| from transformers import AutoConfig, PretrainedConfig | |
| class FloSmolVConfig(PretrainedConfig): | |
| model_type = "flosmolV" | |
| def __init__( | |
| self, | |
| vision_config=None, | |
| llm_config=None, | |
| **kwargs, | |
| ): | |
| super().__init__(**kwargs) | |
| if vision_config is None: | |
| vision_config = AutoConfig.from_pretrained("microsoft/Florence-2-base", trust_remote_code=True) | |
| if llm_config is None: | |
| llm_config = AutoConfig.from_pretrained("HuggingFaceTB/SmolLM-360M-Instruct", trust_remote_code=True) | |
| self.vision_config = vision_config | |
| self.llm_config = llm_config |