| """ |
| MotherConfig — HuggingFace-compatible configuration for MOTHER CORE. |
| |
| This is a thin metadata wrapper. The actual architecture parameters |
| come from mother_core.config.ModelConfig. This class exists so HF tools |
| can read 'what architecture am I loading' without needing to import |
| MOTHER CORE internals. |
| """ |
| from transformers import PretrainedConfig |
|
|
|
|
| class MotherConfig(PretrainedConfig): |
| model_type = "mother_core" |
| keys_to_ignore_at_inference = ["past_key_values"] |
|
|
| def __init__( |
| self, |
| vocab_size: int = 50258, |
| hidden_size: int = 3072, |
| num_hidden_layers: int = 28, |
| num_attention_heads: int = 24, |
| num_key_value_heads: int = 6, |
| ff_mult: float = 4.0, |
| max_position_embeddings: int = 1024, |
| rope_theta: float = 10000.0, |
| rms_norm_eps: float = 1e-5, |
| tie_word_embeddings: bool = False, |
| bos_token_id: int = 1, |
| eos_token_id: int = 2, |
| pad_token_id: int = 0, |
| |
| model_name: str = "MOTHER CORE", |
| model_version: str = "v11", |
| architect: str = "Christopher Kenna", |
| organisation: str = "MediaStream AI Limited", |
| sovereign_nation: str = "United Kingdom", |
| **kwargs, |
| ): |
| self.vocab_size = vocab_size |
| self.hidden_size = hidden_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.ff_mult = ff_mult |
| self.max_position_embeddings = max_position_embeddings |
| self.rope_theta = rope_theta |
| self.rms_norm_eps = rms_norm_eps |
| self.model_name = model_name |
| self.model_version = model_version |
| self.architect = architect |
| self.organisation = organisation |
| self.sovereign_nation = sovereign_nation |
| super().__init__( |
| tie_word_embeddings=tie_word_embeddings, |
| bos_token_id=bos_token_id, |
| eos_token_id=eos_token_id, |
| pad_token_id=pad_token_id, |
| **kwargs, |
| ) |
|
|
