| from transformers import PretrainedConfig | |
| class TalkieConfig(PretrainedConfig): | |
| model_type = "talkie" | |
| def __init__( | |
| self, | |
| vocab_size: int = 65536, | |
| hidden_size: int = 5120, | |
| num_hidden_layers: int = 40, | |
| num_attention_heads: int = 40, | |
| head_dim: int = 128, | |
| intermediate_size: int = 13696, | |
| max_position_embeddings: int = 2048, | |
| rope_theta: float = 1_000_000.0, | |
| tie_word_embeddings: bool = False, | |
| **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.head_dim = head_dim | |
| self.intermediate_size = intermediate_size | |
| self.max_position_embeddings = max_position_embeddings | |
| self.rope_theta = rope_theta | |
| super().__init__(tie_word_embeddings=tie_word_embeddings, **kwargs) | |