| from transformers import PretrainedConfig | |
| class Qwen3MoEConfig(PretrainedConfig): | |
| model_type = "qwen3moe" | |
| def __init__( | |
| self, | |
| router_model_path=None, | |
| expert_model_paths=None, | |
| labels=None, | |
| torch_dtype="auto", | |
| tokenizer_path=None, | |
| **kwargs, | |
| ): | |
| super().__init__(**kwargs) | |
| self.router_model_path = router_model_path | |
| self.expert_model_paths = expert_model_paths or {} | |
| self.labels = labels or [] | |
| self.torch_dtype = torch_dtype | |
| self.tokenizer_path = tokenizer_path |