# Copyright 2021 The Eleuther AI and 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 GPT Neo model."""

from typing import Union

import torch
from torch import nn
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss

from ... import initialization as init
from ...activations import ACT2FN
from ...cache_utils import Cache, DynamicCache
from ...generation import GenerationMixin
from ...modeling_attn_mask_utils import AttentionMaskConverter
from ...modeling_flash_attention_utils import flash_attn_supports_top_left_mask, is_flash_attn_available
from ...modeling_layers import GradientCheckpointingLayer
from ...modeling_outputs import (
    BaseModelOutputWithPast,
    BaseModelOutputWithPastAndCrossAttentions,
    CausalLMOutputWithCrossAttentions,
    CausalLMOutputWithPast,
    QuestionAnsweringModelOutput,
    SequenceClassifierOutputWithPast,
    TokenClassifierOutput,
)
from ...modeling_utils import PreTrainedModel
from ...utils import (
    auto_docstring,
    is_torch_flex_attn_available,
    logging,
)
from ...utils.generic import is_flash_attention_requested
from .configuration_gpt_neo import GPTNeoConfig


if is_torch_flex_attn_available():
    from torch.nn.attention.flex_attention import BlockMask

    from ...integrations.flex_attention import make_flex_block_causal_mask


if is_flash_attn_available():
    from ...modeling_flash_attention_utils import _flash_attention_forward


logger = logging.get_logger(__name__)


class GPTNeoSelfAttention(nn.Module):
    def __init__(self, config, attention_type, layer_id=None):
        super().__init__()
        self.config = config

        max_positions = config.max_position_embeddings
        bias = torch.tril(torch.ones((max_positions, max_positions), dtype=bool)).view(
            1, 1, max_positions, max_positions
        )

        # local causal self attention is a sliding window where each token can only attend to the previous
        # window_size tokens. This is implemented by updating the causal mask such that for each token
        # all other tokens are masked except the previous window_size tokens.
        self.attention_type = attention_type
        if attention_type == "local":
            bias = torch.bitwise_xor(bias, torch.tril(bias, -config.window_size))

        self.register_buffer("bias", bias, persistent=False)

        self.attn_dropout = nn.Dropout(float(config.attention_dropout))
        self.resid_dropout = nn.Dropout(float(config.resid_dropout))
        self.is_causal = True
        self.layer_id = layer_id

        self.embed_dim = config.hidden_size
        self.num_heads = config.num_heads
        self.head_dim = self.embed_dim // self.num_heads
        if self.head_dim * self.num_heads != self.embed_dim:
            raise ValueError(
                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:"
                f" {self.num_heads})."
            )

        self.k_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=False)
        self.v_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=False)
        self.q_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=False)
        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=True)

    def _split_heads(self, tensor, num_heads, attn_head_size):
        """
        Splits hidden_size dim into attn_head_size and num_heads
        """
        new_shape = tensor.size()[:-1] + (num_heads, attn_head_size)
        tensor = tensor.view(new_shape)
        return tensor.permute(0, 2, 1, 3)  # (batch, head, seq_length, head_features)

    def _merge_heads(self, tensor, num_heads, attn_head_size):
        """
        Merges attn_head_size dim and num_attn_heads dim into hidden_size
        """
        tensor = tensor.permute(0, 2, 1, 3).contiguous()
        new_shape = tensor.size()[:-2] + (num_heads * attn_head_size,)
        return tensor.view(new_shape)

    def _attn(self, query, key, value, attention_mask=None):
        # Keep the attention weights computation in fp32 to avoid overflow issues
        query = query.to(torch.float32)
        key = key.to(torch.float32)

        attn_weights = torch.matmul(query, key.transpose(-1, -2))

        # Apply sliding window masking for local attention layers
        query_length, key_length = query.size(-2), key.size(-2)
        causal_mask = self.bias[:, :, key_length - query_length : key_length, :key_length]
        mask_value = torch.finfo(attn_weights.dtype).min
        # Need to be a tensor, otherwise we get error: `RuntimeError: expected scalar type float but found double`.
        # Need to be on the same device, otherwise `RuntimeError: ..., x and y to be on the same device`
        mask_value = torch.tensor(mask_value, dtype=attn_weights.dtype, device=attn_weights.device)
        attn_weights = torch.where(causal_mask, attn_weights, mask_value)

        if attention_mask is not None:  # no matter the length, we just slice it
            causal_mask = attention_mask[:, :, :, : key.shape[-2]]
            attn_weights = attn_weights + causal_mask

        attn_weights = nn.functional.softmax(attn_weights, dim=-1)
        attn_weights = attn_weights.to(value.dtype)
        attn_weights = self.attn_dropout(attn_weights)

        attn_output = torch.matmul(attn_weights, value)

        return attn_output, attn_weights

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        layer_past=None,
        use_cache=False,
        output_attentions=False,
        cache_position=None,
    ):
        query = self.q_proj(hidden_states)
        key = self.k_proj(hidden_states)
        value = self.v_proj(hidden_states)

        query = self._split_heads(query, self.num_heads, self.head_dim)
        key = self._split_heads(key, self.num_heads, self.head_dim)
        value = self._split_heads(value, self.num_heads, self.head_dim)

        if layer_past is not None:
            cache_kwargs = {"cache_position": cache_position}
            key, value = layer_past.update(key, value, self.layer_id, cache_kwargs)

        attn_output, attn_weights = self._attn(query, key, value, attention_mask)

        attn_output = self._merge_heads(attn_output, self.num_heads, self.head_dim)
        attn_output = self.out_proj(attn_output)
        attn_output = self.resid_dropout(attn_output)

        return attn_output, attn_weights


class GPTNeoFlashAttention2(GPTNeoSelfAttention):
    """
    GPTNeo flash attention module. This module inherits from `GPTNeoSelfAttention` as the weights of the module stays
    untouched. The only required change would be on the forward pass where it needs to correctly call the public API of
    flash attention and deal with padding tokens in case the input contains any of them.
    """

    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)

        # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
        # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignment, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
        # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
        self._flash_attn_uses_top_left_mask = flash_attn_supports_top_left_mask()

    def forward(
        self,
        hidden_states,
        attention_mask=None,
        layer_past=None,
        use_cache=False,
        output_attentions=False,
        cache_position=None,
    ):
        bsz, _, _ = hidden_states.size()

        query = self.q_proj(hidden_states)
        key = self.k_proj(hidden_states)
        value = self.v_proj(hidden_states)

        query = self._split_heads(query, self.num_heads, self.head_dim)
        key = self._split_heads(key, self.num_heads, self.head_dim)
        value = self._split_heads(value, self.num_heads, self.head_dim)

        if layer_past is not None:
            cache_kwargs = {"cache_position": cache_position}
            key, value = layer_past.update(key, value, self.layer_id, cache_kwargs)

        query_length = query.shape[2]
        tgt_len = key.shape[2]

        # Flash attention requires the input to have the shape
        # batch_size x seq_length x head_dim x hidden_dim
        query = query.transpose(1, 2).view(bsz, query_length, self.num_heads, self.head_dim)
        key = key.transpose(1, 2).view(bsz, tgt_len, self.num_heads, self.head_dim)
        value = value.transpose(1, 2).view(bsz, tgt_len, self.num_heads, self.head_dim)

        attn_dropout = self.config.attention_dropout if self.training else 0.0

        if attention_mask is not None:  # no matter the length, we just slice it
            attention_mask = attention_mask[:, :, :, : key.shape[-2]]

        # In PEFT, usually we cast the layer norms in float32 for training stability reasons
        # therefore the input hidden states gets silently casted in float32. Hence, we need
        # cast them back in the correct dtype just to be sure everything works as expected.
        # This might slowdown training & inference so it is recommended to not cast the LayerNorms
        # in fp32. (LlamaRMSNorm handles it correctly)

        device_type = query.device.type if query.device.type != "mps" else "cpu"
        if query.dtype == torch.float32:
            if torch.is_autocast_enabled():
                target_dtype = torch.get_autocast_dtype(device_type)
            # Handle the case where the model is quantized
            elif hasattr(self.config, "_is_quantized"):
                target_dtype = self.config.dtype
            else:
                target_dtype = self.q_proj.weight.dtype

            logger.warning_once(
                f"The input hidden states seems to be silently casted in float32, this might be related to"
                f" the fact you have upcasted embedding or layer norm layers in float32. We will cast back the input in"
                f" {target_dtype}."
            )

            query = query.to(target_dtype)
            key = key.to(target_dtype)
            value = value.to(target_dtype)

        attn_output = _flash_attention_forward(
            query,
            key,
            value,
            attention_mask,
            query_length,
            dropout=attn_dropout,
            softmax_scale=1.0,
            is_causal=self.is_causal,
            use_top_left_mask=self._flash_attn_uses_top_left_mask,
        )

        attn_weights_reshaped = attn_output.reshape(bsz, query_length, self.num_heads * self.head_dim)
        attn_output = self.out_proj(attn_weights_reshaped)
        attn_output = self.resid_dropout(attn_output)

        return attn_output, attn_weights_reshaped


GPT_NEO_ATTENTION_CLASSES = {
    "eager": GPTNeoSelfAttention,
    "flash_attention_2": GPTNeoFlashAttention2,
}


class GPTNeoAttention(nn.Module):
    def __init__(self, config, layer_id=0):
        super().__init__()
        self.layer_id = layer_id
        self.attention_layers = config.attention_layers
        self.attention_type = self.attention_layers[layer_id]

        if self.attention_type in ["global", "local"]:
            self.attention = GPT_NEO_ATTENTION_CLASSES[config._attn_implementation](
                config, self.attention_type, layer_id
            )
        else:
            raise NotImplementedError(
                "Only attn layer types 'global' and 'local' exist, but got `config.attention_layers`: "
                f"{config.attention_layers}. Select attn layer types from ['global', 'local'] only."
            )

    def forward(
        self,
        hidden_states,
        layer_past=None,
        attention_mask=None,
        use_cache=False,
        output_attentions=False,
        cache_position=None,
    ):
        return self.attention(
            hidden_states,
            attention_mask=attention_mask,
            layer_past=layer_past,
            use_cache=use_cache,
            output_attentions=output_attentions,
            cache_position=cache_position,
        )


class GPTNeoMLP(nn.Module):
    def __init__(self, intermediate_size, config):  # in MLP: intermediate_size= 4 * hidden_size
        super().__init__()
        embed_dim = config.hidden_size
        self.c_fc = nn.Linear(embed_dim, intermediate_size)
        self.c_proj = nn.Linear(intermediate_size, embed_dim)
        self.act = ACT2FN[config.activation_function]
        self.dropout = nn.Dropout(float(config.resid_dropout))

    def forward(self, hidden_states):
        hidden_states = self.c_fc(hidden_states)
        hidden_states = self.act(hidden_states)
        hidden_states = self.c_proj(hidden_states)
        hidden_states = self.dropout(hidden_states)
        return hidden_states


class GPTNeoBlock(GradientCheckpointingLayer):
    def __init__(self, config, layer_id=None):
        super().__init__()
        hidden_size = config.hidden_size
        inner_dim = config.intermediate_size if config.intermediate_size is not None else 4 * hidden_size
        self.ln_1 = nn.LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
        self.attn = GPTNeoAttention(config, layer_id)
        self.ln_2 = nn.LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
        self.mlp = GPTNeoMLP(inner_dim, config)

    def forward(
        self,
        hidden_states,
        layer_past=None,
        attention_mask=None,
        use_cache=False,
        output_attentions=False,
        cache_position=None,
    ):
        residual = hidden_states
        hidden_states = self.ln_1(hidden_states)
        attn_output, attn_weights = self.attn(
            hidden_states,
            layer_past=layer_past,
            attention_mask=attention_mask,
            use_cache=use_cache,
            output_attentions=output_attentions,
            cache_position=cache_position,
        )

        # residual connection
        hidden_states = attn_output + residual

        residual = hidden_states
        hidden_states = self.ln_2(hidden_states)
        feed_forward_hidden_states = self.mlp(hidden_states)
        # residual connection
        hidden_states = residual + feed_forward_hidden_states

        return hidden_states, attn_weights


@auto_docstring
class GPTNeoPreTrainedModel(PreTrainedModel):
    config: GPTNeoConfig
    base_model_prefix = "transformer"
    supports_gradient_checkpointing = True
    _no_split_modules = ["GPTNeoBlock"]
    _skip_keys_device_placement = "past_key_values"
    _supports_flash_attn = True
    _can_compile_fullgraph = False  # TODO: needs a hybrid cache

    def _init_weights(self, module):
        super()._init_weights(module)
        if isinstance(module, GPTNeoSelfAttention):
            max_positions = module.config.max_position_embeddings
            bias = torch.tril(torch.ones((max_positions, max_positions), dtype=bool)).view(
                1, 1, max_positions, max_positions
            )
            if module.attention_type == "local":
                bias = torch.bitwise_xor(bias, torch.tril(bias, -module.config.window_size))
            init.copy_(module.bias, bias)


@auto_docstring
class GPTNeoModel(GPTNeoPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)

        self.embed_dim = config.hidden_size
        self.wte = nn.Embedding(config.vocab_size, self.embed_dim)
        self.wpe = nn.Embedding(config.max_position_embeddings, self.embed_dim)
        self.drop = nn.Dropout(float(config.embed_dropout))
        self.h = nn.ModuleList([GPTNeoBlock(config, layer_id=i) for i in range(config.num_layers)])
        self.ln_f = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_epsilon)

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

    def get_input_embeddings(self):
        return self.wte

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

    @auto_docstring
    def forward(
        self,
        input_ids: torch.Tensor | None = None,
        past_key_values: Cache | None = None,
        attention_mask: torch.Tensor | None = None,
        token_type_ids: torch.Tensor | None = None,
        position_ids: torch.Tensor | None = None,
        inputs_embeds: torch.Tensor | None = None,
        use_cache: bool | None = None,
        output_attentions: bool | None = None,
        output_hidden_states: bool | None = None,
        return_dict: bool | None = None,
        cache_position: torch.LongTensor | None = None,
        **kwargs,
    ) -> tuple[torch.Tensor] | BaseModelOutputWithPastAndCrossAttentions:
        r"""
        input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`):
            `input_ids_length` = `sequence_length` if `past_key_values` is `None` else
            `past_key_values.get_seq_length()` (`sequence_length` of input past key value states). Indices of input
            sequence tokens in the vocabulary.

            If `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 [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
            [`PreTrainedTokenizer.__call__`] for details.

            [What are input IDs?](../glossary#input-ids)
        """
        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
        )
        use_cache = use_cache if use_cache is not None else self.config.use_cache
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        if (input_ids is None) ^ (inputs_embeds is not None):
            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")

        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 inputs_embeds is None:
            inputs_embeds = self.wte(input_ids)

        if use_cache and past_key_values is None:
            past_key_values = DynamicCache(config=self.config)

        seq_length = inputs_embeds.shape[1]
        if cache_position is None:
            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
            cache_position = torch.arange(past_seen_tokens, past_seen_tokens + seq_length, device=inputs_embeds.device)

        if position_ids is None:
            position_ids = cache_position.unsqueeze(0)

        causal_mask = self._update_causal_mask(
            attention_mask, inputs_embeds, cache_position, past_key_values, output_attentions
        )

        position_embeds = self.wpe(position_ids)
        hidden_states = inputs_embeds + position_embeds

        if token_type_ids is not None:
            token_type_ids = token_type_ids.view(-1, seq_length)
            token_type_embeds = self.wte(token_type_ids)
            hidden_states = hidden_states + token_type_embeds

        hidden_states = self.drop(hidden_states)
        output_shape = (-1, seq_length, hidden_states.size(-1))

        all_self_attentions = () if output_attentions else None
        all_hidden_states = () if output_hidden_states else None
        for i, block in enumerate(self.h):
            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            outputs = block(
                hidden_states,
                layer_past=past_key_values,
                attention_mask=causal_mask,
                use_cache=use_cache,
                output_attentions=output_attentions,
                cache_position=cache_position,
            )

            hidden_states = outputs[0]
            if output_attentions:
                all_self_attentions = all_self_attentions + (outputs[1],)

        hidden_states = self.ln_f(hidden_states)

        hidden_states = hidden_states.view(output_shape)
        # Add last hidden state
        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] if v is not None
            )

        return BaseModelOutputWithPast(
            last_hidden_state=hidden_states,
            past_key_values=past_key_values,
            hidden_states=all_hidden_states,
            attentions=all_self_attentions,
        )

    # Copied from transformers.models.gptj.modeling_gptj.GPTJModel._update_causal_mask
    def _update_causal_mask(
        self,
        attention_mask: Union[torch.Tensor, "BlockMask"],
        input_tensor: torch.Tensor,
        cache_position: torch.Tensor,
        past_key_values: Cache,
        output_attentions: bool = False,
    ):
        if is_flash_attention_requested(self.config):
            if attention_mask is not None and (attention_mask == 0.0).any():
                return attention_mask
            return None
        if self.config._attn_implementation == "flex_attention":
            if isinstance(attention_mask, torch.Tensor):
                attention_mask = make_flex_block_causal_mask(attention_mask)
            return attention_mask

        # For SDPA, when possible, we will rely on its `is_causal` argument instead of its `attn_mask` argument, in
        # order to dispatch on Flash Attention 2. This feature is not compatible with static cache, as SDPA will fail
        # to infer the attention mask.
        past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
        using_compilable_cache = past_key_values.is_compileable if past_key_values is not None else False

        # When output attentions is True, sdpa implementation's forward method calls the eager implementation's forward
        if self.config._attn_implementation == "sdpa" and not using_compilable_cache and not output_attentions:
            if AttentionMaskConverter._ignore_causal_mask_sdpa(
                attention_mask,
                inputs_embeds=input_tensor,
                past_key_values_length=past_seen_tokens,
                is_training=self.training,
            ):
                return None

        dtype = input_tensor.dtype
        sequence_length = input_tensor.shape[1]
        if using_compilable_cache:
            target_length = past_key_values.get_max_cache_shape()
        else:
            target_length = (
                attention_mask.shape[-1]
                if isinstance(attention_mask, torch.Tensor)
                else past_seen_tokens + sequence_length + 1
            )

        # In case the provided `attention` mask is 2D, we generate a causal mask here (4D).
        causal_mask = self._prepare_4d_causal_attention_mask_with_cache_position(
            attention_mask,
            sequence_length=sequence_length,
            target_length=target_length,
            dtype=dtype,
            cache_position=cache_position,
            batch_size=input_tensor.shape[0],
        )

        if (
            self.config._attn_implementation == "sdpa"
            and attention_mask is not None
            and attention_mask.device.type in ["cuda", "xpu", "npu"]
            and not output_attentions
        ):
            # Attend to all tokens in fully masked rows in the causal_mask, for example the relevant first rows when
            # using left padding. This is required by F.scaled_dot_product_attention memory-efficient attention path.
            # Details: https://github.com/pytorch/pytorch/issues/110213
            min_dtype = torch.finfo(dtype).min
            causal_mask = AttentionMaskConverter._unmask_unattended(causal_mask, min_dtype)

        return causal_mask

    @staticmethod
    # Copied from transformers.models.gptj.modeling_gptj.GPTJModel._prepare_4d_causal_attention_mask_with_cache_position
    def _prepare_4d_causal_attention_mask_with_cache_position(
        attention_mask: torch.Tensor,
        sequence_length: int,
        target_length: int,
        dtype: torch.dtype,
        cache_position: torch.Tensor,
        batch_size: int,
        **kwargs,
    ):
        """
        Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape
        `(batch_size, key_value_length)`, or if the input `attention_mask` is already 4D, do nothing.

        Args:
            attention_mask (`torch.Tensor`):
                A 2D attention mask of shape `(batch_size, key_value_length)` or a 4D attention mask of shape
                `(batch_size, 1, query_length, key_value_length)`.
            sequence_length (`int`):
                The sequence length being processed.
            target_length (`int`):
                The target length: when generating with static cache, the mask should be as long as the static cache,
                to account for the 0 padding, the part of the cache that is not filled yet.
            dtype (`torch.dtype`):
                The dtype to use for the 4D attention mask.
            cache_position (`torch.Tensor`):
                Indices depicting the position of the input sequence tokens in the sequence.
            batch_size (`torch.Tensor`):
                Batch size.
        """
        if attention_mask is not None and attention_mask.dim() == 4:
            # In this case we assume that the mask comes already in inverted form and requires no inversion or slicing.
            causal_mask = attention_mask
        else:
            min_dtype = torch.finfo(dtype).min
            causal_mask = torch.full(
                (sequence_length, target_length), fill_value=min_dtype, dtype=dtype, device=cache_position.device
            )
            if sequence_length != 1:
                causal_mask = torch.triu(causal_mask, diagonal=1)
            causal_mask *= torch.arange(target_length, device=cache_position.device) > cache_position.reshape(-1, 1)
            causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
            if attention_mask is not None:
                causal_mask = causal_mask.clone()  # copy to contiguous memory for in-place edit
                mask_length = attention_mask.shape[-1]
                padding_mask = causal_mask[:, :, :, :mask_length] + attention_mask[:, None, None, :].to(
                    causal_mask.device
                )
                padding_mask = padding_mask == 0
                causal_mask[:, :, :, :mask_length] = causal_mask[:, :, :, :mask_length].masked_fill(
                    padding_mask, min_dtype
                )

        return causal_mask


@auto_docstring(
    custom_intro="""
    The GPT Neo Model transformer with a language modeling head on top (linear layer with weights tied to the input
    embeddings).
    """
)
class GPTNeoForCausalLM(GPTNeoPreTrainedModel, GenerationMixin):
    _tied_weights_keys = {"lm_head.weight": "transformer.wte.weight"}

    def __init__(self, config):
        super().__init__(config)
        self.transformer = GPTNeoModel(config)
        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)

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

    @auto_docstring
    def forward(
        self,
        input_ids: torch.Tensor | None = None,
        past_key_values: Cache | None = None,
        attention_mask: torch.Tensor | None = None,
        token_type_ids: torch.Tensor | None = None,
        position_ids: torch.Tensor | None = None,
        inputs_embeds: torch.Tensor | None = None,
        labels: torch.Tensor | None = None,
        use_cache: bool | None = None,
        output_attentions: bool | None = None,
        output_hidden_states: bool | None = None,
        return_dict: bool | None = None,
        cache_position: torch.LongTensor | None = None,
        logits_to_keep: int | torch.Tensor = 0,
        **kwargs,
    ) -> tuple[torch.Tensor] | CausalLMOutputWithCrossAttentions:
        r"""
        input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`):
            `input_ids_length` = `sequence_length` if `past_key_values` is `None` else
            `past_key_values.get_seq_length()` (`sequence_length` of input past key value states). Indices of input
            sequence tokens in the vocabulary.

            If `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 [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
            [`PreTrainedTokenizer.__call__`] for details.

            [What are input IDs?](../glossary#input-ids)
        labels (`torch.LongTensor` of shape `(batch_size, input_ids_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,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            cache_position=cache_position,
        )

        hidden_states = transformer_outputs[0]
        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
        logits = self.lm_head(hidden_states[:, slice_indices, :])

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

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

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


@auto_docstring(
    custom_intro="""
    The GPTNeo Model transformer with a sequence classification head on top (linear layer).

    [`GPTNeoForSequenceClassification`] uses the last token in order to do the classification, as other causal models
    (e.g. GPT-1) do.

    Since it does classification on the last token, it requires to know the position of the last token. If a
    `pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If
    no `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 `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in
    each row of the batch).
    """
)
class GPTNeoForSequenceClassification(GPTNeoPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels
        self.transformer = GPTNeoModel(config)
        self.score = nn.Linear(config.hidden_size, self.num_labels, bias=False)

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

    @auto_docstring
    def forward(
        self,
        input_ids: torch.Tensor | None = None,
        past_key_values: Cache | None = None,
        attention_mask: torch.Tensor | None = None,
        token_type_ids: torch.Tensor | None = None,
        position_ids: torch.Tensor | None = None,
        inputs_embeds: torch.Tensor | None = None,
        labels: torch.Tensor | None = None,
        use_cache: bool | None = None,
        output_attentions: bool | None = None,
        output_hidden_states: bool | None = None,
        return_dict: bool | None = None,
        **kwargs,
    ) -> tuple[torch.Tensor] | SequenceClassifierOutputWithPast:
        r"""
        input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`):
            `input_ids_length` = `sequence_length` if `past_key_values` is `None` else
            `past_key_values.get_seq_length()` (`sequence_length` of input past key value states). Indices of input
            sequence tokens in the vocabulary.

            If `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 [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
            [`PreTrainedTokenizer.__call__`] for details.

            [What are input IDs?](../glossary#input-ids)
        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

        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,
            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.score(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]

        if self.config.pad_token_id is None and batch_size != 1:
            raise ValueError("Cannot handle batch sizes > 1 if no padding token is defined.")
        if self.config.pad_token_id is None:
            last_non_pad_token = -1
        elif input_ids is not None:
            # To handle both left- and right- padding, we take the rightmost token that is not equal to pad_token_id
            non_pad_mask = (input_ids != self.config.pad_token_id).to(logits.device, torch.int32)
            token_indices = torch.arange(input_ids.shape[-1], device=logits.device, dtype=torch.int32)
            last_non_pad_token = (token_indices * non_pad_mask).argmax(-1)
        else:
            last_non_pad_token = -1
            logger.warning_once(
                f"{self.__class__.__name__} will not detect padding tokens in `inputs_embeds`. Results may be "
                "unexpected if using padding tokens in conjunction with `inputs_embeds.`"
            )

        pooled_logits = logits[torch.arange(batch_size, device=logits.device), last_non_pad_token]

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

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


@auto_docstring
class GPTNeoForTokenClassification(GPTNeoPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels

        self.transformer = GPTNeoModel(config)
        self.dropout = nn.Dropout(config.classifier_dropout)
        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: torch.LongTensor | None = None,
        past_key_values: Cache | None = None,
        attention_mask: torch.FloatTensor | None = None,
        token_type_ids: torch.LongTensor | None = None,
        position_ids: torch.LongTensor | None = None,
        inputs_embeds: torch.FloatTensor | None = None,
        labels: torch.LongTensor | None = None,
        use_cache: bool | None = None,
        output_attentions: bool | None = None,
        output_hidden_states: bool | None = None,
        return_dict: bool | None = None,
        **kwargs,
    ) -> tuple | TokenClassifierOutput:
        r"""
        input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`):
            `input_ids_length` = `sequence_length` if `past_key_values` is `None` else
            `past_key_values.get_seq_length()` (`sequence_length` of input past key value states). Indices of input
            sequence tokens in the vocabulary.

            If `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 [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
            [`PreTrainedTokenizer.__call__`] for details.

            [What are input IDs?](../glossary#input-ids)
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *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

        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,
            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]
        hidden_states = self.dropout(hidden_states)
        logits = self.classifier(hidden_states)

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

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

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


@auto_docstring
class GPTNeoForQuestionAnswering(GPTNeoPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels
        self.transformer = GPTNeoModel(config)
        self.qa_outputs = nn.Linear(config.hidden_size, 2)

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

    @auto_docstring
    def forward(
        self,
        input_ids: torch.LongTensor | None = None,
        attention_mask: torch.FloatTensor | None = None,
        token_type_ids: torch.LongTensor | None = None,
        position_ids: torch.LongTensor | None = None,
        inputs_embeds: torch.FloatTensor | None = None,
        start_positions: torch.LongTensor | None = None,
        end_positions: torch.LongTensor | None = None,
        output_attentions: bool | None = None,
        output_hidden_states: bool | None = None,
        return_dict: bool | None = None,
        **kwargs,
    ) -> tuple | QuestionAnsweringModelOutput:
        r"""
        input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`):
            `input_ids_length` = `sequence_length` if `past_key_values` is `None` else
            `past_key_values.get_seq_length()` (`sequence_length` of input past key value states). Indices of input
            sequence tokens in the vocabulary.

            If `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 [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
            [`PreTrainedTokenizer.__call__`] for details.

            [What are input IDs?](../glossary#input-ids)
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.transformer(
            input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            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).contiguous()
        end_logits = end_logits.squeeze(-1).contiguous()

        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[2:]
            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__ = [
    "GPTNeoForCausalLM",
    "GPTNeoForQuestionAnswering",
    "GPTNeoForSequenceClassification",
    "GPTNeoForTokenClassification",
    "GPTNeoModel",
    "GPTNeoPreTrainedModel",
]