modeling_iquestcoder.py

"""
Modified MIT License 

Software Copyright© 2025 IQuest Research

Our only modification is that, if the Software (or any derivative works
thereof) is used for any of your commercial products or services, you shall
prominently display "IQuest Coder" on the user interface of such product or
service.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
"""

from typing import Callable, List, Optional, Tuple, Union

import torch
import torch.nn as nn
import torch.nn.functional as F

from transformers.activations import ACT2FN
from transformers.cache_utils import Cache, DynamicCache, SlidingWindowCache, StaticCache
from transformers.generation import GenerationMixin
from transformers.modeling_attn_mask_utils import AttentionMaskConverter
from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
from transformers.modeling_layers import GradientCheckpointingLayer
from transformers.modeling_outputs import (
    BaseModelOutputWithPast,
    CausalLMOutputWithPast,
    QuestionAnsweringModelOutput,
    SequenceClassifierOutputWithPast,
    TokenClassifierOutput,
)
from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
from transformers.processing_utils import Unpack
from transformers.utils import (
    LossKwargs,
    auto_docstring,
    can_return_tuple,
    is_torch_flex_attn_available,
    logging,
)

from .configuration_iquestcoder import IQuestCoderConfig


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


logger = logging.get_logger(__name__)


# =============================================================================
# Helper Functions
# =============================================================================

def rotate_half(x: torch.Tensor) -> torch.Tensor:
    """Rotates half the hidden dims of the input."""
    x1 = x[..., : x.shape[-1] // 2]
    x2 = x[..., x.shape[-1] // 2 :]
    return torch.cat((-x2, x1), dim=-1)


def apply_rotary_pos_emb(
    q: torch.Tensor,
    k: torch.Tensor,
    cos: torch.Tensor,
    sin: torch.Tensor,
    position_ids: Optional[torch.Tensor] = None,
    unsqueeze_dim: int = 1,
) -> Tuple[torch.Tensor, torch.Tensor]:
    """Applies Rotary Position Embedding to the query and key tensors.

    Args:
        q: The query tensor.
        k: The key tensor.
        cos: The cosine part of the rotary embedding.
        sin: The sine part of the rotary embedding.
        position_ids: Deprecated and unused.
        unsqueeze_dim: The dimension along which to unsqueeze cos and sin.

    Returns:
        Tuple of query and key tensors rotated using the Rotary Position Embedding.
    """
    # Borrowed from OLMo: preserve original dtypes for numerical stability
    q_dtype, k_dtype = q.dtype, k.dtype
    cos = cos.unsqueeze(unsqueeze_dim)
    sin = sin.unsqueeze(unsqueeze_dim)
    q_embed = (q * cos) + (rotate_half(q) * sin)
    k_embed = (k * cos) + (rotate_half(k) * sin)
    return q_embed.to(q_dtype), k_embed.to(k_dtype)


def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
    """
    Expands key/value heads for Grouped Query Attention.
    
    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep).
    The hidden states go from (batch, num_key_value_heads, seqlen, head_dim) to
    (batch, num_attention_heads, seqlen, head_dim).
    """
    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
    if n_rep == 1:
        return hidden_states
    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)


def eager_attention_forward(
    module: nn.Module,
    query: torch.Tensor,
    key: torch.Tensor,
    value: torch.Tensor,
    attention_mask: Optional[torch.Tensor],
    scaling: float,
    dropout: float = 0.0,
    **kwargs,
) -> Tuple[torch.Tensor, torch.Tensor]:
    """Standard eager attention implementation."""
    key_states = repeat_kv(key, module.num_key_value_groups)
    value_states = repeat_kv(value, module.num_key_value_groups)

    attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
    if attention_mask is not None:
        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
        attn_weights = attn_weights + causal_mask

    attn_weights = F.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
    attn_weights = F.dropout(attn_weights, p=dropout, training=module.training)
    attn_output = torch.matmul(attn_weights, value_states)
    attn_output = attn_output.transpose(1, 2).contiguous()

    return attn_output, attn_weights


# =============================================================================
# Model Components
# =============================================================================

class IQuestCoderRMSNorm(nn.Module):
    """Root Mean Square Layer Normalization.
    
    RMSNorm is computationally simpler than LayerNorm while achieving similar
    performance. It normalizes the input by its RMS value.
    """

    def __init__(self, hidden_size: int, eps: float = 1e-6):
        super().__init__()
        self.weight = nn.Parameter(torch.ones(hidden_size))
        self.variance_epsilon = eps

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        input_dtype = hidden_states.dtype
        hidden_states = hidden_states.to(torch.float32)
        variance = hidden_states.pow(2).mean(-1, keepdim=True)
        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
        return self.weight * hidden_states.to(input_dtype)

    def extra_repr(self) -> str:
        return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"


class IQuestCoderRotaryEmbedding(nn.Module):
    """Rotary Position Embedding (RoPE).
    
    Implements rotary positional embeddings as described in the RoFormer paper.
    Supports various RoPE scaling methods for extended context lengths.
    """

    def __init__(self, config: IQuestCoderConfig, device=None):
        super().__init__()
        # BC: "rope_type" was originally "type"
        if hasattr(config, "rope_scaling") and config.rope_scaling is not None:
            self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type"))
        else:
            self.rope_type = "default"
        self.max_seq_len_cached = config.max_position_embeddings
        self.original_max_seq_len = config.max_position_embeddings

        self.config = config
        self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]

        inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device)
        self.register_buffer("inv_freq", inv_freq, persistent=False)
        self.original_inv_freq = self.inv_freq

    @torch.no_grad()
    @dynamic_rope_update
    def forward(self, x: torch.Tensor, position_ids: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to(x.device)
        position_ids_expanded = position_ids[:, None, :].float()

        device_type = x.device.type if isinstance(x.device.type, str) and x.device.type != "mps" else "cpu"
        with torch.autocast(device_type=device_type, enabled=False):
            freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
            emb = torch.cat((freqs, freqs), dim=-1)
            cos = emb.cos() * self.attention_scaling
            sin = emb.sin() * self.attention_scaling

        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)


class IQuestCoderMLP(nn.Module):
    """Feed-forward network with SwiGLU activation.
    
    Uses the gated linear unit variant with SiLU activation for improved
    performance compared to standard FFN.
    """

    def __init__(self, config: IQuestCoderConfig):
        super().__init__()
        self.config = config
        self.hidden_size = config.hidden_size
        self.intermediate_size = config.intermediate_size
        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=config.mlp_bias)
        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=config.mlp_bias)
        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=config.mlp_bias)
        self.act_fn = ACT2FN[config.hidden_act]

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        # SwiGLU: down_proj(act_fn(gate_proj(x)) * up_proj(x))
        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))


class IQuestCoderAttention(nn.Module):
    """Multi-headed attention with support for Grouped Query Attention (GQA).
    
    Features:
    - Grouped Query Attention for memory efficiency
    - Optional QKV clipping for training stability (from OLMo)
    - Optional sliding window attention (from Qwen2)
    - Rotary Position Embeddings
    """

    def __init__(self, config: IQuestCoderConfig, layer_idx: int):
        super().__init__()
        self.config = config
        self.layer_idx = layer_idx
        self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
        self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
        self.scaling = self.head_dim ** -0.5
        self.attention_dropout = config.attention_dropout
        self.is_causal = True

        # Projection layers
        self.q_proj = nn.Linear(
            config.hidden_size, config.num_attention_heads * self.head_dim, bias=config.attention_bias
        )
        self.k_proj = nn.Linear(
            config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
        )
        self.v_proj = nn.Linear(
            config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
        )
        self.o_proj = nn.Linear(
            config.num_attention_heads * self.head_dim, config.hidden_size, bias=config.attention_bias
        )

    def forward(
        self,
        hidden_states: torch.Tensor,
        position_embeddings: Tuple[torch.Tensor, torch.Tensor],
        attention_mask: Optional[torch.Tensor],
        past_key_value: Optional[Cache] = None,
        cache_position: Optional[torch.LongTensor] = None,
        **kwargs: Unpack[FlashAttentionKwargs],
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
        input_shape = hidden_states.shape[:-1]
        hidden_shape = (*input_shape, -1, self.head_dim)

        # Compute Q, K, V projections
        query_states = self.q_proj(hidden_states)
        key_states = self.k_proj(hidden_states)
        value_states = self.v_proj(hidden_states)

        # [OLMo Feature] Optional QKV clipping for training stability
        if self.config.clip_qkv is not None:
            query_states = query_states.clamp(min=-self.config.clip_qkv, max=self.config.clip_qkv)
            key_states = key_states.clamp(min=-self.config.clip_qkv, max=self.config.clip_qkv)
            value_states = value_states.clamp(min=-self.config.clip_qkv, max=self.config.clip_qkv)

        # Reshape to (batch, heads, seq_len, head_dim)
        query_states = query_states.view(hidden_shape).transpose(1, 2)
        key_states = key_states.view(hidden_shape).transpose(1, 2)
        value_states = value_states.view(hidden_shape).transpose(1, 2)

        # Apply rotary position embeddings
        cos, sin = position_embeddings
        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)

        # Update KV cache if provided
        if past_key_value is not None:
            cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)

        # [Qwen2 Feature] Sliding window attention
        sliding_window = None
        if (
            self.config.use_sliding_window
            and getattr(self.config, "sliding_window", None) is not None
            and self.layer_idx >= self.config.max_window_layers
        ):
            sliding_window = self.config.sliding_window

        # Select attention implementation
        attention_interface: Callable = eager_attention_forward
        if self.config._attn_implementation != "eager":
            if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
                logger.warning_once(
                    "`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. "
                    'Falling back to eager attention. This warning can be removed using the argument '
                    '`attn_implementation="eager"` when loading the model.'
                )
            else:
                attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]

        # Compute attention
        attn_output, attn_weights = attention_interface(
            self,
            query_states,
            key_states,
            value_states,
            attention_mask,
            dropout=0.0 if not self.training else self.attention_dropout,
            scaling=self.scaling,
            sliding_window=sliding_window,
            **kwargs,
        )

        # Reshape and project output
        attn_output = attn_output.reshape(*input_shape, -1).contiguous()
        attn_output = self.o_proj(attn_output)

        return attn_output, attn_weights


class IQuestCoderDecoderLayer(GradientCheckpointingLayer):
    """Transformer decoder layer with pre-normalization.
    
    Architecture: Pre-RMSNorm -> Attention -> Residual -> Pre-RMSNorm -> MLP -> Residual
    """

    def __init__(self, config: IQuestCoderConfig, layer_idx: int):
        super().__init__()
        self.hidden_size = config.hidden_size
        self.self_attn = IQuestCoderAttention(config=config, layer_idx=layer_idx)
        self.mlp = IQuestCoderMLP(config)
        self.input_layernorm = IQuestCoderRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_attention_layernorm = IQuestCoderRMSNorm(config.hidden_size, eps=config.rms_norm_eps)

        # Warn if sliding window is enabled but not properly supported
        if config.use_sliding_window and config._attn_implementation != "flash_attention_2":
            logger.warning_once(
                f"Sliding Window Attention is enabled but not implemented for `{config._attn_implementation}`; "
                "unexpected results may be encountered."
            )

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_value: Optional[Cache] = None,
        output_attentions: Optional[bool] = False,
        use_cache: Optional[bool] = False,
        cache_position: Optional[torch.LongTensor] = None,
        position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
        **kwargs: Unpack[FlashAttentionKwargs],
    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
        # Pre-norm + Self Attention
        residual = hidden_states
        hidden_states = self.input_layernorm(hidden_states)

        hidden_states, self_attn_weights = self.self_attn(
            hidden_states=hidden_states,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_value=past_key_value,
            output_attentions=output_attentions,
            use_cache=use_cache,
            cache_position=cache_position,
            position_embeddings=position_embeddings,
            **kwargs,
        )
        hidden_states = residual + hidden_states

        # Pre-norm + MLP
        residual = hidden_states
        hidden_states = self.post_attention_layernorm(hidden_states)
        hidden_states = self.mlp(hidden_states)
        hidden_states = residual + hidden_states

        outputs = (hidden_states,)
        if output_attentions:
            outputs += (self_attn_weights,)

        return outputs


# =============================================================================
# Base Model
# =============================================================================

@auto_docstring
class IQuestCoderPreTrainedModel(PreTrainedModel):
    """Base class for IQuestCoder models."""

    config_class = IQuestCoderConfig
    base_model_prefix = "model"
    supports_gradient_checkpointing = True
    _no_split_modules = ["IQuestCoderDecoderLayer"]
    _skip_keys_device_placement = ["past_key_values"]
    _supports_flash_attn_2 = True
    _supports_sdpa = True
    _supports_flex_attn = True
    _supports_cache_class = True
    _supports_quantized_cache = True
    _supports_static_cache = True
    _supports_attention_backend = True

    def _init_weights(self, module: nn.Module):
        std = self.config.initializer_range
        if isinstance(module, nn.Linear):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.Embedding):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.padding_idx is not None:
                module.weight.data[module.padding_idx].zero_()
        elif isinstance(module, IQuestCoderRMSNorm):
            module.weight.data.fill_(1.0)


@auto_docstring
class IQuestCoderModel(IQuestCoderPreTrainedModel):
    """
    IQuestCoder Model outputting raw hidden-states without any specific head on top.
    
    This model is compatible with LLaMA weights while incorporating features from OLMo and Qwen2.
    """

    def __init__(self, config: IQuestCoderConfig):
        super().__init__(config)
        self.padding_idx = config.pad_token_id
        self.vocab_size = config.vocab_size

        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
        self.layers = nn.ModuleList(
            [IQuestCoderDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
        )
        self.norm = IQuestCoderRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.rotary_emb = IQuestCoderRotaryEmbedding(config=config)
        self.gradient_checkpointing = False

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

    def get_input_embeddings(self) -> nn.Embedding:
        return self.embed_tokens

    def set_input_embeddings(self, value: nn.Embedding):
        self.embed_tokens = value

    @can_return_tuple
    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Cache] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        cache_position: Optional[torch.LongTensor] = None,
        **flash_attn_kwargs: Unpack[FlashAttentionKwargs],
    ) -> BaseModelOutputWithPast:
        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

        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 and use_cache:
            logger.warning_once(
                "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`."
            )
            use_cache = False

        if not isinstance(past_key_values, (type(None), Cache)):
            raise ValueError("The `past_key_values` should be either a `Cache` object or `None`.")

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

        if use_cache and past_key_values is None:
            past_key_values = DynamicCache()

        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 + inputs_embeds.shape[1], 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
        )

        hidden_states = inputs_embeds

        # Create position embeddings to be shared across the decoder layers
        position_embeddings = self.rotary_emb(hidden_states, position_ids)

        # Decoder layers
        all_hidden_states = () if output_hidden_states else None
        all_self_attns = () if output_attentions else None

        for decoder_layer in self.layers[: self.config.num_hidden_layers]:
            if output_hidden_states:
                all_hidden_states += (hidden_states,)

            layer_outputs = decoder_layer(
                hidden_states,
                attention_mask=causal_mask,
                position_ids=position_ids,
                past_key_value=past_key_values,
                output_attentions=output_attentions,
                use_cache=use_cache,
                cache_position=cache_position,
                position_embeddings=position_embeddings,
                **flash_attn_kwargs,
            )

            hidden_states = layer_outputs[0]

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

        hidden_states = self.norm(hidden_states)

        # Add hidden states from the last decoder layer
        if output_hidden_states:
            all_hidden_states += (hidden_states,)

        return BaseModelOutputWithPast(
            last_hidden_state=hidden_states,
            past_key_values=past_key_values if use_cache else None,
            hidden_states=all_hidden_states,
            attentions=all_self_attns,
        )

    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 self.config._attn_implementation == "flash_attention_2":
            if attention_mask is not None and past_key_values is not None:
                is_padding_right = attention_mask[:, -1].sum().item() != input_tensor.size()[0]
                if is_padding_right:
                    raise ValueError(
                        "You are attempting to perform batched generation with padding_side='right'. "
                        "This may lead to unexpected behaviour for Flash Attention version of IQuestCoder. "
                        "Make sure to call `tokenizer.padding_side = 'left'` before tokenizing the input."
                    )
            if attention_mask is not None and 0.0 in attention_mask:
                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

        past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
        using_static_cache = isinstance(past_key_values, StaticCache)
        using_sliding_window_cache = isinstance(past_key_values, SlidingWindowCache)

        if (
            self.config._attn_implementation == "sdpa"
            and not (using_static_cache or using_sliding_window_cache)
            and not output_attentions
        ):
            if AttentionMaskConverter._ignore_causal_mask_sdpa(
                attention_mask,
                inputs_embeds=input_tensor,
                past_key_values_length=past_seen_tokens,
                sliding_window=self.config.sliding_window if self.config.use_sliding_window else None,
                is_training=self.training,
            ):
                return None

        dtype = input_tensor.dtype
        min_dtype = torch.finfo(dtype).min
        sequence_length = input_tensor.shape[1]

        if using_sliding_window_cache or using_static_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
            )

        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],
            config=self.config,
            past_key_values=past_key_values,
        )

        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
        ):
            causal_mask = AttentionMaskConverter._unmask_unattended(causal_mask, min_dtype)

        return causal_mask

    @staticmethod
    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,
        config: IQuestCoderConfig,
        past_key_values: Cache,
    ):
        """Creates a causal 4D mask from a 2D mask, or returns the 4D mask if already provided."""
        if attention_mask is not None and attention_mask.dim() == 4:
            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
            )
            diagonal_attend_mask = torch.arange(target_length, device=cache_position.device) > cache_position.reshape(
                -1, 1
            )
            
            # [Qwen2 Feature] Handle sliding window mask
            if getattr(config, "use_sliding_window", False) and config.sliding_window is not None:
                if not isinstance(past_key_values, SlidingWindowCache) or sequence_length > target_length:
                    sliding_attend_mask = torch.arange(target_length, device=cache_position.device) <= (
                        cache_position.reshape(-1, 1) - config.sliding_window
                    )
                    diagonal_attend_mask.bitwise_or_(sliding_attend_mask)
            
            causal_mask *= diagonal_attend_mask
            causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
            
            if attention_mask is not None:
                causal_mask = causal_mask.clone()
                if attention_mask.shape[-1] > target_length:
                    attention_mask = attention_mask[:, :target_length]
                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


# =============================================================================
# Model Heads
# =============================================================================

class KwargsForCausalLM(FlashAttentionKwargs, LossKwargs):
    ...


@auto_docstring
class IQuestCoderForCausalLM(IQuestCoderPreTrainedModel, GenerationMixin):
    """IQuestCoder Model with a language modeling head on top for causal LM."""

    _tied_weights_keys = ["lm_head.weight"]
    _tp_plan = {"lm_head": "colwise_rep"}
    _pp_plan = {"lm_head": (["hidden_states"], ["logits"])}

    def __init__(self, config: IQuestCoderConfig):
        super().__init__(config)
        self.model = IQuestCoderModel(config)
        self.vocab_size = config.vocab_size
        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)

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

    def get_input_embeddings(self) -> nn.Embedding:
        return self.model.embed_tokens

    def set_input_embeddings(self, value: nn.Embedding):
        self.model.embed_tokens = value

    def get_output_embeddings(self) -> nn.Linear:
        return self.lm_head

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

    def set_decoder(self, decoder: IQuestCoderModel):
        self.model = decoder

    def get_decoder(self) -> IQuestCoderModel:
        return self.model

    @can_return_tuple
    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Cache] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        cache_position: Optional[torch.LongTensor] = None,
        logits_to_keep: Union[int, torch.Tensor] = 0,
        **kwargs: Unpack[KwargsForCausalLM],
    ) -> CausalLMOutputWithPast:
        r"""
        Args:
            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
                config.vocab_size]` or -100 (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]`.

        Example:
            ```python
            >>> from transformers import AutoTokenizer
            >>> from modeling_iquestcoder import IQuestCoderForCausalLM

            >>> model = IQuestCoderForCausalLM.from_pretrained("path/to/IQuestCoder")
            >>> tokenizer = AutoTokenizer.from_pretrained("path/to/IQuestCoder")

            >>> prompt = "Hey, are you conscious? Can you talk to me?"
            >>> inputs = tokenizer(prompt, return_tensors="pt")

            >>> # Generate
            >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
            >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
            "Hey, are you conscious? Can you talk to me?\\nI'm not conscious, but I can talk to you."
            ```
        """
        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
        )

        # Decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
        outputs: BaseModelOutputWithPast = self.model(
            input_ids=input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            cache_position=cache_position,
            **kwargs,
        )

        hidden_states = outputs.last_hidden_state
        # 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)

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


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

    [`IQuestCoderForSequenceClassification`] uses the last token in order to do the classification, as other causal
    models (e.g. GPT-2) do.

    Since it does classification on the last token, it requires to know the position of the last token. If a
    `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.
    """
)
class IQuestCoderForSequenceClassification(IQuestCoderPreTrainedModel):
    """IQuestCoder Model with a sequence classification head."""

    def __init__(self, config: IQuestCoderConfig):
        super().__init__(config)
        self.num_labels = config.num_labels
        self.model = IQuestCoderModel(config)
        self.score = nn.Linear(config.hidden_size, self.num_labels, bias=False)

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

    def get_input_embeddings(self) -> nn.Embedding:
        return self.model.embed_tokens

    def set_input_embeddings(self, value: nn.Embedding):
        self.model.embed_tokens = value

    @can_return_tuple
    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Cache] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
    ) -> SequenceClassifierOutputWithPast:
        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).
        """
        transformer_outputs: BaseModelOutputWithPast = self.model(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
        )
        hidden_states = transformer_outputs.last_hidden_state
        logits = self.score(hidden_states)

        if input_ids is not None:
            batch_size = input_ids.shape[0]
        else:
            batch_size = inputs_embeds.shape[0]

        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:
            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:
            loss = self.loss_function(logits=logits, labels=labels, pooled_logits=pooled_logits, config=self.config)

        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 IQuestCoderForTokenClassification(IQuestCoderPreTrainedModel):
    """IQuestCoder Model with a token classification head."""

    def __init__(self, config: IQuestCoderConfig):
        super().__init__(config)
        self.num_labels = config.num_labels
        self.model = IQuestCoderModel(config)
        if getattr(config, "classifier_dropout", None) is not None:
            classifier_dropout = config.classifier_dropout
        elif getattr(config, "hidden_dropout", None) is not None:
            classifier_dropout = config.hidden_dropout
        else:
            classifier_dropout = 0.1
        self.dropout = nn.Dropout(classifier_dropout)
        self.score = nn.Linear(config.hidden_size, config.num_labels)

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

    def get_input_embeddings(self) -> nn.Embedding:
        return self.model.embed_tokens

    def set_input_embeddings(self, value: nn.Embedding):
        self.model.embed_tokens = value

    @can_return_tuple
    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Cache] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
    ) -> TokenClassifierOutput:
        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).
        """
        outputs: BaseModelOutputWithPast = self.model(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
        )
        sequence_output = outputs.last_hidden_state
        sequence_output = self.dropout(sequence_output)
        logits = self.score(sequence_output)

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

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


@auto_docstring
class IQuestCoderForQuestionAnswering(IQuestCoderPreTrainedModel):
    """IQuestCoder Model with a span classification head for extractive question-answering."""

    base_model_prefix = "transformer"

    def __init__(self, config: IQuestCoderConfig):
        super().__init__(config)
        self.transformer = IQuestCoderModel(config)
        self.qa_outputs = nn.Linear(config.hidden_size, 2)

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

    def get_input_embeddings(self) -> nn.Embedding:
        return self.transformer.embed_tokens

    def set_input_embeddings(self, value: nn.Embedding):
        self.transformer.embed_tokens = value

    @can_return_tuple
    @auto_docstring
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Cache] = 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,
        **kwargs,
    ) -> QuestionAnsweringModelOutput:
        outputs: BaseModelOutputWithPast = self.transformer(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
        )

        sequence_output = outputs.last_hidden_state

        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()

        loss = None
        if start_positions is not None and end_positions is not None:
            loss = self.loss_function(start_logits, end_logits, start_positions, end_positions, **kwargs)

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


__all__ = [
    "IQuestCoderPreTrainedModel",
    "IQuestCoderModel",
    "IQuestCoderForCausalLM",
    "IQuestCoderForSequenceClassification",
    "IQuestCoderForTokenClassification",
    "IQuestCoderForQuestionAnswering",
]