# 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 2.0 RoBERTa model. """


import logging

import tensorflow as tf

from .configuration_roberta import RobertaConfig
from .file_utils import add_start_docstrings, add_start_docstrings_to_callable
from .modeling_tf_bert import TFBertEmbeddings, TFBertMainLayer, gelu
from .modeling_tf_utils import TFPreTrainedModel, get_initializer, shape_list


logger = logging.getLogger(__name__)

TF_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP = {
    "roberta-base": "https://cdn.huggingface.co/roberta-base-tf_model.h5",
    "roberta-large": "https://cdn.huggingface.co/roberta-large-tf_model.h5",
    "roberta-large-mnli": "https://cdn.huggingface.co/roberta-large-mnli-tf_model.h5",
    "distilroberta-base": "https://cdn.huggingface.co/distilroberta-base-tf_model.h5",
}


class TFRobertaEmbeddings(TFBertEmbeddings):
    """
    Same as BertEmbeddings with a tiny tweak for positional embeddings indexing.
    """

    def __init__(self, config, **kwargs):
        super().__init__(config, **kwargs)
        self.padding_idx = 1

    def create_position_ids_from_input_ids(self, x):
        """ Replace non-padding symbols with their position numbers. Position numbers begin at
        padding_idx+1. Padding symbols are ignored. This is modified from fairseq's
        `utils.make_positions`.
        :param tf.Tensor x:
        :return tf.Tensor:
        """
        mask = tf.cast(tf.math.not_equal(x, self.padding_idx), dtype=tf.int32)
        incremental_indicies = tf.math.cumsum(mask, axis=1) * mask
        return incremental_indicies + self.padding_idx

    def create_position_ids_from_inputs_embeds(self, inputs_embeds):
        """ We are provided embeddings directly. We cannot infer which are padded so just generate
        sequential position ids.
        :param tf.Tensor inputs_embeds:
        :return tf.Tensor:
        """
        seq_length = shape_list(inputs_embeds)[1]

        position_ids = tf.range(self.padding_idx + 1, seq_length + self.padding_idx + 1, dtype=tf.int32)[tf.newaxis, :]
        return position_ids

    def _embedding(self, inputs, training=False):
        """Applies embedding based on inputs tensor."""
        input_ids, position_ids, token_type_ids, inputs_embeds = inputs

        if position_ids is None:
            if input_ids is not None:
                # Create the position ids from the input token ids. Any padded tokens remain padded.
                position_ids = self.create_position_ids_from_input_ids(input_ids)
            else:
                position_ids = self.create_position_ids_from_inputs_embeds(inputs_embeds)

        return super()._embedding([input_ids, position_ids, token_type_ids, inputs_embeds], training=training)


class TFRobertaMainLayer(TFBertMainLayer):
    """
    Same as TFBertMainLayer but uses TFRobertaEmbeddings.
    """

    def __init__(self, config, **kwargs):
        super().__init__(config, **kwargs)
        self.embeddings = TFRobertaEmbeddings(config, name="embeddings")

    def get_input_embeddings(self):
        return self.embeddings


class TFRobertaPreTrainedModel(TFPreTrainedModel):
    """ An abstract class to handle weights initialization and
        a simple interface for downloading and loading pretrained models.
    """

    config_class = RobertaConfig
    pretrained_model_archive_map = TF_ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP
    base_model_prefix = "roberta"


ROBERTA_START_DOCSTRING = r"""
    This model is a `tf.keras.Model <https://www.tensorflow.org/api_docs/python/tf/keras/Model>`__ sub-class.
    Use it as a regular TF 2.0 Keras Model and
    refer to the TF 2.0 documentation for all matter related to general usage and behavior.

    .. note::

        TF 2.0 models accepts two formats as inputs:

            - having all inputs as keyword arguments (like PyTorch models), or
            - having all inputs as a list, tuple or dict in the first positional arguments.

        This second option is useful when using :obj:`tf.keras.Model.fit()` method which currently requires having
        all the tensors in the first argument of the model call function: :obj:`model(inputs)`.

        If you choose this second option, there are three possibilities you can use to gather all the input Tensors
        in the first positional argument :

        - a single Tensor with input_ids only and nothing else: :obj:`model(inputs_ids)`
        - a list of varying length with one or several input Tensors IN THE ORDER given in the docstring:
          :obj:`model([input_ids, attention_mask])` or :obj:`model([input_ids, attention_mask, token_type_ids])`
        - a dictionary with one or several input Tensors associated to the input names given in the docstring:
          :obj:`model({'input_ids': input_ids, 'token_type_ids': token_type_ids})`

    Parameters:
        config (:class:`~transformers.RobertaConfig`): Model configuration class with all the parameters of the
            model. Initializing with a config file does not load the weights associated with the model, only the configuration.
            Check out the :meth:`~transformers.PreTrainedModel.from_pretrained` method to load the model weights.
"""

ROBERTA_INPUTS_DOCSTRING = r"""
    Args:
        input_ids (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`):
            Indices of input sequence tokens in the vocabulary.

            Indices can be obtained using :class:`transformers.RobertaTokenizer`.
            See :func:`transformers.PreTrainedTokenizer.encode` and
            :func:`transformers.PreTrainedTokenizer.encode_plus` for details.

            `What are input IDs? <../glossary.html#input-ids>`__
        attention_mask (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
            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.

            `What are attention masks? <../glossary.html#attention-mask>`__
        token_type_ids (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
            Segment token indices to indicate first and second portions of the inputs.
            Indices are selected in ``[0, 1]``: ``0`` corresponds to a `sentence A` token, ``1``
            corresponds to a `sentence B` token

            `What are token type IDs? <../glossary.html#token-type-ids>`__
        position_ids (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
            Indices of positions of each input sequence tokens in the position embeddings.
            Selected in the range ``[0, config.max_position_embeddings - 1]``.

            `What are position IDs? <../glossary.html#position-ids>`__
        head_mask (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(num_heads,)` or :obj:`(num_layers, num_heads)`, `optional`, defaults to :obj:`None`):
            Mask to nullify selected heads of the self-attention modules.
            Mask values selected in ``[0, 1]``:
            :obj:`1` indicates the head is **not masked**, :obj:`0` indicates the head is **masked**.
        inputs_embeds (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, embedding_dim)`, `optional`, defaults to :obj:`None`):
            Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation.
            This is useful if you want more control over how to convert `input_ids` indices into associated vectors
            than the model's internal embedding lookup matrix.
        training (:obj:`boolean`, `optional`, defaults to :obj:`False`):
            Whether to activate dropout modules (if set to :obj:`True`) during training or to de-activate them
            (if set to :obj:`False`) for evaluation.
"""


@add_start_docstrings(
    "The bare RoBERTa Model transformer outputing raw hidden-states without any specific head on top.",
    ROBERTA_START_DOCSTRING,
)
class TFRobertaModel(TFRobertaPreTrainedModel):
    def __init__(self, config, *inputs, **kwargs):
        super().__init__(config, *inputs, **kwargs)
        self.roberta = TFRobertaMainLayer(config, name="roberta")

    @add_start_docstrings_to_callable(ROBERTA_INPUTS_DOCSTRING)
    def call(self, inputs, **kwargs):
        r"""
    Returns:
        :obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.RobertaConfig`) and inputs:
        last_hidden_state (:obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`):
            Sequence of hidden-states at the output of the last layer of the model.
        pooler_output (:obj:`tf.Tensor` of shape :obj:`(batch_size, hidden_size)`):
            Last layer hidden-state of the first token of the sequence (classification token)
            further processed by a Linear layer and a Tanh activation function. The Linear
            layer weights are trained from the next sentence prediction (classification)
            objective during Bert pretraining. This output is usually *not* a good summary
            of the semantic content of the input, you're often better with averaging or pooling
            the sequence of hidden-states for the whole input sequence.
        hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when :obj:`config.output_hidden_states=True`):
            tuple of :obj:`tf.Tensor` (one for the output of the embeddings + one for the output of each layer)
            of shape :obj:`(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
        attentions (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``config.output_attentions=True``):
            tuple of :obj:`tf.Tensor` (one for each layer) of shape
            :obj:`(batch_size, num_heads, sequence_length, sequence_length)`:

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

    Examples::

        import tensorflow as tf
        from transformers import RobertaTokenizer, TFRobertaModel

        tokenizer = RobertaTokenizer.from_pretrained('roberta-base')
        model = TFRobertaModel.from_pretrained('roberta-base')
        input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True))[None, :]  # Batch size 1
        outputs = model(input_ids)
        last_hidden_states = outputs[0]  # The last hidden-state is the first element of the output tuple

        """
        outputs = self.roberta(inputs, **kwargs)
        return outputs


class TFRobertaLMHead(tf.keras.layers.Layer):
    """Roberta Head for masked language modeling."""

    def __init__(self, config, input_embeddings, **kwargs):
        super().__init__(**kwargs)
        self.vocab_size = config.vocab_size
        self.dense = tf.keras.layers.Dense(
            config.hidden_size, kernel_initializer=get_initializer(config.initializer_range), name="dense"
        )
        self.layer_norm = tf.keras.layers.LayerNormalization(epsilon=config.layer_norm_eps, name="layer_norm")
        self.act = tf.keras.layers.Activation(gelu)

        # The output weights are the same as the input embeddings, but there is
        # an output-only bias for each token.
        self.decoder = input_embeddings

    def build(self, input_shape):
        self.bias = self.add_weight(shape=(self.vocab_size,), initializer="zeros", trainable=True, name="bias")
        super().build(input_shape)

    def call(self, features):
        x = self.dense(features)
        x = self.act(x)
        x = self.layer_norm(x)

        # project back to size of vocabulary with bias
        x = self.decoder(x, mode="linear") + self.bias

        return x


@add_start_docstrings("""RoBERTa Model with a `language modeling` head on top. """, ROBERTA_START_DOCSTRING)
class TFRobertaForMaskedLM(TFRobertaPreTrainedModel):
    def __init__(self, config, *inputs, **kwargs):
        super().__init__(config, *inputs, **kwargs)

        self.roberta = TFRobertaMainLayer(config, name="roberta")
        self.lm_head = TFRobertaLMHead(config, self.roberta.embeddings, name="lm_head")

    def get_output_embeddings(self):
        return self.lm_head.decoder

    @add_start_docstrings_to_callable(ROBERTA_INPUTS_DOCSTRING)
    def call(self, inputs, **kwargs):
        r"""
    Return:
        :obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.RobertaConfig`) and inputs:
        prediction_scores (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, config.vocab_size)`):
            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
        hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when :obj:`config.output_hidden_states=True`):
            tuple of :obj:`tf.Tensor` (one for the output of the embeddings + one for the output of each layer)
            of shape :obj:`(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
        attentions (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``config.output_attentions=True``):
            tuple of :obj:`tf.Tensor` (one for each layer) of shape
            :obj:`(batch_size, num_heads, sequence_length, sequence_length)`:

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

    Examples::

        import tensorflow as tf
        from transformers import RobertaTokenizer, TFRobertaForMaskedLM

        tokenizer = RobertaTokenizer.from_pretrained('roberta-base')
        model = TFRobertaForMaskedLM.from_pretrained('roberta-base')
        input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True))[None, :]  # Batch size 1
        outputs = model(input_ids)
        prediction_scores = outputs[0]

        """
        outputs = self.roberta(inputs, **kwargs)

        sequence_output = outputs[0]
        prediction_scores = self.lm_head(sequence_output)

        outputs = (prediction_scores,) + outputs[2:]  # Add hidden states and attention if they are here

        return outputs  # prediction_scores, (hidden_states), (attentions)


class TFRobertaClassificationHead(tf.keras.layers.Layer):
    """Head for sentence-level classification tasks."""

    def __init__(self, config, **kwargs):
        super().__init__(config, **kwargs)
        self.dense = tf.keras.layers.Dense(
            config.hidden_size,
            kernel_initializer=get_initializer(config.initializer_range),
            activation="tanh",
            name="dense",
        )
        self.dropout = tf.keras.layers.Dropout(config.hidden_dropout_prob)
        self.out_proj = tf.keras.layers.Dense(
            config.num_labels, kernel_initializer=get_initializer(config.initializer_range), name="out_proj"
        )

    def call(self, features, training=False):
        x = features[:, 0, :]  # take <s> token (equiv. to [CLS])
        x = self.dropout(x, training=training)
        x = self.dense(x)
        x = self.dropout(x, training=training)
        x = self.out_proj(x)
        return x


@add_start_docstrings(
    """RoBERTa Model transformer with a sequence classification/regression head on top (a linear layer
    on top of the pooled output) e.g. for GLUE tasks. """,
    ROBERTA_START_DOCSTRING,
)
class TFRobertaForSequenceClassification(TFRobertaPreTrainedModel):
    def __init__(self, config, *inputs, **kwargs):
        super().__init__(config, *inputs, **kwargs)
        self.num_labels = config.num_labels

        self.roberta = TFRobertaMainLayer(config, name="roberta")
        self.classifier = TFRobertaClassificationHead(config, name="classifier")

    @add_start_docstrings_to_callable(ROBERTA_INPUTS_DOCSTRING)
    def call(self, inputs, **kwargs):
        r"""
    Return:
        :obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.RobertaConfig`) and inputs:
        logits (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, config.num_labels)`):
            Classification (or regression if config.num_labels==1) scores (before SoftMax).
        hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when :obj:`config.output_hidden_states=True`):
            tuple of :obj:`tf.Tensor` (one for the output of the embeddings + one for the output of each layer)
            of shape :obj:`(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
        attentions (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``config.output_attentions=True``):
            tuple of :obj:`tf.Tensor` (one for each layer) of shape
            :obj:`(batch_size, num_heads, sequence_length, sequence_length)`:

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

    Examples::

        import tensorflow as tf
        from transformers import RobertaTokenizer, TFRobertaForSequenceClassification

        tokenizer = RobertaTokenizer.from_pretrained('roberta-base')
        model = TFRobertaForSequenceClassification.from_pretrained('roberta-base')
        input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True))[None, :]  # Batch size 1
        labels = tf.constant([1])[None, :]  # Batch size 1
        outputs = model(input_ids)
        logits = outputs[0]

        """
        outputs = self.roberta(inputs, **kwargs)

        sequence_output = outputs[0]
        logits = self.classifier(sequence_output, training=kwargs.get("training", False))

        outputs = (logits,) + outputs[2:]

        return outputs  # logits, (hidden_states), (attentions)


@add_start_docstrings(
    """RoBERTa Model with a token classification head on top (a linear layer on top of
    the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks. """,
    ROBERTA_START_DOCSTRING,
)
class TFRobertaForTokenClassification(TFRobertaPreTrainedModel):
    def __init__(self, config, *inputs, **kwargs):
        super().__init__(config, *inputs, **kwargs)
        self.num_labels = config.num_labels

        self.roberta = TFRobertaMainLayer(config, name="roberta")
        self.dropout = tf.keras.layers.Dropout(config.hidden_dropout_prob)
        self.classifier = tf.keras.layers.Dense(
            config.num_labels, kernel_initializer=get_initializer(config.initializer_range), name="classifier"
        )

    @add_start_docstrings_to_callable(ROBERTA_INPUTS_DOCSTRING)
    def call(self, inputs, **kwargs):
        r"""
    Return:
        :obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.RobertaConfig`) and inputs:
        scores (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length, config.num_labels)`):
            Classification scores (before SoftMax).
        hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when :obj:`config.output_hidden_states=True`):
            tuple of :obj:`tf.Tensor` (one for the output of the embeddings + one for the output of each layer)
            of shape :obj:`(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
        attentions (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``config.output_attentions=True``):
            tuple of :obj:`tf.Tensor` (one for each layer) of shape
            :obj:`(batch_size, num_heads, sequence_length, sequence_length)`:

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

    Examples::

        import tensorflow as tf
        from transformers import RobertaTokenizer, TFRobertaForTokenClassification

        tokenizer = RobertaTokenizer.from_pretrained('roberta-base')
        model = TFRobertaForTokenClassification.from_pretrained('roberta-base')
        input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True))[None, :]  # Batch size 1
        outputs = model(input_ids)
        scores = outputs[0]

        """
        outputs = self.roberta(inputs, **kwargs)

        sequence_output = outputs[0]

        sequence_output = self.dropout(sequence_output, training=kwargs.get("training", False))
        logits = self.classifier(sequence_output)

        outputs = (logits,) + outputs[2:]  # add hidden states and attention if they are here

        return outputs  # scores, (hidden_states), (attentions)


@add_start_docstrings(
    """RoBERTa Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of the hidden-states output to compute `span start logits` and `span end logits`). """,
    ROBERTA_START_DOCSTRING,
)
class TFRobertaForQuestionAnswering(TFRobertaPreTrainedModel):
    def __init__(self, config, *inputs, **kwargs):
        super().__init__(config, *inputs, **kwargs)
        self.num_labels = config.num_labels

        self.roberta = TFRobertaMainLayer(config, name="roberta")
        self.qa_outputs = tf.keras.layers.Dense(
            config.num_labels, kernel_initializer=get_initializer(config.initializer_range), name="qa_outputs"
        )

    @add_start_docstrings_to_callable(ROBERTA_INPUTS_DOCSTRING)
    def call(self, inputs, **kwargs):
        r"""
    Return:
        :obj:`tuple(tf.Tensor)` comprising various elements depending on the configuration (:class:`~transformers.RobertaConfig`) and inputs:
        start_scores (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length,)`):
            Span-start scores (before SoftMax).
        end_scores (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length,)`):
            Span-end scores (before SoftMax).
        hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when :obj:`config.output_hidden_states=True`):
            tuple of :obj:`tf.Tensor` (one for the output of the embeddings + one for the output of each layer)
            of shape :obj:`(batch_size, sequence_length, hidden_size)`.

            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
        attentions (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``config.output_attentions=True``):
            tuple of :obj:`tf.Tensor` (one for each layer) of shape
            :obj:`(batch_size, num_heads, sequence_length, sequence_length)`:

            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

    Examples::

        # The checkpoint roberta-base is not fine-tuned for question answering. Please see the
        # examples/question-answering/run_squad.py example to see how to fine-tune a model to a question answering task.

        import tensorflow as tf
        from transformers import RobertaTokenizer, TFRobertaForQuestionAnswering

        tokenizer = RobertaTokenizer.from_pretrained('roberta-base')
        model = TFRobertaForQuestionAnswering.from_pretrained('roberta-base')
        input_ids = tokenizer.encode("Who was Jim Henson?", "Jim Henson was a nice puppet")
        start_scores, end_scores = model(tf.constant(input_ids)[None, :]) # Batch size 1

        all_tokens = tokenizer.convert_ids_to_tokens(input_ids)
        answer = ' '.join(all_tokens[tf.math.argmax(start_scores, 1)[0] : tf.math.argmax(end_scores, 1)[0]+1])

        """
        outputs = self.roberta(inputs, **kwargs)

        sequence_output = outputs[0]

        logits = self.qa_outputs(sequence_output)
        start_logits, end_logits = tf.split(logits, 2, axis=-1)
        start_logits = tf.squeeze(start_logits, axis=-1)
        end_logits = tf.squeeze(end_logits, axis=-1)

        outputs = (start_logits, end_logits,) + outputs[2:]

        return outputs  # start_logits, end_logits, (hidden_states), (attentions)