modelling_maincoder.py

# coding=utf-8
# Copyright 2025 Maincode. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Maincoder model implementation."""

from typing import Callable, Optional, Union

import torch
import torch.nn as nn

from transformers.activations import ACT2FN
from transformers.cache_utils import Cache, DynamicCache
from transformers.generation import GenerationMixin
from transformers.masking_utils import create_causal_mask
from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
from transformers.modeling_layers import GradientCheckpointingLayer
from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
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 TransformersKwargs, auto_docstring, can_return_tuple, logging

from .configuration_maincoder import MaincoderConfig


logger = logging.get_logger(__name__)


class MaincoderRMSNorm(nn.Module):
    """RMSNorm implementation equivalent to T5LayerNorm."""

    def __init__(self, hidden_size, eps=1e-5):
        """
        MatildaPlusRMSNorm is equivalent to T5LayerNorm
        """
        super().__init__()
        self.eps = eps
        self.weight = nn.Parameter(torch.ones(hidden_size))

    def _norm(self, x):
        return x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps)

    def forward(self, x):
        output = self._norm(x.float()).type_as(x)
        return output * self.weight

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


class MaincoderMLP(nn.Module):
    """SwiGLU-style MLP."""

    def __init__(self, config: MaincoderConfig):
        super().__init__()
        self.hidden_size = config.hidden_size
        self.intermediate_size = config.intermediate_size_mlp

        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
        self.act_fn = ACT2FN[config.hidden_act]

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


class MaincoderRotaryEmbedding(nn.Module):
    """Rotary Position Embedding."""

    def __init__(self, config: MaincoderConfig, device=None):
        super().__init__()
        self.rope_type = "llama3" if config.rope_scaling is not None else "default"
        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)

    @torch.no_grad()
    @dynamic_rope_update
    def forward(self, x: torch.Tensor, position_ids: torch.Tensor) -> torch.Tensor:
        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1)
        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.to(x.device) @ position_ids_expanded).transpose(1, 2)
            freqs_cis = torch.polar(torch.ones_like(freqs), freqs)
            freqs_cis = freqs_cis * self.attention_scaling

        return freqs_cis


def apply_rotary_emb(
    xq: torch.Tensor,
    xk: torch.Tensor,
    freqs_cis: torch.Tensor,
) -> tuple[torch.Tensor, torch.Tensor]:
    """Apply rotary embeddings to query and key tensors."""
    xq_ = torch.view_as_complex(xq.float().reshape(*xq.shape[:-1], -1, 2))
    xk_ = torch.view_as_complex(xk.float().reshape(*xk.shape[:-1], -1, 2))

    # Broadcast freqs_cis
    freqs_cis = freqs_cis[:, :, None, :]

    xq_out = torch.view_as_real(xq_ * freqs_cis).flatten(3)
    xk_out = torch.view_as_real(xk_ * freqs_cis).flatten(3)

    return xq_out.type_as(xq), xk_out.type_as(xk)


def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
    """Repeat key/value heads to match query heads for GQA."""
    if n_rep == 1:
        return hidden_states
    batch, num_kv_heads, slen, head_dim = hidden_states.shape
    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_kv_heads, n_rep, slen, head_dim)
    return hidden_states.reshape(batch, num_kv_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]:
    """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:
        attn_weights = attn_weights + attention_mask[:, :, :, : key_states.shape[-2]]

    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 MaincoderAttention(nn.Module):
    """Multi-headed attention with Grouped Query Attention (GQA) and RoPE."""

    def __init__(self, config: MaincoderConfig, layer_idx: int):
        super().__init__()
        self.config = config
        self.layer_idx = layer_idx
        self.head_dim = config.head_dim
        self.num_attention_heads = config.num_attention_heads
        self.num_key_value_heads = config.num_key_value_heads
        self.num_key_value_groups = self.num_attention_heads // self.num_key_value_heads
        self.scaling = self.head_dim**-0.5
        self.attention_dropout = config.attention_dropout

        self.q_proj = nn.Linear(config.hidden_size, self.num_attention_heads * self.head_dim, bias=False)
        self.k_proj = nn.Linear(config.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
        self.v_proj = nn.Linear(config.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
        self.o_proj = nn.Linear(self.num_attention_heads * self.head_dim, config.hidden_size, bias=False)

        # QK normalization
        if config.use_qk_norm:
            self.q_norm = MaincoderRMSNorm(self.head_dim, eps=config.rms_norm_eps)
            self.k_norm = MaincoderRMSNorm(self.head_dim, eps=config.rms_norm_eps)

    def forward(
        self,
        hidden_states: torch.Tensor,
        position_embeddings: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        past_key_values: Optional[Cache] = None,
        cache_position: Optional[torch.LongTensor] = None,
        **kwargs: Unpack[FlashAttentionKwargs],
    ) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
        batch_size, seq_len, _ = hidden_states.shape

        query_states = self.q_proj(hidden_states).view(batch_size, seq_len, self.num_attention_heads, self.head_dim)
        key_states = self.k_proj(hidden_states).view(batch_size, seq_len, self.num_key_value_heads, self.head_dim)
        value_states = self.v_proj(hidden_states).view(batch_size, seq_len, self.num_key_value_heads, self.head_dim)

        # Apply RoPE
        query_states, key_states = apply_rotary_emb(query_states, key_states, position_embeddings)

        # Apply QK normalization
        if hasattr(self, "q_norm"):
            query_states = self.q_norm(query_states)
            key_states = self.k_norm(key_states)

        # Transpose for attention: (batch, heads, seq, head_dim)
        query_states = query_states.transpose(1, 2)
        key_states = key_states.transpose(1, 2)
        value_states = value_states.transpose(1, 2)

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

        # Attention
        attention_fn: Callable = eager_attention_forward
        if self.config._attn_implementation != "eager":
            attention_fn = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]

        attn_output, attn_weights = attention_fn(
            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(batch_size, seq_len, -1)
        attn_output = self.o_proj(attn_output)

        return attn_output, attn_weights


class MaincoderDecoderLayer(GradientCheckpointingLayer):
    """Transformer decoder layer with pre-norm architecture."""

    def __init__(self, config: MaincoderConfig, layer_idx: int):
        super().__init__()
        self.self_attn = MaincoderAttention(config, layer_idx)
        self.feed_forward = MaincoderMLP(config)
        self.input_layernorm = MaincoderRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_attention_layernorm = MaincoderRMSNorm(config.hidden_size, eps=config.rms_norm_eps)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_embeddings: Optional[torch.Tensor] = None,
        past_key_values: Optional[Cache] = None,
        cache_position: Optional[torch.LongTensor] = None,
        **kwargs: Unpack[FlashAttentionKwargs],
    ) -> torch.Tensor:
        # Self Attention
        residual = hidden_states
        hidden_states = self.input_layernorm(hidden_states)
        hidden_states, _ = self.self_attn(
            hidden_states=hidden_states,
            position_embeddings=position_embeddings,
            attention_mask=attention_mask,
            past_key_values=past_key_values,
            cache_position=cache_position,
            **kwargs,
        )
        hidden_states = residual + hidden_states

        # Feed Forward
        residual = hidden_states
        hidden_states = self.post_attention_layernorm(hidden_states)
        hidden_states = self.feed_forward(hidden_states)
        hidden_states = residual + hidden_states

        return hidden_states


@auto_docstring
class MaincoderPreTrainedModel(PreTrainedModel):
    """Base class for Maincoder models."""

    config_class = MaincoderConfig
    base_model_prefix = "model"
    supports_gradient_checkpointing = True
    _no_split_modules = ["MaincoderDecoderLayer"]
    _skip_keys_device_placement = ["past_key_values"]
    _supports_sdpa = True
    _supports_flex_attn = 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, MaincoderRMSNorm):
            module.weight.data.fill_(1.0)


@auto_docstring
class MaincoderModel(MaincoderPreTrainedModel):
    """Maincoder transformer model outputting raw hidden states."""

    def __init__(self, config: MaincoderConfig):
        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(
            [MaincoderDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
        )
        self.norm = MaincoderRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.rotary_emb = MaincoderRotaryEmbedding(config)

        self.post_init()

    @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,
        cache_position: Optional[torch.LongTensor] = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> Union[tuple, BaseModelOutputWithPast]:
        if (input_ids is None) ^ (inputs_embeds is not None):
            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")

        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)

        # Create causal mask
        causal_mask = create_causal_mask(
            config=self.config,
            input_embeds=inputs_embeds,
            attention_mask=attention_mask,
            cache_position=cache_position,
            past_key_values=past_key_values,
        )

        # Position embeddings
        position_embeddings = self.rotary_emb(inputs_embeds, position_ids)

        hidden_states = inputs_embeds
        for layer in self.layers:
            hidden_states = layer(
                hidden_states,
                attention_mask=causal_mask,
                position_embeddings=position_embeddings,
                past_key_values=past_key_values,
                cache_position=cache_position,
                **kwargs,
            )

        hidden_states = self.norm(hidden_states)

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


class MaincoderForCausalLM(MaincoderPreTrainedModel, GenerationMixin):
    """Maincoder model with a causal language modeling head."""

    _tied_weights_keys = ["lm_head.weight"]

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

        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

    @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,
        cache_position: Optional[torch.LongTensor] = None,
        logits_to_keep: Union[int, torch.Tensor] = 0,
        **kwargs: Unpack[TransformersKwargs],
    ) -> Union[tuple, 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
        >>> from modelling_maincoder import MaincoderForCausalLM

        >>> model = MaincoderForCausalLM.from_pretrained("maincoder/maincoder")
        >>> tokenizer = AutoTokenizer.from_pretrained("maincoder/maincoder")

        >>> prompt = "def hello_world():"
        >>> inputs = tokenizer(prompt, return_tensors="pt")

        >>> generate_ids = model.generate(inputs.input_ids, max_length=50)
        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True)[0]
        ```"""
        outputs = 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,
            cache_position=cache_position,
            **kwargs,
        )

        hidden_states = outputs.last_hidden_state

        # Only compute logits for tokens we need
        if isinstance(logits_to_keep, int) and logits_to_keep > 0:
            hidden_states = hidden_states[:, -logits_to_keep:, :]

        logits = self.lm_head(hidden_states)

        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,
        )


__all__ = [
    "MaincoderConfig",
    "MaincoderPreTrainedModel",
    "MaincoderModel",
    "MaincoderForCausalLM",
]