| from unsloth import FastVisionModel | |
| from dotenv import load_dotenv | |
| import os | |
| def save_model(model, tokenizer, local: bool) -> None: | |
| load_dotenv() | |
| if local: | |
| model.save_pretrained() | |
| tokenizer.save_pretrained("ft_llava") | |
| else: | |
| model.push_to_hub(f"{os.getenv("ORG_NAME")}/ft_llava", token = os.getenv("HF_TOKEN")) | |
| return | |
| def save_gguf(model_name: str, local:bool, tokenizer): | |
| model, processor = FastVisionModel.from_pretrained( | |
| model_name= model_name, | |
| load_in_4bit=True, | |
| ) | |
| FastVisionModel.for_inference(model) | |
| if local: | |
| model.save_pretrained_merged("ft_qwen2_vl_2b", tokenizer) | |
| else: | |
| model.push_to_hub_merged(f"{os.getenv("ORG_NAME")}/ft_qwen2_vl_2b", tokenizer, token = f"{os.getenv("HF_TOKEN")}") | |