configuration_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 configuration."""

from typing import Optional

from transformers.configuration_utils import PretrainedConfig
from transformers.utils import logging


logger = logging.get_logger(__name__)


class MaincoderConfig(PretrainedConfig):
    r"""
    Configuration class for Maincoder model.

    Args:
        vocab_size (`int`, *optional*, defaults to 151936):
            Vocabulary size of the Maincoder model.
        hidden_size (`int`, *optional*, defaults to 1536):
            Dimension of the hidden representations.
        intermediate_size (`int`, *optional*, defaults to 4096):
            Dimension of the MLP intermediate representations.
        intermediate_size_mlp (`int`, *optional*, defaults to 4096):
            Dimension of the MLP representations (same as intermediate_size for dense models).
        num_hidden_layers (`int`, *optional*, defaults to 32):
            Number of hidden layers in the Transformer decoder.
        num_attention_heads (`int`, *optional*, defaults to 16):
            Number of attention heads for each attention layer.
        num_key_value_heads (`int`, *optional*, defaults to 4):
            Number of key-value heads for Grouped Query Attention (GQA).
        head_dim (`int`, *optional*, defaults to 96):
            Dimension of each attention head.
        hidden_act (`str`, *optional*, defaults to `"silu"`):
            The activation function in the MLP.
        max_position_embeddings (`int`, *optional*, defaults to 2048):
            Maximum sequence length the model can handle.
        initializer_range (`float`, *optional*, defaults to 0.02):
            Standard deviation for weight initialization.
        rms_norm_eps (`float`, *optional*, defaults to 1e-05):
            Epsilon for RMS normalization layers.
        use_cache (`bool`, *optional*, defaults to `True`):
            Whether to use key-value cache for generation.
        pad_token_id (`int`, *optional*, defaults to 151643):
            Padding token id.
        bos_token_id (`int`, *optional*):
            Beginning of sequence token id.
        eos_token_id (`int`, *optional*, defaults to 151643):
            End of sequence token id.
        tie_word_embeddings (`bool`, *optional*, defaults to `True`):
            Whether to tie input and output embeddings.
        rope_theta (`float`, *optional*, defaults to 1000000.0):
            Base period for RoPE embeddings.
        rope_scaling (`Dict`, *optional*):
            RoPE scaling configuration for extended context.
        attention_dropout (`float`, *optional*, defaults to 0.0):
            Dropout probability for attention weights.
        use_qk_norm (`bool`, *optional*, defaults to `True`):
            Whether to apply RMS normalization to query and key.

    Example:
    ```python
    >>> from configuration_maincoder import MaincoderConfig
    >>> from modelling_maincoder import MaincoderForCausalLM

    >>> config = MaincoderConfig()
    >>> model = MaincoderForCausalLM(config)
    ```
    """

    model_type = "maincoder"
    keys_to_ignore_at_inference = ["past_key_values"]

    def __init__(
        self,
        vocab_size: int = 151936,
        hidden_size: int = 1536,
        intermediate_size: int = 4096,
        intermediate_size_mlp: int = 4096,
        num_hidden_layers: int = 32,
        num_attention_heads: int = 16,
        num_key_value_heads: Optional[int] = 4,
        head_dim: Optional[int] = 96,
        hidden_act: str = "silu",
        max_position_embeddings: int = 2048,
        initializer_range: float = 0.02,
        rms_norm_eps: float = 1e-5,
        use_cache: bool = True,
        pad_token_id: Optional[int] = 151643,
        bos_token_id: Optional[int] = None,
        eos_token_id: int = 151643,
        tie_word_embeddings: bool = True,
        rope_theta: float = 1000000.0,
        rope_scaling: Optional[dict] = None,
        attention_dropout: float = 0.0,
        use_qk_norm: bool = True,
        **kwargs,
    ):
        self.vocab_size = vocab_size
        self.hidden_size = hidden_size
        self.intermediate_size = intermediate_size
        self.intermediate_size_mlp = intermediate_size_mlp
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.max_position_embeddings = max_position_embeddings
        self.initializer_range = initializer_range
        self.rms_norm_eps = rms_norm_eps
        self.use_cache = use_cache
        self.rope_theta = rope_theta
        self.rope_scaling = rope_scaling
        self.attention_dropout = attention_dropout
        self.use_qk_norm = use_qk_norm
        self.hidden_act = hidden_act

        # GQA configuration
        self.num_key_value_heads = num_key_value_heads if num_key_value_heads is not None else num_attention_heads
        self.head_dim = head_dim if head_dim is not None else self.hidden_size // self.num_attention_heads

        super().__init__(
            pad_token_id=pad_token_id,
            bos_token_id=bos_token_id,
            eos_token_id=eos_token_id,
            tie_word_embeddings=tie_word_embeddings,
            **kwargs,
        )


__all__ = ["MaincoderConfig"]