers.Dense(
                num_classes, kernel_initializer=get_initializer(initializer_range), name="summary"
            )

        self.has_activation = hasattr(config, "summary_activation") and config.summary_activation == "tanh"
        if self.has_activation:
            self.activation = tf.keras.activations.tanh

        self.has_first_dropout = hasattr(config, "summary_first_dropout") and config.summary_first_dropout > 0
        if self.has_first_dropout:
            self.first_dropout = tf.keras.layers.Dropout(config.summary_first_dropout)

        self.has_last_dropout = hasattr(config, "summary_last_dropout") and config.summary_last_dropout > 0
        if self.has_last_dropout:
            self.last_dropout = tf.keras.layers.Dropout(config.summary_last_dropout)

    def call(self, inputs, training=False):
        """ hidden_states: float Tensor in shape [bsz, seq_len, hidden_size], the hidden-states of the last layer.
            cls_index: [optional] position of the classification token if summary_type == 'cls_index',
                shape (bsz,) or more generally (bsz, ...) where ... are optional leading dimensions of hidden_states.
                if summary_type == 'cls_index' and cls_index is None:
                    we take the last token of the sequence as classification token
        """
        if not isinstance(inputs, (dict, tuple, list)):
            hidden_states = inputs
            cls_index = None
        elif isinstance(inputs, (tuple, list)):
            hidden_states = inputs[0]
            cls_index = inputs[1] if len(inputs) > 1 else None
            assert len(inputs) <= 2, "Too many inputs."
        else:
            hidden_states = inputs.get("hidden_states")
            cls_index = inputs.get("cls_index", None)

        if self.summary_type == "last":
            output = hidden_states[:, -1]
        elif self.summary_type == "first":
            output = hidden_states[:, 0]
        elif self.summary_type == "mean":
            output = tf.reduce_mean(hidden_states, axis=1)
        elif self.summary_type == "cls_index":
            hidden_shape = shape_list(hidden_states)  # e.g. [batch, num choices, seq length, hidden dims]
            if cls_index is None:
                cls_index = tf.fill(
                    hidden_shape[:-2], hidden_shape[-2] - 1
                )  # A tensor full of shape [batch] or [batch, num choices] full of sequence length
            cls_shape = shape_list(cls_index)
            if len(cls_shape) <= len(hidden_shape) - 2:
                cls_index = cls_index[..., tf.newaxis]
            # else:
            # cls_index = cls_index[..., tf.newaxis]
            # cls_index = cls_index.expand((-1,) * (cls_index.dim()-1) + (hidden_states.size(-1),))
            # shape of cls_index: (bsz, XX, 1, hidden_size) where XX are optional leading dim of hidden_states
            output = tf.gather(hidden_states, cls_index, batch_dims=len(hidden_shape) - 2)
            output = tf.squeeze(
                output, axis=len(hidden_shape) - 2
            )  # shape of output: (batch, num choices, hidden_size)
        elif self.summary_type == "attn":
            raise NotImplementedError

        if self.has_first_dropout:
            output = self.first_dropout(output, training=training)

        if self.has_summary:
            output = self.summary(output)

        if self.has_activation:
            output = self.activation(output)

        if self.has_last_dropout:
            output = self.last_dropout(output, training=training)

        return output


def shape_list(x):
    """Deal with dynamic shape in tensorflow cleanly."""
    static = x.shape.as_list()
    dynamic = tf.shape(x)
    return [dynamic[i] if s is None else s for i, s in enumerate(static)]


def sample_without_replacement(logits, num_samples):
    """
        categorical sampling witouth replacement is currently not implemented
        the gumbel-max trick will do for now
        see https://github.com/tensorflow/tensorflow/issues/9260 for more info
    """
    z = -tf.math.log(tf.random.uniform(shape_list(logits), 0, 1))
    _, indices = tf.nn.top_k(logits + z, num_samples)
    return indices


def get_initializer(initializer_range=0.02):
    """Creates a `tf.initializers.truncated_normal` with the given range.
    Args:
        initializer_range: float, initializer range for stddev.
    Returns:
        TruncatedNormal initializer with stddev = `initializer_range`.
    """
    return tf.keras.initializers.TruncatedNormal(stddev=initializer_range)


def cast_bool_to_primitive(bool_variable, default_tensor_to_true=False):
    """Function arguments can be inserted as boolean tensor
        and bool variables to cope with keras serialization
        we need to cast `output_attentions` to correct bool
        if it is a tensor

    Args:
        default_tensor_to_true: bool, if tensor should default to True
        in case tensor has no numpy attribute
    """
    # if bool variable is tensor and has numpy value
    if tf.is_tensor(bool_variable):
        if hasattr(bool_variable, "numpy"):
            return bool(bool_variable.numpy())
        elif default_tensor_to_true:
            return True

    # else variable is bool
    return bool_variable