modeling_aether_micro.py

#!/usr/bin/env python3
"""
AETHER-Micro Model Implementation (Hugging Face Standard)

모듈화 구조:
- utils.py: Helper functions
- normalization.py: RMSNorm
- embeddings.py: RoPE
- attention.py: Multi-Head Attention
- router.py: Wu-Xing Router
- moe.py: Heterogeneous MoE
- layers.py: Decoder Layer
- modeling_aether_micro.py: Main Model (이 파일)
"""

import torch
import torch.nn as nn
import torch.utils.checkpoint
from typing import Optional, Tuple, Union

from transformers import PreTrainedModel, GenerationMixin
from transformers.modeling_outputs import CausalLMOutputWithPast, BaseModelOutputWithPast

from .configuration_aether_micro import AETHERMicroConfig
from .normalization import AETHERMicroRMSNorm
from .layers import AETHERMicroDecoderLayer
from .quality_head import AETHERMicroQualityHead
from .mtp_loss import MTPLoss


# ========================================
# PreTrained Model Base Class
# ========================================

class AETHERMicroPreTrainedModel(PreTrainedModel):
    """
    AETHER-Micro PreTrained Model Base Class

    모든 AETHER-Micro 모델의 기본 클래스입니다.
    HF의 save_pretrained, from_pretrained 기능을 제공합니다.
    """

    config_class = AETHERMicroConfig
    base_model_prefix = "model"
    supports_gradient_checkpointing = True
    _no_split_modules = ["AETHERMicroDecoderLayer"]
    _skip_keys_device_placement = "past_key_values"

    def _init_weights(self, module):
        """
        Initialize weights

        Args:
            module: nn.Module to initialize
        """
        std = self.config.initializer_range if hasattr(self.config, 'initializer_range') else 0.02

        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_()

    def _set_gradient_checkpointing(self, module, value=False):
        """Enable gradient checkpointing"""
        if isinstance(module, AETHERMicroModel):
            module.gradient_checkpointing = value


# ========================================
# Main Transformer Model
# ========================================

class AETHERMicroModel(AETHERMicroPreTrainedModel):
    """
    Main Transformer Model

    Structure:
    - Embedding layer
    - 24 Decoder layers
    - Output RMSNorm
    """

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

        # Embedding
        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)

        # Decoder layers
        self.layers = nn.ModuleList([
            AETHERMicroDecoderLayer(config)
            for _ in range(config.num_hidden_layers)
        ])

        # Output normalization
        self.norm = AETHERMicroRMSNorm(config.hidden_size, eps=config.rms_norm_eps)

        self.gradient_checkpointing = False

        # Initialize weights
        self.post_init()

    def get_input_embeddings(self):
        return self.embed_tokens

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

    def forward(
        self,
        input_ids: torch.LongTensor = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        disable_ltl: Optional[bool] = False,
    ) -> Union[Tuple, BaseModelOutputWithPast]:
        """
        Args:
            input_ids: (batch_size, sequence_length)
            attention_mask: (batch_size, sequence_length)
            position_ids: (batch_size, sequence_length)
            inputs_embeds: (batch_size, sequence_length, hidden_size)

        Returns:
            BaseModelOutputWithPast or tuple
        """
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        use_cache = use_cache if use_cache is not None else self.config.use_cache
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        # Retrieve input_ids and inputs_embeds
        if input_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
        elif input_ids is not None:
            batch_size, seq_length = input_ids.shape
        elif inputs_embeds is not None:
            batch_size, seq_length, _ = inputs_embeds.shape
        else:
            raise ValueError("You have to specify either input_ids or inputs_embeds")

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

        hidden_states = inputs_embeds

        # Position IDs
        if position_ids is None:
            device = input_ids.device if input_ids is not None else inputs_embeds.device
            position_ids = torch.arange(
                0, seq_length, dtype=torch.long, device=device
            )
            position_ids = position_ids.unsqueeze(0).expand(batch_size, -1)

        # Attention mask (causal)
        if attention_mask is None:
            attention_mask = torch.ones(
                (batch_size, seq_length), dtype=torch.bool, device=hidden_states.device
            )

        # Causal mask: lower triangular
        attention_mask = self._prepare_decoder_attention_mask(
            attention_mask, (batch_size, seq_length), hidden_states, 0
        )

        # 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:
            if output_hidden_states:
                all_hidden_states += (hidden_states,)

            if self.gradient_checkpointing and self.training:
                # PyTorch 2.7+ non-reentrant mode (권장)
                # decoder_layer.forward()가 항상 단일 tensor 반환하도록 수정됨
                hidden_states = torch.utils.checkpoint.checkpoint(
                    decoder_layer,
                    hidden_states,
                    attention_mask,
                    position_ids,
                    disable_ltl,
                    use_reentrant=False
                )
            else:
                hidden_states = decoder_layer(
                    hidden_states,
                    attention_mask=attention_mask,
                    position_ids=position_ids,
                    disable_ltl=disable_ltl,
                )

        # Output normalization
        hidden_states = self.norm(hidden_states)

        # Add last hidden state
        if output_hidden_states:
            all_hidden_states += (hidden_states,)

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

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

    def _prepare_decoder_attention_mask(self, attention_mask, input_shape, inputs_embeds, past_key_values_length):
        """
        Prepare causal attention mask

        Args:
            attention_mask: (batch_size, seq_length)
            input_shape: (batch_size, seq_length)
            inputs_embeds: embeddings tensor
            past_key_values_length: 0 for training

        Returns:
            combined_attention_mask: (batch_size, 1, seq_length, seq_length)
        """
        # Create causal mask
        # [batch_size, seq_length] -> [batch_size, 1, tgt_seq_length, src_seq_length]
        combined_attention_mask = None
        if input_shape[-1] > 1:
            combined_attention_mask = _make_causal_mask(
                input_shape,
                inputs_embeds.dtype,
                device=inputs_embeds.device,
                past_key_values_length=past_key_values_length,
            )

        if attention_mask is not None:
            # [batch_size, seq_length] -> [batch_size, 1, tgt_seq_length, src_seq_length]
            expanded_attn_mask = _expand_mask(attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]).to(
                inputs_embeds.device
            )
            combined_attention_mask = (
                expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
            )

        return combined_attention_mask


# ========================================
# Causal Language Model
# ========================================

class AETHERMicroForCausalLM(AETHERMicroPreTrainedModel, GenerationMixin):
    """
    AETHER-Micro Causal Language Model

    Structure:
    - AETHERMicroModel (base transformer)
    - LM Head (hidden → vocab)
    - Loss computation
    """

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

        # Quality Head (Block 3)
        if config.enable_quality_head:
            self.quality_head = AETHERMicroQualityHead(config)

        # MTP Loss (Block 5)
        if config.enable_mtp_loss:
            self.mtp_loss = MTPLoss(config)

        # 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

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

    def get_decoder(self):
        return self.model

    def forward(
        self,
        input_ids: torch.LongTensor = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[list] = 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,
        return_dict: Optional[bool] = None,
        disable_ltl: Optional[bool] = False,
    ) -> Union[Tuple, CausalLMOutputWithPast]:
        """
        Args:
            input_ids: (batch_size, sequence_length)
            labels: (batch_size, sequence_length) - for training

        Returns:
            CausalLMOutputWithPast with loss, logits
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        # Forward through base model
        outputs = self.model(
            input_ids=input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            disable_ltl=disable_ltl,
        )

        hidden_states = outputs[0] if not return_dict else outputs.last_hidden_state
        logits = self.lm_head(hidden_states)
        logits = logits.float()

        # Quality Head (Block 3)
        quality_scores = None
        if hasattr(self, 'quality_head'):
            quality_scores = self.quality_head(hidden_states)

        loss = None
        mtp_metrics = None
        if labels is not None:
            if hasattr(self, 'mtp_loss') and self.config.enable_mtp_loss:
                # MTP Loss (Block 5)
                loss, mtp_metrics = self.mtp_loss(hidden_states, labels)
            else:
                # Standard NTP Loss
                # Shift so that tokens < n predict n
                shift_logits = logits[..., :-1, :].contiguous()
                shift_labels = labels[..., 1:].contiguous()
                # Flatten the tokens
                loss_fct = nn.CrossEntropyLoss()
                shift_logits = shift_logits.view(-1, self.config.vocab_size)
                shift_labels = shift_labels.view(-1)
                # Enable model parallelism
                shift_labels = shift_labels.to(shift_logits.device)
                loss = loss_fct(shift_logits, shift_labels)

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

        return CausalLMOutputWithPast(
            loss=loss,
            logits=logits,
            past_key_values=outputs.past_key_values if hasattr(outputs, 'past_key_values') else None,
            hidden_states=outputs.hidden_states if hasattr(outputs, 'hidden_states') else None,
            attentions=outputs.attentions if hasattr(outputs, 'attentions') else None,
        )

    def prepare_inputs_for_generation(
        self, input_ids, past_key_values=None, attention_mask=None, inputs_embeds=None, **kwargs
    ):
        """Prepare inputs for generation"""
        if past_key_values:
            input_ids = input_ids[:, -1:]

        position_ids = kwargs.get("position_ids", None)
        if attention_mask is not None and position_ids is None:
            # create position_ids on the fly for batch generation
            position_ids = attention_mask.long().cumsum(-1) - 1
            position_ids.masked_fill_(attention_mask == 0, 1)
            if past_key_values:
                position_ids = position_ids[:, -1].unsqueeze(-1)

        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
        if inputs_embeds is not None and past_key_values is None:
            model_inputs = {"inputs_embeds": inputs_embeds}
        else:
            model_inputs = {"input_ids": input_ids}

        model_inputs.update(
            {
                "position_ids": position_ids,
                "past_key_values": past_key_values,
                "use_cache": kwargs.get("use_cache"),
                "attention_mask": attention_mask,
            }
        )
        return model_inputs

    @staticmethod
    def _reorder_cache(past_key_values, beam_idx):
        """Reorder cache for beam search"""
        reordered_past = ()
        for layer_past in past_key_values:
            reordered_past += (
                tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past),
            )
        return reordered_past


# ========================================
# Helper Functions for Attention Mask
# ========================================

def _make_causal_mask(
    input_ids_shape: torch.Size, dtype: torch.dtype, device: torch.device, past_key_values_length: int = 0
):
    """
    Make causal mask used for bi-directional self-attention.
    """
    bsz, tgt_len = input_ids_shape
    mask = torch.full((tgt_len, tgt_len), torch.finfo(dtype).min, device=device)
    mask_cond = torch.arange(mask.size(-1), device=device)
    mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
    mask = mask.to(dtype)

    if past_key_values_length > 0:
        mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype, device=device), mask], dim=-1)
    return mask[None, None, :, :].expand(bsz, 1, tgt_len, tgt_len + past_key_values_length)


def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
    """
    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
    """
    bsz, src_len = mask.size()
    tgt_len = tgt_len if tgt_len is not None else src_len

    expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)

    inverted_mask = 1.0 - expanded_mask

    return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)


# ========================================
# Export all classes
# ========================================

__all__ = [
    "AETHERMicroConfig",
    "AETHERMicroPreTrainedModel",
    "AETHERMicroModel",
    "AETHERMicroForCausalLM",
]