| """ | |
| Configuration class for Ivme-Conversate-S-v2-Instruct. | |
| Mirrors train_conversate_s_v2_instruct.ModelConfig exactly (same field | |
| names, same defaults), so config.json produced from a real training run's | |
| ModelConfig round-trips into this class with no field remapping needed. | |
| """ | |
| from transformers import PretrainedConfig | |
| class IvmeConversateSV2InstructConfig(PretrainedConfig): | |
| model_type = "ivme_conversate_s_v2_instruct" | |
| def __init__( | |
| self, | |
| vocab_size: int = 8000, | |
| d_model: int = 224, | |
| n_layers: int = 9, | |
| n_heads: int = 7, | |
| d_ff: int = 896, | |
| max_seq_len: int = 1024, | |
| norm_eps: float = 1e-5, | |
| rope_theta: float = 10000.0, | |
| **kwargs, | |
| ): | |
| self.vocab_size = vocab_size | |
| self.d_model = d_model | |
| self.n_layers = n_layers | |
| self.n_heads = n_heads | |
| self.d_ff = d_ff | |
| self.max_seq_len = max_seq_len | |
| self.norm_eps = norm_eps | |
| self.rope_theta = rope_theta | |
| kwargs.setdefault("tie_word_embeddings", True) | |
| super().__init__(**kwargs) | |