# coding=utf-8
# Copyright 2021 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 RoFormer model."""

import math
import os
from typing import Callable, Optional, Union

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

from ...activations import ACT2FN, get_activation
from ...cache_utils import Cache, DynamicCache, EncoderDecoderCache
from ...generation import GenerationMixin
from ...modeling_layers import GradientCheckpointingLayer
from ...modeling_outputs import (
    BaseModelOutputWithPastAndCrossAttentions,
    CausalLMOutputWithCrossAttentions,
    MaskedLMOutput,
    MultipleChoiceModelOutput,
    QuestionAnsweringModelOutput,
    SequenceClassifierOutput,
    TokenClassifierOutput,
)
from ...modeling_utils import PreTrainedModel
from ...pytorch_utils import apply_chunking_to_forward, find_pruneable_heads_and_indices, prune_linear_layer
from ...utils import auto_docstring, logging
from ...utils.deprecation import deprecate_kwarg
from .configuration_roformer import RoFormerConfig


logger = logging.get_logger(__name__)


# Copied from transformers.models.marian.modeling_marian.MarianSinusoidalPositionalEmbedding with Marian->RoFormer
class RoFormerSinusoidalPositionalEmbedding(nn.Embedding):
    """This module produces sinusoidal positional embeddings of any length."""

    def __init__(self, num_positions: int, embedding_dim: int, padding_idx: Optional[int] = None) -> None:
        super().__init__(num_positions, embedding_dim)

    def _init_weight(self):
        """
        Identical to the XLM create_sinusoidal_embeddings except features are not interleaved. The cos features are in
        the 2nd half of the vector. [dim // 2:]
        """
        n_pos, dim = self.weight.shape
        position_enc = np.array(
            [[pos / np.power(10000, 2 * (j // 2) / dim) for j in range(dim)] for pos in range(n_pos)]
        )
        out = torch.empty(n_pos, dim, dtype=self.weight.dtype, requires_grad=False)
        sentinel = dim // 2 if dim % 2 == 0 else (dim // 2) + 1
        out[:, 0:sentinel] = torch.FloatTensor(np.sin(position_enc[:, 0::2]))
        out[:, sentinel:] = torch.FloatTensor(np.cos(position_enc[:, 1::2]))
        self.weight = nn.Parameter(out, requires_grad=False)

    @torch.no_grad()
    def forward(
        self, input_ids_shape: torch.Size, past_key_values_length: int = 0, position_ids: Optional[torch.Tensor] = None
    ) -> torch.Tensor:
        """`input_ids_shape` is expected to be [bsz x seqlen]."""
        if position_ids is None:
            bsz, seq_len = input_ids_shape[:2]
            position_ids = torch.arange(
                past_key_values_length, past_key_values_length + seq_len, dtype=torch.long, device=self.weight.device
            )
        return super().forward(position_ids)


def load_tf_weights_in_roformer(model, config, tf_checkpoint_path):
    """Load tf checkpoints in a pytorch model."""
    try:
        import re

        import numpy as np
        import tensorflow as tf
    except ImportError:
        logger.error(
            "Loading a TensorFlow model in PyTorch, requires TensorFlow to be installed. Please see "
            "https://www.tensorflow.org/install/ for installation instructions."
        )
        raise
    tf_path = os.path.abspath(tf_checkpoint_path)
    logger.info(f"Converting TensorFlow checkpoint from {tf_path}")
    # Load weights from TF model
    init_vars = tf.train.list_variables(tf_path)
    names = []
    arrays = []
    for name, shape in init_vars:
        logger.info(f"Loading TF weight {name} with shape {shape}")
        array = tf.train.load_variable(tf_path, name)
        names.append(name.replace("bert", "roformer"))
        arrays.append(array)

    for name, array in zip(names, arrays):
        name = name.split("/")
        # adam_v and adam_m are variables used in AdamWeightDecayOptimizer to calculated m and v
        # which are not required for using pretrained model
        if any(
            n in ["adam_v", "adam_m", "AdamWeightDecayOptimizer", "AdamWeightDecayOptimizer_1", "global_step"]
            for n in name
        ):
            logger.info(f"Skipping {'/'.join(name)}")
            continue
        pointer = model
        for m_name in name:
            if re.fullmatch(r"[A-Za-z]+_\d+", m_name):
                scope_names = re.split(r"_(\d+)", m_name)
            else:
                scope_names = [m_name]
            if scope_names[0] == "kernel" or scope_names[0] == "gamma":
                pointer = getattr(pointer, "weight")
            elif scope_names[0] == "output_bias" or scope_names[0] == "beta":
                pointer = getattr(pointer, "bias")
            elif scope_names[0] == "output_weights":
                pointer = getattr(pointer, "weight")
            elif scope_names[0] == "squad":
                pointer = getattr(pointer, "classifier")
            else:
                try:
                    pointer = getattr(pointer, scope_names[0])
                except AttributeError:
                    logger.info(f"Skipping {'/'.join(name)}")
                    continue
            if len(scope_names) >= 2:
                num = int(scope_names[1])
                pointer = pointer[num]
        if m_name[-11:] == "_embeddings":
            pointer = getattr(pointer, "weight")
        elif m_name == "kernel":
            array = np.transpose(array)
        try:
            if not pointer.shape == array.shape:
                raise ValueError(f"Pointer shape {pointer.shape} and array shape {array.shape} mismatched")
        except AssertionError as e:
            e.args += (pointer.shape, array.shape)
            raise
        logger.info(f"Initialize PyTorch weight {name}")
        pointer.data = torch.from_numpy(array)
    return model


class RoFormerEmbeddings(nn.Module):
    """Construct the embeddings from word and token_type embeddings."""

    def __init__(self, config):
        super().__init__()
        self.word_embeddings = nn.Embedding(config.vocab_size, config.embedding_size, padding_idx=config.pad_token_id)
        self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.embedding_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.embedding_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

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

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

        if token_type_ids is None:
            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=inputs_embeds.device)

        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 RoFormerSelfAttention(nn.Module):
    def __init__(self, config, layer_idx=None):
        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 = int(config.hidden_size / config.num_attention_heads)
        self.all_head_size = self.num_attention_heads * self.attention_head_size

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

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

        self.is_decoder = config.is_decoder
        self.rotary_value = config.rotary_value
        self.layer_idx = layer_idx

    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
    def forward(
        self,
        hidden_states,
        attention_mask=None,
        sinusoidal_pos=None,
        head_mask=None,
        encoder_hidden_states=None,
        past_key_values=None,
        output_attentions=False,
        cache_position=None,
    ):
        batch_size, seq_length, _ = hidden_states.shape
        query_layer = (
            self.query(hidden_states)
            .view(batch_size, -1, self.num_attention_heads, self.attention_head_size)
            .transpose(1, 2)
        )
        # If this is instantiated as a cross-attention module, the keys
        # and values come from an encoder; the attention mask needs to be
        # such that the encoder's padding tokens are not attended to.
        is_cross_attention = encoder_hidden_states is not None

        is_updated = False
        if past_key_values is not None:
            if isinstance(past_key_values, EncoderDecoderCache):
                is_updated = past_key_values.is_updated.get(self.layer_idx)
                if is_cross_attention:
                    # after the first generated id, we can subsequently re-use all key/value_layer from cache
                    curr_past_key_value = past_key_values.cross_attention_cache
                else:
                    curr_past_key_value = past_key_values.self_attention_cache
            else:
                curr_past_key_value = past_key_values

        current_states = encoder_hidden_states if is_cross_attention else hidden_states
        if is_cross_attention and past_key_values is not None and is_updated:
            # reuse k,v, cross_attentions
            key_layer = curr_past_key_value.layers[self.layer_idx].keys
            value_layer = curr_past_key_value.layers[self.layer_idx].values
        else:
            key_layer = (
                self.key(current_states)
                .view(batch_size, -1, self.num_attention_heads, self.attention_head_size)
                .transpose(1, 2)
            )
            value_layer = (
                self.value(current_states)
                .view(batch_size, -1, self.num_attention_heads, self.attention_head_size)
                .transpose(1, 2)
            )

            # Apply RoPE if self attention
            if not is_cross_attention and sinusoidal_pos is not None:
                if self.rotary_value:
                    query_layer, key_layer, value_layer = self.apply_rotary_position_embeddings(
                        sinusoidal_pos, query_layer, key_layer, value_layer
                    )
                else:
                    query_layer, key_layer = self.apply_rotary_position_embeddings(
                        sinusoidal_pos, query_layer, key_layer
                    )

            if past_key_values is not None:
                # save all key/value_layer to cache to be re-used for fast auto-regressive generation
                cache_position = cache_position if not is_cross_attention else None
                key_layer, value_layer = curr_past_key_value.update(
                    key_layer, value_layer, self.layer_idx, {"cache_position": cache_position}
                )
                # set flag that curr layer for cross-attn is already updated so we can re-use in subsequent calls
                if is_cross_attention and isinstance(past_key_values, EncoderDecoderCache):
                    past_key_values.is_updated[self.layer_idx] = True

        # 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))

        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
        if attention_mask is not None:
            # Apply the attention mask is (precomputed for all layers in RoFormerModel 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).contiguous()
        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
        context_layer = context_layer.view(*new_context_layer_shape)

        return context_layer, attention_probs

    @staticmethod
    def apply_rotary_position_embeddings(sinusoidal_pos, query_layer, key_layer, value_layer=None):
        # https://kexue.fm/archives/8265
        # sin [batch_size, num_heads, sequence_length, embed_size_per_head//2]
        # cos [batch_size, num_heads, sequence_length, embed_size_per_head//2]
        sin, cos = sinusoidal_pos.chunk(2, dim=-1)
        # sin [θ0,θ1,θ2......θd/2-1] -> sin_pos [θ0,θ0,θ1,θ1,θ2,θ2......θd/2-1,θd/2-1]
        sin_pos = torch.stack([sin, sin], dim=-1).reshape_as(sinusoidal_pos)
        # cos [θ0,θ1,θ2......θd/2-1] -> cos_pos [θ0,θ0,θ1,θ1,θ2,θ2......θd/2-1,θd/2-1]
        cos_pos = torch.stack([cos, cos], dim=-1).reshape_as(sinusoidal_pos)
        # rotate_half_query_layer [-q1,q0,-q3,q2......,-qd-1,qd-2]
        rotate_half_query_layer = torch.stack([-query_layer[..., 1::2], query_layer[..., ::2]], dim=-1).reshape_as(
            query_layer
        )
        query_layer = query_layer * cos_pos + rotate_half_query_layer * sin_pos
        # rotate_half_key_layer [-k1,k0,-k3,k2......,-kd-1,kd-2]
        rotate_half_key_layer = torch.stack([-key_layer[..., 1::2], key_layer[..., ::2]], dim=-1).reshape_as(key_layer)
        key_layer = key_layer * cos_pos + rotate_half_key_layer * sin_pos
        if value_layer is not None:
            # rotate_half_value_layer [-v1,v0,-v3,v2......,-vd-1,vd-2]
            rotate_half_value_layer = torch.stack([-value_layer[..., 1::2], value_layer[..., ::2]], dim=-1).reshape_as(
                value_layer
            )
            value_layer = value_layer * cos_pos + rotate_half_value_layer * sin_pos
            return query_layer, key_layer, value_layer
        return query_layer, key_layer


# Copied from transformers.models.bert.modeling_bert.BertSelfOutput with Bert->RoFormer
class RoFormerSelfOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        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: torch.Tensor, input_tensor: torch.Tensor) -> torch.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 RoFormerAttention(nn.Module):
    def __init__(self, config, layer_idx=None):
        super().__init__()
        self.self = RoFormerSelfAttention(config, layer_idx=layer_idx)
        self.output = RoFormerSelfOutput(config)
        self.pruned_heads = set()

    # Copied from transformers.models.bert.modeling_bert.BertAttention.prune_heads
    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)

    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
    def forward(
        self,
        hidden_states,
        attention_mask=None,
        sinusoidal_pos=None,
        head_mask=None,
        encoder_hidden_states=None,
        past_key_values=None,
        output_attentions=False,
        cache_position=None,
    ):
        self_outputs = self.self(
            hidden_states,
            attention_mask=attention_mask,
            sinusoidal_pos=sinusoidal_pos,
            head_mask=head_mask,
            encoder_hidden_states=encoder_hidden_states,
            past_key_values=past_key_values,
            output_attentions=output_attentions,
            cache_position=cache_position,
        )
        attention_output = self.output(self_outputs[0], hidden_states)
        outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them
        return outputs


# Copied from transformers.models.bert.modeling_bert.BertIntermediate with Bert->RoFormer
class RoFormerIntermediate(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
        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: torch.Tensor) -> torch.Tensor:
        hidden_states = self.dense(hidden_states)
        hidden_states = self.intermediate_act_fn(hidden_states)
        return hidden_states


# Copied from transformers.models.bert.modeling_bert.BertOutput with Bert->RoFormer
class RoFormerOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
        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: torch.Tensor, input_tensor: torch.Tensor) -> torch.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 RoFormerLayer(GradientCheckpointingLayer):
    def __init__(self, config, layer_idx=None):
        super().__init__()
        self.chunk_size_feed_forward = config.chunk_size_feed_forward
        self.seq_len_dim = 1
        self.attention = RoFormerAttention(config, layer_idx)
        self.is_decoder = config.is_decoder
        self.add_cross_attention = config.add_cross_attention
        if self.add_cross_attention:
            if not self.is_decoder:
                raise ValueError(f"{self} should be used as a decoder model if cross attention is added")
            self.crossattention = RoFormerAttention(config, layer_idx)
        self.intermediate = RoFormerIntermediate(config)
        self.output = RoFormerOutput(config)

    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
    def forward(
        self,
        hidden_states,
        attention_mask=None,
        sinusoidal_pos=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_values=None,
        output_attentions=False,
        cache_position=None,
    ):
        self_attention_outputs = self.attention(
            hidden_states,
            attention_mask=attention_mask,
            sinusoidal_pos=sinusoidal_pos,
            head_mask=head_mask,
            output_attentions=output_attentions,
            past_key_values=past_key_values,
            cache_position=cache_position,
        )
        attention_output = self_attention_outputs[0]
        outputs = self_attention_outputs[1:]  # add self attentions if we output attention weights

        if self.is_decoder and encoder_hidden_states is not None:
            if not hasattr(self, "crossattention"):
                raise ValueError(
                    f"If `encoder_hidden_states` are passed, {self} has to be instantiated with cross-attention "
                    "layers by setting `config.add_cross_attention=True`"
                )

            cross_attention_outputs = self.crossattention(
                attention_output,
                attention_mask=encoder_attention_mask,
                sinusoidal_pos=sinusoidal_pos,
                head_mask=head_mask,
                encoder_hidden_states=encoder_hidden_states,
                past_key_values=past_key_values,
                output_attentions=output_attentions,
                cache_position=cache_position,
            )
            attention_output = cross_attention_outputs[0]
            outputs = outputs + cross_attention_outputs[1:]  # add cross 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
        )
        return (layer_output,) + 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 RoFormerEncoder(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.embed_positions = RoFormerSinusoidalPositionalEmbedding(
            config.max_position_embeddings, config.hidden_size // config.num_attention_heads
        )
        self.layer = nn.ModuleList([RoFormerLayer(config, layer_idx=i) for i in range(config.num_hidden_layers)])
        self.gradient_checkpointing = False

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        head_mask=None,
        encoder_hidden_states=None,
        encoder_attention_mask=None,
        past_key_values=None,
        use_cache=None,
        output_attentions=False,
        output_hidden_states=False,
        return_dict=True,
        cache_position=None,
    ):
        if self.gradient_checkpointing and self.training:
            if use_cache:
                logger.warning_once(
                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                )
                use_cache = False

        if use_cache and past_key_values is None:
            past_key_values = EncoderDecoderCache(DynamicCache(config=self.config), DynamicCache(config=self.config))
        if use_cache and isinstance(past_key_values, tuple):
            logger.warning_once(
                "Passing a tuple of `past_key_values` is deprecated and will be removed in Transformers v4.58.0. "
                "You should pass an instance of `EncoderDecoderCache` instead, e.g. "
                "`past_key_values=EncoderDecoderCache.from_legacy_cache(past_key_values)`."
            )
            past_key_values = EncoderDecoderCache.from_legacy_cache(past_key_values)

        all_hidden_states = () if output_hidden_states else None
        all_self_attentions = () if output_attentions else None
        all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None

        past_key_values_length = past_key_values.get_seq_length() if past_key_values is not None else 0

        # [sequence_length, embed_size_per_head] -> [batch_size, num_heads, sequence_length, embed_size_per_head]
        sinusoidal_pos = self.embed_positions(hidden_states.shape[:-1], past_key_values_length)[None, None, :, :]

        for i, layer_module in enumerate(self.layer):
            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            layer_head_mask = head_mask[i] if head_mask is not None else None

            layer_outputs = layer_module(
                hidden_states,
                attention_mask,
                sinusoidal_pos,
                layer_head_mask,
                encoder_hidden_states,
                encoder_attention_mask,
                past_key_values,
                output_attentions,
                cache_position,
            )

            hidden_states = layer_outputs[0]
            if output_attentions:
                all_self_attentions = all_self_attentions + (layer_outputs[1],)
                if self.config.add_cross_attention:
                    all_cross_attentions = all_cross_attentions + (layer_outputs[2],)

        if output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        if not return_dict:
            return tuple(
                v
                for v in [
                    hidden_states,
                    past_key_values,
                    all_hidden_states,
                    all_self_attentions,
                    all_cross_attentions,
                ]
                if v is not None
            )
        return BaseModelOutputWithPastAndCrossAttentions(
            last_hidden_state=hidden_states,
            past_key_values=past_key_values,
            hidden_states=all_hidden_states,
            attentions=all_self_attentions,
            cross_attentions=all_cross_attentions,
        )


# Copied from transformers.models.xlm.modeling_xlm.XLMSequenceSummary with XLM->RoFormer
class RoFormerSequenceSummary(nn.Module):
    r"""
    Compute a single vector summary of a sequence hidden states.

    Args:
        config ([`RoFormerConfig`]):
            The config used by the model. Relevant arguments in the config class of the model are (refer to the actual
            config class of your model for the default values it uses):

            - **summary_type** (`str`) -- The method to use to make this summary. Accepted values are:

                - `"last"` -- 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** (`bool`) -- Add a projection after the vector extraction.
            - **summary_proj_to_labels** (`bool`) -- If `True`, the projection outputs to `config.num_labels` classes
              (otherwise to `config.hidden_size`).
            - **summary_activation** (`Optional[str]`) -- Set to `"tanh"` to add a tanh activation to the output,
              another string or `None` will add no activation.
            - **summary_first_dropout** (`float`) -- Optional dropout probability before the projection and activation.
            - **summary_last_dropout** (`float`)-- Optional dropout probability after the projection and activation.
    """

    def __init__(self, config: RoFormerConfig):
        super().__init__()

        self.summary_type = getattr(config, "summary_type", "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.summary = nn.Identity()
        if hasattr(config, "summary_use_proj") and config.summary_use_proj:
            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 = nn.Linear(config.hidden_size, num_classes)

        activation_string = getattr(config, "summary_activation", None)
        self.activation: Callable = get_activation(activation_string) if activation_string else nn.Identity()

        self.first_dropout = nn.Identity()
        if hasattr(config, "summary_first_dropout") and config.summary_first_dropout > 0:
            self.first_dropout = nn.Dropout(config.summary_first_dropout)

        self.last_dropout = nn.Identity()
        if hasattr(config, "summary_last_dropout") and config.summary_last_dropout > 0:
            self.last_dropout = nn.Dropout(config.summary_last_dropout)

    def forward(
        self, hidden_states: torch.FloatTensor, cls_index: Optional[torch.LongTensor] = None
    ) -> torch.FloatTensor:
        """
        Compute a single vector summary of a sequence hidden states.

        Args:
            hidden_states (`torch.FloatTensor` of shape `[batch_size, seq_len, hidden_size]`):
                The hidden states of the last layer.
            cls_index (`torch.LongTensor` of shape `[batch_size]` or `[batch_size, ...]` where ... are optional leading dimensions of `hidden_states`, *optional*):
                Used if `summary_type == "cls_index"` and takes the last token of the sequence as classification token.

        Returns:
            `torch.FloatTensor`: The summary of the sequence hidden states.
        """
        if self.summary_type == "last":
            output = hidden_states[:, -1]
        elif self.summary_type == "first":
            output = hidden_states[:, 0]
        elif self.summary_type == "mean":
            output = hidden_states.mean(dim=1)
        elif self.summary_type == "cls_index":
            if cls_index is None:
                cls_index = torch.full_like(
                    hidden_states[..., :1, :],
                    hidden_states.shape[-2] - 1,
                    dtype=torch.long,
                )
            else:
                cls_index = cls_index.unsqueeze(-1).unsqueeze(-1)
                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 = hidden_states.gather(-2, cls_index).squeeze(-2)  # shape (bsz, XX, hidden_size)
        elif self.summary_type == "attn":
            raise NotImplementedError

        output = self.first_dropout(output)
        output = self.summary(output)
        output = self.activation(output)
        output = self.last_dropout(output)

        return output


class RoFormerPredictionHeadTransform(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.embedding_size)
        if isinstance(config.hidden_act, str):
            self.transform_act_fn = ACT2FN[config.hidden_act]
        else:
            self.transform_act_fn = config.hidden_act
        self.LayerNorm = nn.LayerNorm(config.embedding_size, eps=config.layer_norm_eps)

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


class RoFormerLMPredictionHead(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.transform = RoFormerPredictionHeadTransform(config)

        # The output weights are the same as the input embeddings, but there is
        # an output-only bias for each token.
        self.decoder = nn.Linear(config.embedding_size, config.vocab_size, bias=False)

        self.bias = nn.Parameter(torch.zeros(config.vocab_size))

        # Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
        self.decoder.bias = self.bias

    def _tie_weights(self) -> None:
        self.decoder.bias = self.bias

    def forward(self, hidden_states):
        hidden_states = self.transform(hidden_states)
        hidden_states = self.decoder(hidden_states)
        return hidden_states


# Copied from transformers.models.bert.modeling_bert.BertOnlyMLMHead with Bert->RoFormer
class RoFormerOnlyMLMHead(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.predictions = RoFormerLMPredictionHead(config)

    def forward(self, sequence_output: torch.Tensor) -> torch.Tensor:
        prediction_scores = self.predictions(sequence_output)
        return prediction_scores


@auto_docstring
class RoFormerPreTrainedModel(PreTrainedModel):
    config: RoFormerConfig
    load_tf_weights = load_tf_weights_in_roformer
    base_model_prefix = "roformer"
    supports_gradient_checkpointing = True

    def _init_weights(self, module):
        """Initialize the weights"""
        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=self.config.initializer_range)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, RoFormerSinusoidalPositionalEmbedding):
            module._init_weight()
        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)
        elif isinstance(module, RoFormerLMPredictionHead):
            module.bias.data.zero_()


@auto_docstring(
    custom_intro="""

    The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of
    cross-attention is added between the self-attention layers, following the architecture described in [Attention is
    all you need](https://huggingface.co/papers/1706.03762) by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
    Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.

    To behave as an decoder the model needs to be initialized with the `is_decoder` argument of the configuration set
    to `True`. To be used in a Seq2Seq model, the model needs to initialized with both `is_decoder` argument and
    `add_cross_attention` set to `True`; an `encoder_hidden_states` is then expected as an input to the forward pass.
    """
)
class RoFormerModel(RoFormerPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.config = config
        self.embeddings = RoFormerEmbeddings(config)

        if config.embedding_size != config.hidden_size:
            self.embeddings_project = nn.Linear(config.embedding_size, config.hidden_size)

        self.encoder = RoFormerEncoder(config)

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

    def get_input_embeddings(self):
        return self.embeddings.word_embeddings

    def set_input_embeddings(self, value):
        self.embeddings.word_embeddings = value

    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} See base
        class PreTrainedModel
        """
        for layer, heads in heads_to_prune.items():
            self.encoder.layer[layer].attention.prune_heads(heads)

    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.FloatTensor] = None,
        token_type_ids: Optional[torch.LongTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        encoder_hidden_states: Optional[torch.FloatTensor] = None,
        encoder_attention_mask: Optional[torch.FloatTensor] = None,
        past_key_values: Optional[Cache] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        cache_position: Optional[torch.Tensor] = None,
    ) -> Union[BaseModelOutputWithPastAndCrossAttentions, tuple[torch.Tensor]]:
        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 self.config.is_decoder:
            use_cache = use_cache if use_cache is not None else self.config.use_cache
        else:
            use_cache = False

        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:
            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
            input_shape = input_ids.size()
        elif inputs_embeds is not None:
            input_shape = inputs_embeds.size()[:-1]
        else:
            raise ValueError("You have to specify either input_ids or inputs_embeds")

        batch_size, seq_length = input_shape
        device = input_ids.device if input_ids is not None else inputs_embeds.device

        past_key_values_length = 0
        if past_key_values is not None:
            past_key_values_length = (
                past_key_values[0][0].shape[-2]
                if not isinstance(past_key_values, Cache)
                else past_key_values.get_seq_length()
            )

        if attention_mask is None:
            attention_mask = torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)
        if token_type_ids is None:
            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=device)

        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
        # ourselves in which case we just need to make it broadcastable to all heads.
        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape)

        # If a 2D or 3D attention mask is provided for the cross-attention
        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
        if self.config.is_decoder and encoder_hidden_states is not None:
            encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
            if encoder_attention_mask is None:
                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
            encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
        else:
            encoder_extended_attention_mask = None

        # Prepare head mask if needed
        # 1.0 in head_mask indicate we keep the head
        # attention_probs has shape bsz x n_heads x N x N
        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)

        embedding_output = self.embeddings(
            input_ids=input_ids, token_type_ids=token_type_ids, inputs_embeds=inputs_embeds
        )
        if hasattr(self, "embeddings_project"):
            embedding_output = self.embeddings_project(embedding_output)

        encoder_outputs = self.encoder(
            embedding_output,
            attention_mask=extended_attention_mask,
            head_mask=head_mask,
            encoder_hidden_states=encoder_hidden_states,
            encoder_attention_mask=encoder_extended_attention_mask,
            past_key_values=past_key_values,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            cache_position=cache_position,
        )
        sequence_output = encoder_outputs[0]

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

        return BaseModelOutputWithPastAndCrossAttentions(
            last_hidden_state=sequence_output,
            past_key_values=encoder_outputs.past_key_values,
            hidden_states=encoder_outputs.hidden_states,
            attentions=encoder_outputs.attentions,
            cross_attentions=encoder_outputs.cross_attentions,
        )


@auto_docstring
class RoFormerForMaskedLM(RoFormerPreTrainedModel):
    _tied_weights_keys = ["cls.predictions.decoder.bias", "cls.predictions.decoder.weight"]

    def __init__(self, config):
        super().__init__(config)

        if config.is_decoder:
            logger.warning(
                "If you want to use `RoFormerForMaskedLM` make sure `config.is_decoder=False` for "
                "bi-directional self-attention."
            )

        self.roformer = RoFormerModel(config)
        self.cls = RoFormerOnlyMLMHead(config)

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

    def get_output_embeddings(self):
        return self.cls.predictions.decoder

    def set_output_embeddings(self, new_embeddings):
        self.cls.predictions.decoder = new_embeddings
        self.cls.predictions.bias = new_embeddings.bias

    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.FloatTensor] = None,
        token_type_ids: Optional[torch.LongTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        encoder_hidden_states: Optional[torch.FloatTensor] = None,
        encoder_attention_mask: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[MaskedLMOutput, tuple[torch.Tensor]]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for computing the masked language modeling loss. Indices should be in `[-100, 0, ...,
            config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored (masked), the
            loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.roformer(
            input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            encoder_hidden_states=encoder_hidden_states,
            encoder_attention_mask=encoder_attention_mask,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

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

        masked_lm_loss = None
        if labels is not None:
            loss_fct = CrossEntropyLoss()  # -100 index = padding token
            masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))

        if not return_dict:
            output = (prediction_scores,) + outputs[1:]
            return ((masked_lm_loss,) + output) if masked_lm_loss is not None else output

        return MaskedLMOutput(
            loss=masked_lm_loss,
            logits=prediction_scores,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

    def prepare_inputs_for_generation(self, input_ids, attention_mask=None, **model_kwargs):
        input_shape = input_ids.shape
        effective_batch_size = input_shape[0]

        #  add a dummy token
        assert self.config.pad_token_id is not None, "The PAD token should be defined for generation"
        attention_mask = torch.cat([attention_mask, attention_mask.new_zeros((attention_mask.shape[0], 1))], dim=-1)
        dummy_token = torch.full(
            (effective_batch_size, 1), self.config.pad_token_id, dtype=torch.long, device=input_ids.device
        )
        input_ids = torch.cat([input_ids, dummy_token], dim=1)

        return {"input_ids": input_ids, "attention_mask": attention_mask}


@auto_docstring(
    custom_intro="""
    RoFormer Model with a `language modeling` head on top for CLM fine-tuning.
    """
)
class RoFormerForCausalLM(RoFormerPreTrainedModel, GenerationMixin):
    _tied_weights_keys = ["cls.predictions.decoder.bias", "cls.predictions.decoder.weight"]

    def __init__(self, config):
        super().__init__(config)

        if not config.is_decoder:
            logger.warning("If you want to use `RoFormerForCausalLM` as a standalone, add `is_decoder=True.`")

        self.roformer = RoFormerModel(config)
        self.cls = RoFormerOnlyMLMHead(config)

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

    def get_output_embeddings(self):
        return self.cls.predictions.decoder

    def set_output_embeddings(self, new_embeddings):
        self.cls.predictions.decoder = new_embeddings
        self.cls.predictions.bias = new_embeddings.bias

    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.FloatTensor] = None,
        token_type_ids: Optional[torch.LongTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        encoder_hidden_states: Optional[torch.FloatTensor] = None,
        encoder_attention_mask: Optional[torch.FloatTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        cross_attn_head_mask: Optional[torch.Tensor] = None,
        past_key_values: Optional[Cache] = None,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        cache_position: Optional[torch.Tensor] = None,
        **kwargs,
    ) -> Union[CausalLMOutputWithCrossAttentions, tuple[torch.Tensor]]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for computing the left-to-right language modeling loss (next word prediction). Indices should be in
            `[-100, 0, ..., config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are
            ignored (masked), the loss is only computed for the tokens with labels n `[0, ..., config.vocab_size]`.

        Example:

        ```python
        >>> from transformers import AutoTokenizer, RoFormerForCausalLM, RoFormerConfig
        >>> import torch

        >>> tokenizer = AutoTokenizer.from_pretrained("junnyu/roformer_chinese_base")
        >>> config = RoFormerConfig.from_pretrained("junnyu/roformer_chinese_base")
        >>> config.is_decoder = True
        >>> model = RoFormerForCausalLM.from_pretrained("junnyu/roformer_chinese_base", config=config)

        >>> inputs = tokenizer("今天天气非常好。", return_tensors="pt")
        >>> outputs = model(**inputs)

        >>> prediction_logits = outputs.logits
        ```"""
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.roformer(
            input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            encoder_hidden_states=encoder_hidden_states,
            encoder_attention_mask=encoder_attention_mask,
            past_key_values=past_key_values,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            cache_position=cache_position,
        )

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

        lm_loss = None
        if labels is not None:
            lm_loss = self.loss_function(
                prediction_scores,
                labels,
                vocab_size=self.config.vocab_size,
                **kwargs,
            )

        if not return_dict:
            output = (prediction_scores,) + outputs[1:]
            return ((lm_loss,) + output) if lm_loss is not None else output

        return CausalLMOutputWithCrossAttentions(
            loss=lm_loss,
            logits=prediction_scores,
            past_key_values=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
            cross_attentions=outputs.cross_attentions,
        )


class RoFormerClassificationHead(nn.Module):
    """Head for sentence-level classification tasks."""

    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
        self.out_proj = nn.Linear(config.hidden_size, config.num_labels)

        self.config = config

    def forward(self, features, **kwargs):
        x = features[:, 0, :]  # take <s> token (equiv. to [CLS])
        x = self.dropout(x)
        x = self.dense(x)
        x = ACT2FN[self.config.hidden_act](x)
        x = self.dropout(x)
        x = self.out_proj(x)
        return x


@auto_docstring(
    custom_intro="""
    RoFormer Model transformer with a sequence classification/regression head on top (a linear layer on top of the
    pooled output) e.g. for GLUE tasks.
    """
)
class RoFormerForSequenceClassification(RoFormerPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels
        self.roformer = RoFormerModel(config)
        self.classifier = RoFormerClassificationHead(config)

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

    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.FloatTensor] = None,
        token_type_ids: Optional[torch.LongTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[SequenceClassifierOutput, tuple[torch.Tensor]]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
            `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

        outputs = self.roformer(
            input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        sequence_output = outputs[0]
        logits = self.classifier(sequence_output)

        loss = None
        if labels is not None:
            if self.config.problem_type is None:
                if self.num_labels == 1:
                    self.config.problem_type = "regression"
                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
                    self.config.problem_type = "single_label_classification"
                else:
                    self.config.problem_type = "multi_label_classification"

            if self.config.problem_type == "regression":
                loss_fct = MSELoss()
                if self.num_labels == 1:
                    loss = loss_fct(logits.squeeze(), labels.squeeze())
                else:
                    loss = loss_fct(logits, labels)
            elif self.config.problem_type == "single_label_classification":
                loss_fct = CrossEntropyLoss()
                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
            elif self.config.problem_type == "multi_label_classification":
                loss_fct = BCEWithLogitsLoss()
                loss = loss_fct(logits, labels)
        if not return_dict:
            output = (logits,) + outputs[1:]
            return ((loss,) + output) if loss is not None else output

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


@auto_docstring
class RoFormerForMultipleChoice(RoFormerPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)

        self.roformer = RoFormerModel(config)
        self.sequence_summary = RoFormerSequenceSummary(config)
        self.classifier = nn.Linear(config.hidden_size, 1)

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

    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.FloatTensor] = None,
        token_type_ids: Optional[torch.LongTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[MultipleChoiceModelOutput, tuple[torch.Tensor]]:
        r"""
        input_ids (`torch.LongTensor` of shape `(batch_size, num_choices, sequence_length)`):
            Indices of input sequence tokens in the vocabulary.

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

            [What are input IDs?](../glossary#input-ids)
        token_type_ids (`torch.LongTensor` of shape `(batch_size, num_choices, 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#token-type-ids)
        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, num_choices, sequence_length, hidden_size)`, *optional*):
            Optionally, instead of passing `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.
        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for computing the multiple choice classification loss. Indices should be in `[0, ...,
            num_choices-1]` where `num_choices` is the size of the second dimension of the input tensors. (See
            `input_ids` above)
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
        num_choices = input_ids.shape[1] if input_ids is not None else inputs_embeds.shape[1]

        input_ids = input_ids.view(-1, input_ids.size(-1)) if input_ids is not None else None
        attention_mask = attention_mask.view(-1, attention_mask.size(-1)) if attention_mask is not None else None
        token_type_ids = token_type_ids.view(-1, token_type_ids.size(-1)) if token_type_ids is not None else None

        inputs_embeds = (
            inputs_embeds.view(-1, inputs_embeds.size(-2), inputs_embeds.size(-1))
            if inputs_embeds is not None
            else None
        )

        outputs = self.roformer(
            input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        sequence_output = outputs[0]

        pooled_output = self.sequence_summary(sequence_output)
        logits = self.classifier(pooled_output)
        reshaped_logits = logits.view(-1, num_choices)

        loss = None
        if labels is not None:
            loss_fct = CrossEntropyLoss()
            loss = loss_fct(reshaped_logits, labels)

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

        return MultipleChoiceModelOutput(
            loss=loss,
            logits=reshaped_logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )


@auto_docstring
class RoFormerForTokenClassification(RoFormerPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels

        self.roformer = RoFormerModel(config)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
        self.classifier = nn.Linear(config.hidden_size, config.num_labels)

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

    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.FloatTensor] = None,
        token_type_ids: Optional[torch.LongTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[TokenClassifierOutput, tuple[torch.Tensor]]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`.
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.roformer(
            input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        sequence_output = outputs[0]

        sequence_output = self.dropout(sequence_output)
        logits = self.classifier(sequence_output)

        loss = None
        if labels is not None:
            loss_fct = CrossEntropyLoss()
            loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))

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

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


@auto_docstring
class RoFormerForQuestionAnswering(RoFormerPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)

        config.num_labels = 2
        self.num_labels = config.num_labels

        self.roformer = RoFormerModel(config)
        self.qa_outputs = nn.Linear(config.hidden_size, config.num_labels)

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

    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.FloatTensor] = None,
        token_type_ids: Optional[torch.LongTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        start_positions: Optional[torch.LongTensor] = None,
        end_positions: Optional[torch.LongTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[QuestionAnsweringModelOutput, tuple[torch.Tensor]]:
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.roformer(
            input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        sequence_output = outputs[0]

        logits = self.qa_outputs(sequence_output)
        start_logits, end_logits = logits.split(1, dim=-1)
        start_logits = start_logits.squeeze(-1)
        end_logits = end_logits.squeeze(-1)

        total_loss = None
        if start_positions is not None and end_positions is not None:
            # If we are on multi-GPU, split add a dimension
            if len(start_positions.size()) > 1:
                start_positions = start_positions.squeeze(-1)
            if len(end_positions.size()) > 1:
                end_positions = end_positions.squeeze(-1)
            # sometimes the start/end positions are outside our model inputs, we ignore these terms
            ignored_index = start_logits.size(1)
            start_positions = start_positions.clamp(0, ignored_index)
            end_positions = end_positions.clamp(0, ignored_index)

            loss_fct = CrossEntropyLoss(ignore_index=ignored_index)
            start_loss = loss_fct(start_logits, start_positions)
            end_loss = loss_fct(end_logits, end_positions)
            total_loss = (start_loss + end_loss) / 2

        if not return_dict:
            output = (start_logits, end_logits) + outputs[1:]
            return ((total_loss,) + output) if total_loss is not None else output

        return QuestionAnsweringModelOutput(
            loss=total_loss,
            start_logits=start_logits,
            end_logits=end_logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )


__all__ = [
    "RoFormerForCausalLM",
    "RoFormerForMaskedLM",
    "RoFormerForMultipleChoice",
    "RoFormerForQuestionAnswering",
    "RoFormerForSequenceClassification",
    "RoFormerForTokenClassification",
    "RoFormerLayer",
    "RoFormerModel",
    "RoFormerPreTrainedModel",
    "load_tf_weights_in_roformer",
]