modeling_mctct.py

# coding=utf-8
# Copyright 2022 The HuggingFace Inc. team. 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.
""" PyTorch M-CTC-T model."""


import math
from typing import Optional, Tuple, Union

import torch
import torch.utils.checkpoint
from torch import nn

from ....activations import ACT2FN
from ....deepspeed import is_deepspeed_zero3_enabled
from ....file_utils import add_code_sample_docstrings, add_start_docstrings, add_start_docstrings_to_model_forward
from ....modeling_outputs import BaseModelOutput, CausalLMOutput
from ....modeling_utils import (
    PreTrainedModel,
    apply_chunking_to_forward,
    find_pruneable_heads_and_indices,
    prune_linear_layer,
)
from ....utils import logging
from .configuration_mctct import MCTCTConfig


logger = logging.get_logger(__name__)

_HIDDEN_STATES_START_POSITION = 1

_CONFIG_FOR_DOC = "MCTCTConfig"

# Base docstring
_CHECKPOINT_FOR_DOC = "speechbrain/m-ctc-t-large"
_EXPECTED_OUTPUT_SHAPE = [1, 195, 1536]

# CTC docstring
_CTC_EXPECTED_OUTPUT = '"Mr. Quilter is the apostle of the middle classes, and we\'re glad to welcome his gospel."'
_CTC_EXPECTED_LOSS = 1885.65


MCTCT_PRETRAINED_MODEL_ARCHIVE_LIST = [
    "speechbrain/m-ctc-t-large",
    # See all M-CTC-T models at https://huggingface.co/models?filter=mctct
]


# Copied from transformers.models.bart.modeling_bart._expand_mask
def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
    """
    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
    """
    bsz, src_len = mask.size()
    tgt_len = tgt_len if tgt_len is not None else src_len

    expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)

    inverted_mask = 1.0 - expanded_mask

    return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)


class MCTCTConv1dSubsampler(nn.Module):
    """
    Convolutional subsampler: a stack of 1D convolution (along temporal dimension) followed by non-linear activation
    via gated linear units (https://arxiv.org/abs/1911.08460)
    """

    def __init__(self, config):
        super().__init__()
        self.config = config
        self.glu_dim = config.conv_glu_dim

        self.dropout = nn.Dropout(config.conv_dropout)

        self.num_layers = config.num_conv_layers
        self.in_channels = config.input_feat_per_channel * config.input_channels

        if self.num_layers > 1:
            if config.conv_channels is None:
                raise ValueError(
                    "Need to specify `conv_channels` configuration in `MCTCTConfig` to use multiple convolution"
                    " layers."
                )

            self.mid_channels = config.conv_channels
        else:
            self.mid_channels = None

        self.out_channels = config.hidden_size * 2  # considering GLU halving
        self.kernel_size = config.conv_kernel
        self.stride = config.conv_stride

        # NOTE: MCTCT by construction only uses one convolution kernel. I've made this flexible to allow for
        # multiple layers of convolutions, but not sure if this model definition should just restrict it
        # to one layer. This becomes especially relevant when considering the padding like line 1 of forward().
        self.conv_layers = nn.ModuleList(
            nn.Conv1d(
                self.in_channels if i == 0 else self.mid_channels[i],
                self.mid_channels[i] if i < self.num_layers - 1 else self.out_channels,
                kernel_size=k,
                stride=self.stride[i],
                padding="valid",
            )
            for i, k in enumerate(self.kernel_size)
        )

    def forward(self, input_features):
        # NOTE: in reference to the NOTE in __init__, right now it just calculates padding as if
        # there will be just one conv layer.
        padding = sum([size // 2 for size in self.kernel_size])  # (7, 7) -> (3, 3)

        input_features = torch.nn.functional.pad(input_features, (0, 0, padding, padding), "constant", 0)
        hidden_states = input_features.transpose(1, 2).contiguous()  # -> Batch x Frame x Time
        for conv in self.conv_layers:
            hidden_states = conv(hidden_states)
            hidden_states = nn.functional.glu(hidden_states, dim=self.glu_dim)
            hidden_states = self.dropout(hidden_states)

        hidden_states = hidden_states.transpose(1, 2).contiguous()  # -> Batch x Time x Frame
        return hidden_states


class MCTCTEmbeddings(nn.Module):
    """Construct the embeddings from word, position and token_type embeddings."""

    def __init__(self, config):
        super().__init__()
        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
        self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)
        self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.hidden_size)

        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
        # any TensorFlow checkpoint file
        # self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.LayerNorm = MCTCTLayerNorm()
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
        self.register_buffer(
            "position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)), persistent=False
        )
        self.register_buffer(
            "token_type_ids",
            torch.zeros(self.position_ids.size(), dtype=torch.long, device=self.position_ids.device),
            persistent=False,
        )

    def forward(
        self, input_features=None, token_type_ids=None, position_ids=None, inputs_embeds=None, past_key_values_length=0
    ):
        input_shape = input_features.size() if input_features is not None else inputs_embeds.size()[:-1]

        seq_length = input_shape[1]

        if position_ids is None:
            position_ids = self.position_ids[:, past_key_values_length : seq_length + past_key_values_length]

        # Setting the token_type_ids to the registered buffer in constructor where it is all zeros, which usually occurs
        # when its auto-generated, registered buffer helps users when tracing the model without passing token_type_ids, solves
        # issue #5664
        if token_type_ids is None:
            if hasattr(self, "token_type_ids"):
                buffered_token_type_ids = self.token_type_ids[:, :seq_length]
                buffered_token_type_ids_expanded = buffered_token_type_ids.expand(input_shape[0], seq_length)
                token_type_ids = buffered_token_type_ids_expanded
            else:
                token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=self.position_ids.device)

        if inputs_embeds is None:
            inputs_embeds = self.word_embeddings(input_features)

        token_type_embeddings = self.token_type_embeddings(token_type_ids)

        embeddings = inputs_embeds + token_type_embeddings

        embeddings = self.LayerNorm(embeddings)
        embeddings = self.dropout(embeddings)
        return embeddings


class MCTCTSelfAttention(nn.Module):
    def __init__(self, config):
        super().__init__()
        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
            raise ValueError(
                f"The hidden size ({config.hidden_size}) is not a multiple of the number of attention "
                f"heads ({config.num_attention_heads})"
            )

        self.num_attention_heads = config.num_attention_heads
        self.attention_head_size = config.attention_head_dim
        self.all_head_size = self.num_attention_heads * self.attention_head_size

        self.query = nn.Linear(config.hidden_size, self.all_head_size, bias=False)
        self.key = nn.Linear(config.hidden_size, self.all_head_size, bias=False)
        self.value = nn.Linear(config.hidden_size, self.all_head_size, bias=False)

        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)

        self.max_position_embeddings = config.max_position_embeddings
        self.distance_embedding = nn.Embedding(2 * config.max_position_embeddings - 1, self.attention_head_size)

        self.is_decoder = config.is_decoder

    def transpose_for_scores(self, x):
        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
        x = x.view(*new_x_shape)
        return x.permute(0, 2, 1, 3)

    def reshape_fortran(self, x, shape):
        if len(x.shape) > 0:
            x = x.permute(*reversed(range(len(x.shape))))
        return x.reshape(*reversed(shape)).permute(*reversed(range(len(shape))))

    def relative_position_embedding_rotate(self, scores):
        # NOTE: should re-evaluate whether this re-implementation was truly necessary
        # or the reason why my complete re-haul worked was due to some other part
        # of the code. Adding this and the reshape fortrain code seems very undesirable.
        scores = scores.permute(0, 2, 3, 1)  # e.g. [10, 1839, 14, 4]

        batch, hidden_state, seq_len, heads = scores.shape

        # e.g. [10, 1853, 14, 4]
        scores = torch.cat((scores, torch.zeros((batch, seq_len, seq_len, heads), device=scores.device)), dim=1)

        # e.g. [10, 25942, 1, 4]
        scores = self.reshape_fortran(scores, [batch, (hidden_state + seq_len) * seq_len, 1, heads])

        # e.g. [10, 25928, 1, 4]
        scores = scores[:, : (seq_len + hidden_state - 1) * seq_len]

        # e.g. [10, 1852, 14, 4]
        scores = self.reshape_fortran(scores, [batch, hidden_state + seq_len - 1, seq_len, heads])

        halfpoint = hidden_state // 2
        scores = scores[:, halfpoint : halfpoint + seq_len].transpose(1, 2)  # e.g. [10, 14, 14, 4]

        return scores.permute(0, 3, 1, 2)

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        output_attentions=False,
    ):
        mixed_query_layer = self.query(hidden_states)
        mixed_query_layer = mixed_query_layer / math.sqrt(self.attention_head_size)

        key_layer = self.transpose_for_scores(self.key(hidden_states))
        value_layer = self.transpose_for_scores(self.value(hidden_states))

        query_layer = self.transpose_for_scores(mixed_query_layer)

        # Take the dot product between "query" and "key" to get the raw attention scores.
        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))

        # relative key position embeddings
        positional_embedding = self.distance_embedding.weight
        relative_position_scores = torch.einsum("lh, bche -> bcle", positional_embedding, query_layer.transpose(2, 3))

        relative_position_scores = self.relative_position_embedding_rotate(relative_position_scores)
        attention_scores = attention_scores + relative_position_scores

        if attention_mask is not None:
            # Apply the attention mask is (precomputed for all layers in MCTCTModel forward() function)
            attention_scores = attention_scores + attention_mask

        # Normalize the attention scores to probabilities.
        attention_probs = nn.functional.softmax(attention_scores, dim=-1)

        # This is actually dropping out entire tokens to attend to, which might
        # seem a bit unusual, but is taken from the original Transformer paper.
        attention_probs = self.dropout(attention_probs)

        # Mask heads if we want to
        if head_mask is not None:
            attention_probs = attention_probs * head_mask

        context_layer = torch.matmul(attention_probs, value_layer)

        context_layer = context_layer.permute(0, 2, 1, 3).flatten(start_dim=-2)

        outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)

        return outputs


class MCTCTLayerNorm(nn.Module):
    def __init__(self):
        super().__init__()
        self.singleton_weight = nn.Parameter(torch.ones(1))
        self.singleton_bias = nn.Parameter(torch.zeros(1))

    def forward(self, hidden_states):
        return (hidden_states * self.singleton_weight) + self.singleton_bias


class MCTCTSelfOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.dense = nn.Linear(config.hidden_size, config.hidden_size, bias=False)
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states, input_tensor):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


class MCTCTAttention(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.self = MCTCTSelfAttention(config)
        self.output = MCTCTSelfOutput(config)
        self.pruned_heads = set()

    def prune_heads(self, heads):
        if len(heads) == 0:
            return
        heads, index = find_pruneable_heads_and_indices(
            heads, self.self.num_attention_heads, self.self.attention_head_size, self.pruned_heads
        )

        # Prune linear layers
        self.self.query = prune_linear_layer(self.self.query, index)
        self.self.key = prune_linear_layer(self.self.key, index)
        self.self.value = prune_linear_layer(self.self.value, index)
        self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)

        # Update hyper params and store pruned heads
        self.self.num_attention_heads = self.self.num_attention_heads - len(heads)
        self.self.all_head_size = self.self.attention_head_size * self.self.num_attention_heads
        self.pruned_heads = self.pruned_heads.union(heads)

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        output_attentions=False,
    ):
        self_outputs = self.self(
            hidden_states,
            attention_mask,
            head_mask,
            output_attentions,
        )
        attention_output = self.output(self_outputs[0], hidden_states)
        outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them

        return outputs


class MCTCTIntermediate(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
        if isinstance(config.hidden_act, str):
            self.intermediate_act_fn = ACT2FN[config.hidden_act]
        else:
            self.intermediate_act_fn = config.hidden_act

    def forward(self, hidden_states):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.intermediate_act_fn(hidden_states)
        return hidden_states


class MCTCTOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.intermediate_size, config.hidden_size, bias=False)
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states, input_tensor):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


class MCTCTLayer(nn.Module):
    def __init__(self, config: MCTCTConfig):
        super().__init__()

        self.seq_len_dim = 1
        self.chunk_size_feed_forward = config.chunk_size_feed_forward

        self.intermediate = MCTCTIntermediate(config)
        self.attention = MCTCTAttention(config)
        self.is_decoder = config.is_decoder
        self.output = MCTCTOutput(config)

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        output_attentions=False,
    ):
        self_attention_outputs = self.attention(
            hidden_states, attention_mask, head_mask, output_attentions=output_attentions
        )
        attention_output = self_attention_outputs[0]
        outputs = self_attention_outputs[1:]  # add self attentions if we output attention weights

        layer_output = apply_chunking_to_forward(
            self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output
        )

        outputs = (layer_output,) + outputs

        return outputs

    def feed_forward_chunk(self, attention_output):
        intermediate_output = self.intermediate(attention_output)
        layer_output = self.output(intermediate_output, attention_output)
        return layer_output


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

    config_class = MCTCTConfig
    base_model_prefix = "mctct"
    main_input_name = "input_features"
    supports_gradient_checkpointing = True

    def _init_weights(self, module):
        """Initialize the weights"""
        std = self.config.initializer_range
        if isinstance(module, nn.Linear):
            # Slightly different from the TF version which uses truncated_normal for initialization
            # cf https://github.com/pytorch/pytorch/pull/5617
            module.weight.data.normal_(mean=0.0, std=std)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.Embedding):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.padding_idx is not None:
                module.weight.data[module.padding_idx].zero_()
        elif isinstance(module, nn.LayerNorm):
            module.bias.data.zero_()
            module.weight.data.fill_(1.0)
        elif isinstance(module, MCTCTLayerNorm):
            module.singleton_weight.data.fill_(1.0)
            module.singleton_bias.data.zero_()
        if isinstance(module, (nn.Linear, nn.Conv1d)):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.bias is not None:
                module.bias.data.zero_()

    def _get_feat_extract_output_lengths(self, input_lengths: torch.LongTensor):
        """
        Computes the output length of the convolutional layers
        """
        dilation = 1
        for _, kernel_sz, stride in zip(
            range(self.config.num_conv_layers), self.config.conv_kernel, self.config.conv_stride
        ):
            padding = kernel_sz // 2
            input_lengths = input_lengths + 2 * padding - dilation * (kernel_sz - 1) - 1
            input_lengths = torch.div(input_lengths, stride, rounding_mode="trunc") + 1

        return input_lengths

    def _get_feature_vector_attention_mask(self, feature_vector_length, attention_mask):
        # generate creates 3D attention mask, because of the shape of input_features
        # convert it to 2D if thats the case
        if len(attention_mask.shape) > 2:
            attention_mask = attention_mask[:, :, -1]

        # subsampled_lengths = attention_mask.sum(-1)
        subsampled_lengths = self._get_feat_extract_output_lengths(attention_mask.sum(-1))
        bsz = attention_mask.size()[0]
        attention_mask = torch.zeros(
            (bsz, feature_vector_length), dtype=attention_mask.dtype, device=attention_mask.device
        )

        # these two operations makes sure that all values
        # before the output lengths indices are attended to
        attention_mask[(torch.arange(bsz, device=attention_mask.device), subsampled_lengths - 1)] = 1
        attention_mask = attention_mask.flip([-1]).cumsum(-1).flip([-1]).long()
        return attention_mask

    def _set_gradient_checkpointing(self, module, value=False):
        if isinstance(module, (MCTCTEncoder)):
            module.gradient_checkpointing = value


MCTCT_START_DOCSTRING = r"""
    This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) sub-class. Use
    it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and
    behavior.

    Parameters:
        config ([`MCTCTConfig`]): 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 [`~PreTrainedModel.from_pretrained`] method to load the model weights.
"""

MCTCT_INPUTS_DOCSTRING = r"""
    Args:
        input_features (`torch.LongTensor` of shape `({0})`):
            Indices of input sequence tokens in the vocabulary.

            Indices can be obtained using [`Wav2Vec2CTCTokenizer`]. See [`PreTrainedTokenizer.encode`] and
            [`PreTrainedTokenizer.__call__`] for details.

            [What are input IDs?](../glossary#input-ids)
        attention_mask (`torch.FloatTensor` of shape `({0})`, *optional*):
            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:

            - 1 for tokens that are **not masked**,
            - 0 for tokens that are **masked**.

            [What are attention masks?](../glossary#attention-mask)
        head_mask (`torch.FloatTensor` of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*):
            Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`:

            - 1 indicates the head is **not masked**,
            - 0 indicates the head is **masked**.
        output_attentions (`bool`, *optional*):
            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
            tensors for more detail.
        output_hidden_states (`bool`, *optional*):
            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
            more detail.
        return_dict (`bool`, *optional*):
            Whether or not to return a [`~file_utils.ModelOutput`] instead of a plain tuple.
"""


class MCTCTEncoder(MCTCTPreTrainedModel):
    def __init__(self, config: MCTCTConfig):
        super().__init__(config)
        self.hidden_dropout_prob = config.hidden_dropout_prob

        self.layer_norm = MCTCTLayerNorm()
        self.conv = MCTCTConv1dSubsampler(config)
        self.layers = nn.ModuleList([MCTCTLayer(config) for _ in range(config.num_hidden_layers)])

        self.gradient_checkpointing = False

    def forward(
        self,
        input_features: torch.Tensor,
        attention_mask: torch.Tensor,
        head_mask: torch.Tensor,
        output_attentions: bool = False,
        output_hidden_states: bool = False,
        return_dict: bool = True,
    ) -> Union[Tuple, BaseModelOutput]:
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        input_features = self.layer_norm(input_features)

        inputs_embeds = self.conv(input_features)

        # subsample attention mask if necessary
        if attention_mask is not None:
            attention_mask = self._get_feature_vector_attention_mask(inputs_embeds.shape[1], attention_mask)

        hidden_states = nn.functional.dropout(inputs_embeds, p=self.hidden_dropout_prob, training=self.training)

        # expand attention_mask
        if attention_mask is not None:
            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
            attention_mask = _expand_mask(attention_mask, inputs_embeds.dtype)

        encoder_states = () if output_hidden_states else None
        all_attentions = () if output_attentions else None

        # check if head_mask has a correct number of layers specified if desired
        if head_mask is not None:
            if head_mask.size()[0] != len(self.layers):
                raise ValueError(
                    f"The head_mask should be specified for {len(self.layers)} layers, "
                    f"but it is for {head_mask.size()[0]}."
                )

        deepspeed_zero3_is_enabled = is_deepspeed_zero3_enabled()
        for idx, encoder_layer in enumerate(self.layers):
            if output_hidden_states:
                encoder_states = encoder_states + (hidden_states,)

            # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
            dropout_probability = torch.rand([])

            skip_the_layer = True if self.training and (dropout_probability < self.config.layerdrop) else False
            if not skip_the_layer or deepspeed_zero3_is_enabled:
                # under deepspeed zero3 all gpus must run in sync
                if self.gradient_checkpointing and self.training:

                    def create_custom_forward(module):
                        def custom_forward(*inputs):
                            return module(*inputs, output_attentions)

                        return custom_forward

                    layer_outputs = torch.utils.checkpoint.checkpoint(
                        create_custom_forward(encoder_layer),
                        hidden_states,
                        attention_mask,
                        (head_mask[idx] if head_mask is not None else None),
                    )
                else:
                    layer_outputs = encoder_layer(
                        hidden_states=hidden_states,
                        attention_mask=attention_mask,
                        output_attentions=output_attentions,
                    )

                hidden_states = layer_outputs[0]

            if skip_the_layer:
                layer_outputs = (None, None)

            if output_attentions:
                all_attentions = all_attentions + (layer_outputs[1],)

        if output_hidden_states:
            encoder_states = encoder_states + (hidden_states,)

        if not return_dict:
            return tuple(v for v in [hidden_states, encoder_states, all_attentions] if v is not None)
        return BaseModelOutput(
            last_hidden_state=hidden_states, hidden_states=encoder_states, attentions=all_attentions
        )


@add_start_docstrings(
    "The bare M-CTC-T Model transformer outputting raw hidden-states without any specific head on top.",
    MCTCT_START_DOCSTRING,
)
class MCTCTModel(MCTCTPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.config = config

        self.encoder = MCTCTEncoder(config)

        # Initialize weights and apply final processing
        self.post_init()

    @add_start_docstrings_to_model_forward(MCTCT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
    @add_code_sample_docstrings(
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=BaseModelOutput,
        config_class=_CONFIG_FOR_DOC,
        modality="audio",
        expected_output=_EXPECTED_OUTPUT_SHAPE,
    )
    def forward(
        self,
        input_features: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        head_mask: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, BaseModelOutput]:
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        if input_features is None:
            raise ValueError("You have to specify input_features.")

        encoder_outputs = self.encoder(
            input_features,
            attention_mask=attention_mask,
            head_mask=head_mask,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )
        sequence_output = encoder_outputs[0]

        if not return_dict:
            return (sequence_output,) + encoder_outputs[1:]

        return BaseModelOutput(
            last_hidden_state=sequence_output,
            hidden_states=encoder_outputs.hidden_states,
            attentions=encoder_outputs.attentions,
        )


@add_start_docstrings(
    """MCTCT Model with a `language modeling` head on top for Connectionist Temporal Classification (CTC).""",
    MCTCT_START_DOCSTRING,
)
class MCTCTForCTC(MCTCTPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)

        self.mctct = MCTCTModel(config)

        if config.vocab_size is None:
            raise ValueError(
                f"You are trying to instantiate {self.__class__} with a configuration that "
                "does not define the vocabulary size of the language model head. Please "
                "instantiate the model as follows: `MCTCTForCTC.from_pretrained(..., vocab_size=vocab_size)`. "
                "or define `vocab_size` of your model's configuration."
            )
        output_hidden_size = config.hidden_size

        self.ctc_head = nn.Linear(output_hidden_size, config.vocab_size)

        # Initialize weights and apply final processing
        self.post_init()

    @add_start_docstrings_to_model_forward(MCTCT_INPUTS_DOCSTRING)
    @add_code_sample_docstrings(
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=CausalLMOutput,
        config_class=_CONFIG_FOR_DOC,
        expected_output=_CTC_EXPECTED_OUTPUT,
        expected_loss=_CTC_EXPECTED_LOSS,
    )
    def forward(
        self,
        input_features: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        head_mask: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        labels: Optional[torch.LongTensor] = None,
    ) -> Union[Tuple, CausalLMOutput]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, target_length)`, *optional*):
            Labels for connectionist temporal classification. Note that `target_length` has to be smaller or equal to
            the sequence length of the output logits. Indices are selected in `[-100, 0, ..., config.vocab_size - 1]`.
            All labels set to `-100` are ignored (masked), the loss is only computed for labels in `[0, ...,
            config.vocab_size - 1]`.
        """

        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
        outputs = self.mctct(
            input_features,
            attention_mask=attention_mask,
            head_mask=head_mask,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        hidden_states = outputs[0]

        logits = self.ctc_head(hidden_states)

        loss = None
        if labels is not None:
            if labels.max() >= self.config.vocab_size:
                raise ValueError(f"Label values must be <= vocab_size: {self.config.vocab_size}")

            # retrieve loss input_lengths from attention_mask
            attention_mask = (
                attention_mask
                if attention_mask is not None
                else torch.ones(input_features.shape[:-1], dtype=torch.long)
            )
            input_lengths = self._get_feat_extract_output_lengths(attention_mask.sum(-1)).to(torch.long)
            # assuming that padded tokens are filled with -100
            # when not being attended to
            labels_mask = labels >= 0
            target_lengths = labels_mask.sum(-1)
            flattened_targets = labels.masked_select(labels_mask)

            # ctc_loss doesn't support fp16
            log_probs = nn.functional.log_softmax(logits, dim=-1, dtype=torch.float32).transpose(0, 1)

            with torch.backends.cudnn.flags(enabled=False):
                loss = nn.functional.ctc_loss(
                    log_probs,
                    flattened_targets,
                    input_lengths,
                    target_lengths,
                    blank=self.config.pad_token_id,
                    reduction=self.config.ctc_loss_reduction,
                    zero_infinity=self.config.ctc_zero_infinity,
                )

        if not return_dict:
            output = (logits,) + outputs[_HIDDEN_STATES_START_POSITION:]
            return ((loss,) + output) if loss is not None else output

        return CausalLMOutput(
            loss=loss, logits=logits, hidden_states=outputs.hidden_states, attentions=outputs.attentions
        )