| from typing import Optional, List, Any, Dict | |
| from transformers.configuration_utils import PretrainedConfig | |
| class Step1Config(PretrainedConfig): | |
| model_type = "step1" | |
| keys_to_ignore_at_inference = ["past_key_values"] | |
| def __init__( | |
| self, | |
| hidden_size: int = 5120, | |
| intermediate_size: int = 13312, | |
| num_attention_heads: int = 40, | |
| num_attention_groups: int = 8, | |
| num_hidden_layers: int = 48, | |
| max_seq_len: int = 4096, | |
| vocab_size: int = 65536, | |
| rms_norm_eps: float = 1e-5, | |
| bos_token_id: int = 1, | |
| eos_token_id: int = 3, | |
| pad_token_id: int = 0, | |
| **kwargs, | |
| ) -> None: | |
| self.hidden_size = hidden_size | |
| self.intermediate_size = intermediate_size | |
| self.num_attention_heads = num_attention_heads | |
| self.num_attention_groups = num_attention_groups | |
| self.num_hidden_layers = num_hidden_layers | |
| self.max_seq_len = max_seq_len | |
| self.vocab_size = vocab_size | |
| self.rms_norm_eps = rms_norm_eps | |
| super().__init__( | |
| bos_token_id=bos_token_id, | |
| pad_token_id=pad_token_id, | |
| eos_token_id=eos_token_id, | |
| **kwargs | |
| ) | |
| __all__ = ["Step1Config"] | |