public/gpt-2/transformers/models/ctrl/modeling_ctrl.py

# coding=utf-8
# Copyright 2018 Salesforce and 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.
""" PyTorch CTRL model."""

from typing import Tuple

import numpy as np
import torch
from torch import nn
from torch.nn import CrossEntropyLoss, MSELoss

from ...file_utils import add_code_sample_docstrings, add_start_docstrings, add_start_docstrings_to_model_forward
from ...modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast, SequenceClassifierOutput
from ...modeling_utils import Conv1D, PreTrainedModel, find_pruneable_heads_and_indices, prune_linear_layer
from ...utils import logging
from .configuration_ctrl import CTRLConfig


logger = logging.get_logger(__name__)

_CHECKPOINT_FOR_DOC = "ctrl"
_CONFIG_FOR_DOC = "CTRLConfig"
_TOKENIZER_FOR_DOC = "CTRLTokenizer"

CTRL_PRETRAINED_MODEL_ARCHIVE_LIST = [
    "ctrl"
    # See all CTRL models at https://huggingface.co/models?filter=ctrl
]


def angle_defn(pos, i, d_model_size):
    angle_rates = 1 / torch.pow(10000, (2 * (i // 2)) / d_model_size)
    return pos * angle_rates


def positional_encoding(position, d_model_size, dtype):
    # create the sinusoidal pattern for the positional encoding
    angle_rads = angle_defn(
        torch.arange(position, dtype=dtype).unsqueeze(1),
        torch.arange(d_model_size, dtype=dtype).unsqueeze(0),
        d_model_size,
    )

    sines = torch.sin(angle_rads[:, 0::2])
    cosines = torch.cos(angle_rads[:, 1::2])

    pos_encoding = torch.cat([sines, cosines], dim=-1)
    return pos_encoding


def scaled_dot_product_attention(q, k, v, mask, attention_mask=None, head_mask=None):
    # calculate attention
    matmul_qk = torch.matmul(q, k.permute(0, 1, 3, 2))

    dk = k.shape[-1]
    scaled_attention_logits = matmul_qk / np.sqrt(dk)

    if mask is not None:
        nd, ns = scaled_attention_logits.size(-2), scaled_attention_logits.size(-1)
        scaled_attention_logits += mask[ns - nd : ns, :ns] * -1e4

    if attention_mask is not None:
        # Apply the attention mask
        scaled_attention_logits = scaled_attention_logits + attention_mask

    attention_weights = torch.softmax(scaled_attention_logits, dim=-1)

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

    output = torch.matmul(attention_weights, v)

    return output, attention_weights


class MultiHeadAttention(nn.Module):
    def __init__(self, d_model_size, num_heads):
        super().__init__()
        self.num_heads = num_heads
        self.d_model_size = d_model_size

        self.depth = int(d_model_size / self.num_heads)

        self.Wq = nn.Linear(d_model_size, d_model_size)
        self.Wk = nn.Linear(d_model_size, d_model_size)
        self.Wv = nn.Linear(d_model_size, d_model_size)

        self.dense = nn.Linear(d_model_size, d_model_size)
        self.pruned_heads = set()

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

        # Prune linear layers
        self.Wq = prune_linear_layer(self.Wq, index)
        self.Wk = prune_linear_layer(self.Wk, index)
        self.Wv = prune_linear_layer(self.Wv, index)
        self.dense = prune_linear_layer(self.dense, index, dim=1)

        # Update hyper params
        self.num_heads = self.num_heads - len(heads)
        self.d_model_size = attention_head_size * self.num_heads
        self.pruned_heads = self.pruned_heads.union(heads)

    def split_into_heads(self, x, batch_size):
        x = x.reshape(batch_size, -1, self.num_heads, self.depth)
        return x.permute([0, 2, 1, 3])

    def forward(
        self,
        v,
        k,
        q,
        mask,
        layer_past=None,
        attention_mask=None,
        head_mask=None,
        use_cache=False,
        output_attentions=False,
    ):
        batch_size = q.shape[0]

        q = self.Wq(q)
        k = self.Wk(k)
        v = self.Wv(v)

        q = self.split_into_heads(q, batch_size)
        k = self.split_into_heads(k, batch_size)
        v = self.split_into_heads(v, batch_size)
        if layer_past is not None:
            past_key, past_value = layer_past[0], layer_past[1]
            k = torch.cat((past_key, k), dim=-2)
            v = torch.cat((past_value, v), dim=-2)

        if use_cache is True:
            present = torch.stack((k, v))
        else:
            present = (None,)

        output = scaled_dot_product_attention(q, k, v, mask, attention_mask, head_mask)
        scaled_attention = output[0].permute([0, 2, 1, 3])
        attn = output[1]
        original_size_attention = scaled_attention.reshape(batch_size, -1, self.d_model_size)
        output = self.dense(original_size_attention)

        outputs = (output, present)
        if output_attentions:
            outputs = outputs + (attn,)
        return outputs


def point_wise_feed_forward_network(d_model_size, dff):
    return nn.Sequential(nn.Linear(d_model_size, dff), nn.ReLU(), nn.Linear(dff, d_model_size))


class EncoderLayer(nn.Module):
    def __init__(self, d_model_size, num_heads, dff, rate=0.1):
        super().__init__()

        self.multi_head_attention = MultiHeadAttention(d_model_size, num_heads)
        self.ffn = point_wise_feed_forward_network(d_model_size, dff)

        self.layernorm1 = nn.LayerNorm(d_model_size, eps=1e-6)
        self.layernorm2 = nn.LayerNorm(d_model_size, eps=1e-6)

        self.dropout1 = nn.Dropout(rate)
        self.dropout2 = nn.Dropout(rate)

    def forward(
        self, x, mask, layer_past=None, attention_mask=None, head_mask=None, use_cache=False, output_attentions=False
    ):
        normed = self.layernorm1(x)
        attn_outputs = self.multi_head_attention(
            normed,
            normed,
            normed,
            mask,
            layer_past=layer_past,
            attention_mask=attention_mask,
            head_mask=head_mask,
            use_cache=use_cache,
            output_attentions=output_attentions,
        )
        attn_output = attn_outputs[0]
        attn_output = self.dropout1(attn_output)
        out1 = x + attn_output

        out2 = self.layernorm2(out1)
        ffn_output = self.ffn(out2)
        ffn_output = self.dropout2(ffn_output)
        out2 = out1 + ffn_output

        outputs = (out2,) + attn_outputs[1:]
        return outputs


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

    config_class = CTRLConfig
    base_model_prefix = "transformer"

    def _init_weights(self, module):
        """Initialize the weights."""
        if isinstance(module, (nn.Linear, Conv1D)):
            # 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=self.config.initializer_range)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.Embedding):
            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
            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)


CTRL_START_DOCSTRING = r"""

    This model inherits from :class:`~transformers.PreTrainedModel`. Check the superclass documentation for the generic
    methods the library implements for all its model (such as downloading or saving, resizing the input embeddings,
    pruning heads etc.)

    This model is also a PyTorch `torch.nn.Module <https://pytorch.org/docs/stable/nn.html#torch.nn.Module>`__
    subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to
    general usage and behavior.

    Parameters:
        config (:class:`~transformers.CTRLConfig`): 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.
"""

CTRL_INPUTS_DOCSTRING = r"""
    Args:
        input_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`):
            :obj:`input_ids_length` = ``sequence_length`` if :obj:`past_key_values` is ``None`` else
            ``past_key_values[0].shape[-2]`` (``sequence_length`` of input past key value states). Indices of input
            sequence tokens in the vocabulary.

            If :obj:`past_key_values` is used, only input IDs that do not have their past calculated should be passed
            as ``input_ids``.

            Indices can be obtained using :class:`~transformers.CTRLTokenizer`. See
            :meth:`transformers.PreTrainedTokenizer.__call__` and :meth:`transformers.PreTrainedTokenizer.encode` for
            details.

            `What are input IDs? <../glossary.html#input-ids>`__
        past_key_values (:obj:`Tuple[Tuple[torch.FloatTensor]]` of length :obj:`config.n_layers`):
            Contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model (see
            :obj:`past_key_values` output below). Can be used to speed up sequential decoding. The ``input_ids`` which
            have their past given to this model should not be passed as input ids as they have already been computed.
        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `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.html#attention-mask>`__
        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
            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:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
            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:`torch.FloatTensor` of shape :obj:`(num_heads,)` or :obj:`(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**.

        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
            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 :obj:`input_ids` indices into associated
            vectors than the model's internal embedding lookup matrix.
        use_cache (:obj:`bool`, `optional`):
            If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
            decoding (see :obj:`past_key_values`).
        output_attentions (:obj:`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 (:obj:`bool`, `optional`):
            Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for
            more detail.
        return_dict (:obj:`bool`, `optional`):
            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
"""


@add_start_docstrings(
    "The bare CTRL Model transformer outputting raw hidden-states without any specific head on top.",
    CTRL_START_DOCSTRING,
)
class CTRLModel(CTRLPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)

        self.d_model_size = config.n_embd
        self.num_layers = config.n_layer

        self.pos_encoding = positional_encoding(config.n_positions, self.d_model_size, torch.float)

        self.w = nn.Embedding(config.vocab_size, config.n_embd)

        self.dropout = nn.Dropout(config.embd_pdrop)
        self.h = nn.ModuleList(
            [EncoderLayer(config.n_embd, config.n_head, config.dff, config.resid_pdrop) for _ in range(config.n_layer)]
        )
        self.layernorm = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)

        self.init_weights()

    def get_input_embeddings(self):
        return self.w

    def set_input_embeddings(self, new_embeddings):
        self.w = new_embeddings

    def _prune_heads(self, heads_to_prune):
        """
        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer}
        """
        for layer, heads in heads_to_prune.items():
            self.h[layer].multi_head_attention.prune_heads(heads)

    @add_start_docstrings_to_model_forward(CTRL_INPUTS_DOCSTRING)
    @add_code_sample_docstrings(
        tokenizer_class=_TOKENIZER_FOR_DOC,
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=BaseModelOutputWithPast,
        config_class=_CONFIG_FOR_DOC,
    )
    def forward(
        self,
        input_ids=None,
        past_key_values=None,
        attention_mask=None,
        token_type_ids=None,
        position_ids=None,
        head_mask=None,
        inputs_embeds=None,
        use_cache=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
    ):

        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        use_cache = use_cache if use_cache is not None else self.config.use_cache
        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_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
        elif input_ids is not None:
            input_shape = input_ids.size()
            input_ids = input_ids.view(-1, input_shape[-1])
            batch_size = input_ids.shape[0]
        elif inputs_embeds is not None:
            input_shape = inputs_embeds.size()[:-1]
            batch_size = inputs_embeds.shape[0]
        else:
            raise ValueError("You have to specify either input_ids or inputs_embeds")

        device = input_ids.device if input_ids is not None else inputs_embeds.device

        if past_key_values is None:
            past_length = 0
            past_key_values = tuple([None] * len(self.h))
        else:
            past_length = past_key_values[0][0].size(-2)
        if position_ids is None:
            position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device)
            position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])

        # Attention mask.
        if attention_mask is not None:
            assert batch_size > 0, "batch_size has to be defined and > 0"
            attention_mask = attention_mask.view(batch_size, -1)
            # We create a 3D attention mask from a 2D tensor mask.
            # Sizes are [batch_size, 1, 1, to_seq_length]
            # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length]
            # this attention mask is more simple than the triangular masking of causal attention
            # used in OpenAI GPT, we just need to prepare the broadcast dimension here.
            attention_mask = attention_mask.unsqueeze(1).unsqueeze(2)

            # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
            # masked positions, this operation will create a tensor which is 0.0 for
            # positions we want to attend and -10000.0 for masked positions.
            # Since we are adding it to the raw scores before the softmax, this is
            # effectively the same as removing these entirely.
            attention_mask = attention_mask.to(dtype=self.dtype)  # fp16 compatibility
            attention_mask = (1.0 - attention_mask) * -10000.0

        # Prepare head mask if needed
        head_mask = self.get_head_mask(head_mask, self.config.n_layer)

        if token_type_ids is not None:
            token_type_ids = token_type_ids.view(-1, input_shape[-1])
            token_type_embeds = self.w(token_type_ids)
            token_type_embeds *= np.sqrt(self.d_model_size)
        else:
            token_type_embeds = 0
        position_ids = position_ids.view(-1, input_shape[-1])

        if inputs_embeds is None:
            inputs_embeds = self.w(input_ids)
        # inputs_embeds = embedded.unsqueeze(0) if len(input_ids.shape)<2 else embedded
        seq_len = input_shape[-1]
        mask = torch.triu(torch.ones(seq_len + past_length, seq_len + past_length), 1).to(device)

        inputs_embeds *= np.sqrt(self.d_model_size)

        pos_embeds = self.pos_encoding[position_ids, :].to(device)

        hidden_states = inputs_embeds + pos_embeds + token_type_embeds

        hidden_states = self.dropout(hidden_states)

        presents = () if use_cache else None
        all_hidden_states = () if output_hidden_states else None
        all_attentions = () if output_attentions else None
        for i, (h, layer_past) in enumerate(zip(self.h, past_key_values)):
            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)
            outputs = h(
                hidden_states,
                mask,
                layer_past=layer_past,
                attention_mask=attention_mask,
                head_mask=head_mask[i],
                use_cache=use_cache,
                output_attentions=output_attentions,
            )
            hidden_states, present = outputs[:2]
            if use_cache is True:
                presents = presents + (present,)

            if output_attentions:
                all_attentions += (outputs[2],)

        hidden_states = self.layernorm(hidden_states)
        if output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        if not return_dict:
            return tuple(v for v in [hidden_states, presents, all_hidden_states, all_attentions] if v is not None)

        return BaseModelOutputWithPast(
            last_hidden_state=hidden_states,
            past_key_values=presents,
            hidden_states=all_hidden_states,
            attentions=all_attentions,
        )


@add_start_docstrings(
    """
    The CTRL Model transformer with a language modeling head on top (linear layer with weights tied to the input
    embeddings).
    """,
    CTRL_START_DOCSTRING,
)
class CTRLLMHeadModel(CTRLPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.transformer = CTRLModel(config)
        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=True)

        self.init_weights()

    def get_output_embeddings(self):
        return self.lm_head

    def set_output_embeddings(self, new_embeddings):
        self.lm_head = new_embeddings

    def prepare_inputs_for_generation(self, input_ids, past=None, use_cache=None, **kwargs):
        # only last token for inputs_ids if past is defined in kwargs
        if past:
            input_ids = input_ids[:, -1].unsqueeze(-1)

        return {"input_ids": input_ids, "past_key_values": past, "use_cache": use_cache}

    @add_start_docstrings_to_model_forward(CTRL_INPUTS_DOCSTRING)
    @add_code_sample_docstrings(
        tokenizer_class=_TOKENIZER_FOR_DOC,
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=CausalLMOutputWithPast,
        config_class=_CONFIG_FOR_DOC,
    )
    def forward(
        self,
        input_ids=None,
        past_key_values=None,
        attention_mask=None,
        token_type_ids=None,
        position_ids=None,
        head_mask=None,
        inputs_embeds=None,
        labels=None,
        use_cache=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
    ):
        r"""
        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
            Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
            ``labels = input_ids`` Indices are selected in ``[-100, 0, ..., config.vocab_size]`` All labels set to
            ``-100`` are ignored (masked), the loss is only computed for labels in ``[0, ..., config.vocab_size]``
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        transformer_outputs = self.transformer(
            input_ids,
            past_key_values=past_key_values,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        hidden_states = transformer_outputs[0]

        lm_logits = self.lm_head(hidden_states)

        loss = None
        if labels is not None:
            # Shift so that tokens < n predict n
            shift_logits = lm_logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            # Flatten the tokens
            loss_fct = CrossEntropyLoss()
            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))

        if not return_dict:
            output = (lm_logits,) + transformer_outputs[1:]
            return ((loss,) + output) if loss is not None else output

        return CausalLMOutputWithPast(
            loss=loss,
            logits=lm_logits,
            past_key_values=transformer_outputs.past_key_values,
            hidden_states=transformer_outputs.hidden_states,
            attentions=transformer_outputs.attentions,
        )

    @staticmethod
    def _reorder_cache(past: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor) -> Tuple[Tuple[torch.Tensor]]:
        """
        This function is used to re-order the :obj:`past_key_values` cache if
        :meth:`~transformers.PreTrainedModel.beam_search` or :meth:`~transformers.PreTrainedModel.beam_sample` is
        called. This is required to match :obj:`past_key_values` with the correct beam_idx at every generation step.
        """
        return tuple(
            tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past)
            for layer_past in past
        )


@add_start_docstrings(
    """
    The CTRL Model transformer with a sequence classification head on top (linear layer).
    :class:`~transformers.CTRLForSequenceClassification` uses the last token in order to do the classification, as
    other causal models (e.g. GPT-2) do. Since it does classification on the last token, it requires to know the
    position of the last token. If a :obj:`pad_token_id` is defined in the configuration, it finds the last token that
    is not a padding token in each row. If no :obj:`pad_token_id` is defined, it simply takes the last value in each
    row of the batch. Since it cannot guess the padding tokens when :obj:`inputs_embeds` are passed instead of
    :obj:`input_ids`, it does the same (take the last value in each row of the batch).
    """,
    CTRL_START_DOCSTRING,
)
class CTRLForSequenceClassification(CTRLPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels
        self.transformer = CTRLModel(config)
        self.classifier = nn.Linear(config.n_embd, self.num_labels, bias=False)

        self.init_weights()

    @add_start_docstrings_to_model_forward(CTRL_INPUTS_DOCSTRING)
    @add_code_sample_docstrings(
        tokenizer_class=_TOKENIZER_FOR_DOC,
        checkpoint=_CHECKPOINT_FOR_DOC,
        output_type=SequenceClassifierOutput,
        config_class=_CONFIG_FOR_DOC,
    )
    def forward(
        self,
        input_ids=None,
        past_key_values=None,
        attention_mask=None,
        token_type_ids=None,
        position_ids=None,
        head_mask=None,
        inputs_embeds=None,
        labels=None,
        use_cache=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
    ):
        r"""
        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`):
            Labels for computing the sequence classification/regression loss. Indices should be in :obj:`[0, ...,
            config.num_labels - 1]`. If :obj:`config.num_labels == 1` a regression loss is computed (Mean-Square loss),
            If :obj:`config.num_labels > 1` a classification loss is computed (Cross-Entropy).
        """

        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        transformer_outputs = self.transformer(
            input_ids,
            past_key_values=past_key_values,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        hidden_states = transformer_outputs[0]
        logits = self.classifier(hidden_states)

        if input_ids is not None:
            batch_size, sequence_length = input_ids.shape[:2]
        else:
            batch_size, sequence_length = inputs_embeds.shape[:2]

        assert (
            self.config.pad_token_id is not None or batch_size == 1
        ), "Cannot handle batch sizes > 1 if no padding token is defined."

        if self.config.pad_token_id is None:
            sequence_lengths = -1
        else:
            if input_ids is not None:
                sequence_lengths = torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1
            else:
                sequence_lengths = -1
                logger.warning(
                    f"{self.__class__.__name__} will not detect padding tokens in `inputs_embeds`. Results may be "
                    f"unexpected if using padding tokens in conjunction with `inputs_embeds.`"
                )

        pooled_logits = logits[range(batch_size), sequence_lengths]

        loss = None
        if labels is not None:
            if self.num_labels == 1:
                #  We are doing regression
                loss_fct = MSELoss()
                loss = loss_fct(pooled_logits.view(-1), labels.to(self.dtype).view(-1))
            else:
                loss_fct = CrossEntropyLoss()
                loss = loss_fct(pooled_logits.view(-1, self.num_labels), labels.view(-1))

        if not return_dict:
            output = (pooled_logits,) + transformer_outputs[2:]
            return ((loss,) + output) if loss is not None else output

        return SequenceClassifierOutput(
            loss=loss,
            logits=pooled_logits,
            hidden_states=transformer_outputs.hidden_states,
            attentions=transformer_outputs.attentions,
        )