# coding=utf-8
# This file was created for the HyperCLOVA X SEED 14B Think architecture.
# partially copied and modified from https://github.com/huggingface/transformers
# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
#
# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
# and OPT implementations in this library. It has been modified from its
# original forms to accommodate minor architectural differences compared
# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
#
# 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.
from typing import Callable, Optional, Union

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

from transformers.activations import ACT2FN
from transformers.cache_utils import Cache, DynamicCache
from transformers.generation import GenerationMixin
from transformers.modeling_attn_mask_utils import AttentionMaskConverter
from transformers.integrations import use_kernel_forward_from_hub
from transformers.generation.logits_process import LogitsProcessor, LogitsProcessorList
from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
from typing import List, Iterable, Optional, Union, Tuple
from collections import deque
import os
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.pytorch_utils import ALL_LAYERNORM_LAYERS
from transformers.utils import LossKwargs, auto_docstring, can_return_tuple, is_torch_flex_attn_available, logging
from .configuration_hyperclovax import HyperCLOVAXConfig
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__)
 
# ================= DeepConf: confidence-based online early stop =================
class DeepConfEOSLogitsProcessor(LogitsProcessor):
    """
    Per-sample early stop: at each step, compute token_conf = mean(logprob of top-r),
    maintain group_conf = mean of last `window` token_conf; if group_conf < threshold,
    force EOS for THAT sample by setting EOS logprob=0 and others to -inf.
    """
    def __init__(
        self,
        eos_token_ids: List[int],
        window: int = 512,
        top_r: int = 5,
        threshold: float = -3.5,
        warmup_tokens: int = 0,
        prefer_eos_ids: Optional[List[int]] = None,
        require_prev_id: Optional[int] = None,
    ):
        self.eos_ids: List[int] = sorted({int(i) for i in (eos_token_ids or []) if i is not None and i >= 0})
        self.window: int = max(int(window), 1)
        self.top_r: int = max(int(top_r), 1)
        self.threshold: float = float(threshold)
        self.warmup_tokens: int = max(int(warmup_tokens), 0)
        self.prefer_eos_ids: List[int] = sorted({int(i) for i in (prefer_eos_ids or []) if i is not None and i >= 0})
        self.require_prev_id = require_prev_id
        self._buffers: Optional[List[deque]] = None
        self._verbose: bool = os.getenv("HYPERCLOVA_DEEPCONF_VERBOSE", "0").strip().lower() in {"1","on","true"}
        self._every: int = max(int(os.getenv("HYPERCLOVA_DEEPCONF_REPORT_EVERY", "64")), 1)
        self._tick: int = 0
        self._stops: int = 0

    def _ensure(self, bsz: int) -> None:
        if self._buffers is None or len(self._buffers) != bsz:
            self._buffers = [deque(maxlen=self.window) for _ in range(bsz)]

    @torch.no_grad()
    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor) -> torch.FloatTensor:
        bsz, vocab = scores.shape
        self._ensure(bsz)
        logprobs = torch.log_softmax(scores, dim=-1)         # (B, V)
        k = min(self.top_r, vocab)
        token_conf = torch.topk(logprobs, k=k, dim=-1).values.mean(dim=-1)  # (B,)

        stopped = False
        for i, c in enumerate(token_conf.tolist()):
            buf = self._buffers[i]; buf.append(c)
            group_conf = sum(buf) / len(buf)
            # --- warmup gate: do not early-stop until we have enough tokens ---
            if len(buf) < self.warmup_tokens:
                continue

            # ChatML protection: only force preferred EOS after the required previous token
            prev_ok = True
            if self.require_prev_id is not None:
                prev_tok = int(input_ids[i, -1].item()) if input_ids is not None and input_ids.size(1) > 0 else -1
                prev_ok = (prev_tok == self.require_prev_id)

            if group_conf < self.threshold and (self.prefer_eos_ids or self.eos_ids) and prev_ok:
                # Prefer ChatML end tokens if available; else fall back to config eos
                targets = self.prefer_eos_ids if self.prefer_eos_ids else self.eos_ids
                scores[i].fill_(-float("inf"))
                for eid in targets:
                    if 0 <= eid < vocab:
                        scores[i, eid] = 0.0
                self._stops += 1
                stopped = True

        if self._verbose:
            self._tick += 1
            if self._tick % self._every == 0:
                try:
                    gcs = [(sum(b)/len(b)) if b else float("nan") for b in (self._buffers or [])]
                    valid = [x for x in gcs if not (x != x)]
                    mean_gc = float(sum(valid)/max(1, len(valid)))
                except Exception:
                    mean_gc = float("nan")
                print(f"[DeepConf] step={self._tick} mean_gc={mean_gc:.4f} stops={self._stops}")
        return scores

# (optional) Offline helper: Lowest Group Confidence (LGC)
def deepconf_lgc_from_scores(scores_list: Iterable[torch.Tensor], top_r: int = 5, window: int = 2048) -> float:
    tensors = [s for s in scores_list]
    if not tensors: return float("-inf")
    with torch.no_grad():
        vals = [
            torch.topk(torch.log_softmax(s, dim=-1), k=min(top_r, s.size(-1)), dim=-1).values.mean(dim=-1)
            for s in tensors
        ]  # each (B,)
        conf = torch.stack(vals).squeeze(-1)  # (T,) if B=1
        w = min(int(window), conf.numel())
        kernel = torch.ones(1,1,w, device=conf.device) / w
        run = torch.nn.functional.conv1d(conf.view(1,1,-1), weight=kernel).squeeze()
        return float(run.min().item())
# ============================================================================== 


@use_kernel_forward_from_hub("RMSNorm")
class HyperCLOVAXRMSNorm(nn.Module):
    def __init__(self, hidden_size, eps=1e-6):
        """
        HyperCLOVAXRMSNorm is equivalent to T5LayerNorm
        """
        super().__init__()
        self.weight = nn.Parameter(torch.ones(hidden_size))
        self.variance_epsilon = eps

    def forward(self, hidden_states):
        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):
        return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"

ALL_LAYERNORM_LAYERS.append(HyperCLOVAXRMSNorm)
class HyperCLOVAXRotaryEmbedding(nn.Module):
    def __init__(self, config: HyperCLOVAXConfig, 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  # power user: used with advanced RoPE types (e.g. dynamic rope)
    def forward(self, x, position_ids):
        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):  # Force float32
            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)


def rotate_half(x):
    """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, k, cos, sin, position_ids=None, unsqueeze_dim=1):
    """Applies Rotary Position Embedding to the query and key tensors.

    Args:
        q (`torch.Tensor`): The query tensor.
        k (`torch.Tensor`): The key tensor.
        cos (`torch.Tensor`): The cosine part of the rotary embedding.
        sin (`torch.Tensor`): The sine part of the rotary embedding.
        position_ids (`torch.Tensor`, *optional*):
            Deprecated and unused.
        unsqueeze_dim (`int`, *optional*, defaults to 1):
            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
    Returns:
        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
    """
    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, k_embed


class HyperCLOVAXMLP(nn.Module):
    def __init__(self, config):
        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):
        down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
        return down_proj


def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
    """
    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,
):
    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 = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
    attn_weights = nn.functional.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


class HyperCLOVAXAttention(nn.Module):
    """Multi-headed attention from 'Attention Is All You Need' paper"""

    def __init__(self, config: HyperCLOVAXConfig, 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 = getattr(config, "attention_multiplier", self.head_dim**-0.5) # MuP
        self.attention_dropout = config.attention_dropout
        self.is_causal = True

        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)

        query_states = self.q_proj(hidden_states).view(hidden_shape).transpose(1, 2)
        key_states = self.k_proj(hidden_states).view(hidden_shape).transpose(1, 2)
        value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)

        cos, sin = position_embeddings
        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)

        if past_key_value is not None:
            # sin and cos are specific to RoPE models; cache_position needed for the static cache
            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)

        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]

        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,
            **kwargs,
        )

        attn_output = attn_output.reshape(*input_shape, -1).contiguous()
        attn_output = self.o_proj(attn_output)
        return attn_output, attn_weights


class HyperCLOVAXDecoderLayer(GradientCheckpointingLayer):
    def __init__(self, config: HyperCLOVAXConfig, layer_idx: int):
        super().__init__()
        self.hidden_size = config.hidden_size

        self.self_attn = HyperCLOVAXAttention(config=config, layer_idx=layer_idx)

        self.mlp = HyperCLOVAXMLP(config)
        self.input_layernorm = HyperCLOVAXRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_attention_layernorm = HyperCLOVAXRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.use_post_norm = getattr(config, "use_post_norm", False)

        # Peri-LN (post-norm)
        if self.use_post_norm:
            self.post_norm1 = HyperCLOVAXRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
            self.post_norm2 = HyperCLOVAXRMSNorm(config.hidden_size, eps=config.rms_norm_eps)

        self.residual_multiplier = getattr(config, "residual_multiplier", 1.0) # MuP

    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,  # necessary, but kept here for BC
        **kwargs: Unpack[FlashAttentionKwargs],
    ) -> tuple[torch.FloatTensor, Optional[tuple[torch.FloatTensor, torch.FloatTensor]]]:
        residual = hidden_states
        hidden_states = self.input_layernorm(hidden_states)

        # Self Attention
        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,
        )

        if self.use_post_norm: # Peri-LN
            hidden_states = self.post_norm1(hidden_states)

        hidden_states = residual + hidden_states * self.residual_multiplier # MuP

        # Fully Connected
        residual = hidden_states
        hidden_states = self.post_attention_layernorm(hidden_states)
        hidden_states = self.mlp(hidden_states)

        if self.use_post_norm: # Peri-LN
            hidden_states = self.post_norm2(hidden_states)

        hidden_states = residual + hidden_states * self.residual_multiplier # MuP

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

        return outputs


@auto_docstring
class HyperCLOVAXPreTrainedModel(PreTrainedModel):
    config_class = HyperCLOVAXConfig
    base_model_prefix = "model"
    supports_gradient_checkpointing = True
    _no_split_modules = ["HyperCLOVAXDecoderLayer"]
    _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):
        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, HyperCLOVAXRMSNorm):
            module.weight.data.fill_(1.0)


@auto_docstring
class HyperCLOVAXModel(HyperCLOVAXPreTrainedModel):
    def __init__(self, config: HyperCLOVAXConfig):
        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(
            [HyperCLOVAXDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
        )
        self.norm = HyperCLOVAXRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.rotary_emb = HyperCLOVAXRotaryEmbedding(config=config)
        self.gradient_checkpointing = False

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

        # MuP
        self.embedding_multiplier = getattr(config, "embedding_multiplier", 1.0)

    def get_input_embeddings(self):
        return self.embed_tokens

    def set_input_embeddings(self, value):
        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

        # TODO (joao): remove this exception in v4.56 -- it exists for users that try to pass a legacy cache
        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)

        inputs_embeds = inputs_embeds * self.embedding_multiplier # MuP

        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 (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
    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


class KwargsForCausalLM(FlashAttentionKwargs, LossKwargs): ...


@auto_docstring
class HyperCLOVAXForCausalLM(HyperCLOVAXPreTrainedModel, GenerationMixin):
    _tied_weights_keys = ["lm_head.weight"]
    _tp_plan = {"lm_head": "colwise_rep"}
    _pp_plan = {"lm_head": (["hidden_states"], ["logits"])}

    def __init__(self, config):
        super().__init__(config)
        self.model = HyperCLOVAXModel(config)
        self.vocab_size = config.vocab_size
        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
        self.logits_scaling = getattr(config, "logits_scaling", 1.0)

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

    def get_input_embeddings(self):
        return self.model.embed_tokens

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

    def get_output_embeddings(self):
        return self.lm_head

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

    # -------- DeepConf helpers ----------
    def _dc_collect_eos(self, explicit: Optional[Union[int, List[int]]] = None, **kwargs) -> List[int]:
        ids: List[int] = []
        if explicit is not None:
            ids.extend([int(x) for x in (explicit if isinstance(explicit, (list,tuple)) else [explicit])])
        else:
            if getattr(self.config, "eos_token_id", None) is not None:
                ids.append(int(self.config.eos_token_id))
            if getattr(self.config, "eos_token_id_list", None):
                ids.extend(int(x) for x in self.config.eos_token_id_list if x is not None)
        extra = os.getenv("HYPERCLOVA_DEEPCONF_EOS_IDS", "").strip()
        if extra:
            ids.extend(int(tok) for tok in extra.split(",") if tok.strip().isdigit())
        return sorted({i for i in ids if i >= 0})

    def _dc_enabled(self) -> bool:
        enabled = True
        env = os.getenv("HYPERCLOVA_DEEPCONF", "").strip().lower()
        if env in {"0","off","false"}: enabled = False
        elif env in {"1","on","true"}: enabled = True
        cfg_en = getattr(self.config, "deepconf_enable", None)
        if cfg_en is not None:
            enabled = bool(cfg_en)           # If config is specified, it takes precedence
        if getattr(self.config, "deepconf_disable", False):
            enabled = False                  # Force OFF flag
        return enabled

    def _dc_params(self) -> Tuple[int,int,float,int]:
        def env_int(k, d): v=os.getenv(k); return int(v) if v not in (None,"") else d
        def env_flt(k, d): v=os.getenv(k); return float(v) if v not in (None,"") else d
        window = env_int("HYPERCLOVA_DEEPCONF_WINDOW", getattr(self.config, "deepconf_window", 512))
        top_r  = env_int("HYPERCLOVA_DEEPCONF_TOPR",   getattr(self.config, "deepconf_top_r", 5))
        thr    = env_flt("HYPERCLOVA_DEEPCONF_THRESH", getattr(self.config, "deepconf_threshold", -3.5))
        warmup = env_int("HYPERCLOVA_DEEPCONF_WARMUP", getattr(self.config, "deepconf_warmup_tokens", 0))
        return window, top_r, thr, warmup

    def deepconf_generate(self, *args,
                          eos_token_id: Optional[Union[int, List[int]]] = None,
                          window: int = 512, top_r: int = 5, threshold: float = -3.5,
                          warmup_tokens: int = 0,
                          **kwargs):
        # Prefer ChatML stop strings if tokenizer+stop_strings are provided
        prefer_ids: List[int] = []
        tok = kwargs.get("tokenizer", None)
        stop_strings = kwargs.get("stop_strings", None)
        if tok is not None and stop_strings:
            for s in stop_strings:
                try:
                    eid = tok.convert_tokens_to_ids(s)
                    if isinstance(eid, int) and eid >= 0:
                        prefer_ids.append(int(eid)); continue
                except Exception:
                    pass
                try:
                    enc = tok.encode(s, add_special_tokens=False)
                    if isinstance(enc, list) and len(enc) == 1:
                        prefer_ids.append(int(enc[0]))
                except Exception:
                    pass
        lp: LogitsProcessorList = kwargs.pop("logits_processor", None) or LogitsProcessorList()
        lp.append(
            DeepConfEOSLogitsProcessor(
                self._dc_collect_eos(eos_token_id, **kwargs),
                window, top_r, threshold,
                warmup_tokens=warmup_tokens,
                prefer_eos_ids=prefer_ids or None
            )
        )
        kwargs["logits_processor"] = lp
        return super().generate(*args, **kwargs)

    # Override generate() to be default ON (auto-attach DeepConf; merge with external lps)
    def generate(self, *args, **kwargs):
        if self._dc_enabled():
            eos_ids = self._dc_collect_eos(kwargs.get("eos_token_id", None), **kwargs)
            # Prefer ChatML end tokens if provided
            prefer_ids: List[int] = []
            tok = kwargs.get("tokenizer", None)
            stop_strings = kwargs.get("stop_strings", None)
            im_end_id = None
            if tok is not None and stop_strings:
                for s in stop_strings:
                    try:
                        eid = tok.convert_tokens_to_ids(s)
                        if isinstance(eid, int) and eid >= 0: prefer_ids.append(int(eid)); continue
                    except Exception: pass
                    try:
                        enc = tok.encode(s, add_special_tokens=False)
                        if isinstance(enc, list) and len(enc) == 1: prefer_ids.append(int(enc[0]))
                    except Exception: pass

            # For ChatML protection: extract <|im_end|> id
            if tok is not None:
                try:
                    im_end_id = tok.convert_tokens_to_ids("<|im_end|>")
                    if not isinstance(im_end_id, int) or im_end_id < 0:
                        im_end_id = None
                except Exception:
                    im_end_id = None

            if eos_ids:
                window, top_r, thr, warmup = self._dc_params()
                require_prev = None
                if (os.getenv("HYPERCLOVA_DEEPCONF_REQUIRE_IM_END", "1").lower() in {"1","on","true"}) and prefer_ids and im_end_id is not None:
                    require_prev = im_end_id
                
                lp: LogitsProcessorList = kwargs.pop("logits_processor", None) or LogitsProcessorList()

                if os.getenv("HYPERCLOVA_DEEPCONF_VERBOSE_ATTACH","0") in {"1","on","true"}:
                    print(f"[DeepConf] attach window={window} top_r={top_r} thr={thr} warmup={warmup} eos={eos_ids} prefer={prefer_ids} require_prev={require_prev}")

                lp.append(
                    DeepConfEOSLogitsProcessor(
                        eos_ids, window, top_r, thr,
                        warmup_tokens=warmup,
                        prefer_eos_ids=prefer_ids or None,
                        require_prev_id=require_prev
                    )
                )
                kwargs["logits_processor"] = lp
        return super().generate(*args, **kwargs)

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

    def get_decoder(self):
        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"""
        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, HyperCLOVAXForCausalLM

        >>> model = HyperCLOVAXForCausalLM.from_pretrained("naver-hyperclovax/{model_name}")
        >>> tokenizer = AutoTokenizer.from_pretrained("naver-hyperclovax/{model_name}")

        >>> 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
        # MuP
        logits = self.lm_head(hidden_states[:, slice_indices, :]) * self.logits_scaling

        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 HyperCLOVAX Model transformer with a sequence classification head on top (linear layer).

    [`HyperCLOVAXForSequenceClassification`] 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. 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 HyperCLOVAXForSequenceClassification(HyperCLOVAXPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels
        self.model = HyperCLOVAXModel(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):
        return self.model.embed_tokens

    def set_input_embeddings(self, value):
        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:
            # 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:
            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 HyperCLOVAXForQuestionAnswering(HyperCLOVAXPreTrainedModel):
    base_model_prefix = "transformer"

    # Copied from transformers.models.bloom.modeling_bloom.BloomForQuestionAnswering.__init__ with Bloom->HyperCLOVAX
    def __init__(self, config):
        super().__init__(config)
        self.transformer = HyperCLOVAXModel(config)
        self.qa_outputs = nn.Linear(config.hidden_size, 2)

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

    def get_input_embeddings(self):
        return self.transformer.embed_tokens

    def set_input_embeddings(self, value):
        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,
        )


@auto_docstring
class HyperCLOVAXForTokenClassification(HyperCLOVAXPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels
        self.model = HyperCLOVAXModel(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):
        return self.model.embed_tokens

    def set_input_embeddings(self, value):
        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,
        )


__all__ = [
    "HyperCLOVAXForCausalLM",
    "HyperCLOVAXModel",
    "HyperCLOVAXPreTrainedModel",
    "HyperCLOVAXForSequenceClassification",
    "HyperCLOVAXForQuestionAnswering",
    "HyperCLOVAXForTokenClassification",
]