aether_pkg/configuration_aether_v2_7way.py

# coding=utf-8
"""
AETHER-V2-7way Configuration.

AETHERV27way configuration.

49-layer 7×7 Latin Square MoE Transformer with 7-aware attention.
"""
from transformers.configuration_utils import PretrainedConfig


class AETHERV27wayConfig(PretrainedConfig):
    """Configuration for AETHER-V2-7way.

    7×7 Latin Square: 49 layers × 7 attention types
    25-expert MoE × top-7 active per token
    5-element cyclic FFN phase ()
    """
    model_type = "aether_v2_7way"
    keys_to_ignore_at_inference = ["past_key_values"]

    def __init__(
        self,
        # Vocab
        vocab_size: int = 151936,
        pad_token_id: int = 151643,

        # Model dimensions
        hidden_size: int = 4096,
        intermediate_size: int = 12288,
        num_hidden_layers: int = 49,

        # Attention
        num_attention_heads: int = 32,
        num_key_value_heads: int = 8,
        head_dim: int = 128,
        attention_dropout: float = 0.0,
        sliding_window_size: int = 2048,
        compress_block_size: int = 64,

        # MoE
        num_experts: int = 25,
        num_experts_per_tok: int = 7,
        expert_intermediate_size: int = 1280,
        use_shared_expert: bool = True,
        router_aux_loss_coef: float = 0.001,

        # Position
        max_position_embeddings: int = 8192,
        rope_theta: float = 1000000.0,

        # Norm + activation
        rms_norm_eps: float = 1e-6,
        hidden_act: str = "silu",
        initializer_range: float = 0.02,

        # Training / inference
        use_cache: bool = True,
        output_router_logits: bool = False,
        tie_word_embeddings: bool = False,

        **kwargs,
    ):
        self.vocab_size = vocab_size
        self.hidden_size = hidden_size
        self.intermediate_size = intermediate_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.num_key_value_heads = num_key_value_heads
        self.head_dim = head_dim
        self.attention_dropout = attention_dropout
        self.sliding_window_size = sliding_window_size
        self.compress_block_size = compress_block_size
        self.num_experts = num_experts
        self.num_experts_per_tok = num_experts_per_tok
        self.expert_intermediate_size = expert_intermediate_size
        self.use_shared_expert = use_shared_expert
        self.router_aux_loss_coef = router_aux_loss_coef
        self.max_position_embeddings = max_position_embeddings
        self.rope_theta = rope_theta
        self.rms_norm_eps = rms_norm_eps
        self.hidden_act = hidden_act
        self.initializer_range = initializer_range
        self.use_cache = use_cache
        self.output_router_logits = output_router_logits

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


__all__ = ["AETHERV27wayConfig"]