| from transformers import PretrainedConfig | |
| class BinaryLLMConfig(PretrainedConfig): | |
| model_type = "binaryllm" | |
| def __init__( | |
| self, | |
| vocab_size: int = 65538, | |
| hidden_size: int = 512, | |
| num_hidden_layers: int = 4, | |
| num_attention_heads: int = 4, | |
| intermediate_size: int = 2048, | |
| max_position_embeddings: int = 2048, | |
| dropout: float = 0.1, | |
| activation: str = "gelu", | |
| bos_token_id: int = 65536, | |
| eos_token_id: int = 65537, | |
| pad_token_id: int = 65537, | |
| **kwargs, | |
| ): | |
| self.vocab_size = int(vocab_size) | |
| self.hidden_size = int(hidden_size) | |
| self.num_hidden_layers = int(num_hidden_layers) | |
| self.num_attention_heads = int(num_attention_heads) | |
| self.intermediate_size = int(intermediate_size) | |
| self.max_position_embeddings = int(max_position_embeddings) | |
| self.dropout = float(dropout) | |
| self.activation = str(activation) | |
| self.bos_token_id = int(bos_token_id) | |
| self.eos_token_id = int(eos_token_id) | |
| self.pad_token_id = int(pad_token_id) | |
| super().__init__(**kwargs) | |