search

yuanzhoulvpi2017 / zero_nlp

中文nlp解决方案(大模型、数据、模型、训练、推理)
MIT License
2.93k stars 360 forks source link

train_llava保存processor的时候出现错误 #184

Open 1ittlesnow opened 1 month ago

1ittlesnow commented 1 month ago

qwen_model_name_or_path = "/data/workspace/huggingface/Qwen2-1.5B-Instruct" autoprocessor = AutoProcessor.from_pretrained(clip_model_name_or_path) autoprocessor.save_pretrained("huggingface/model003")

通过这个保存了processor之后

autoprocessor1 = AutoProcessor.from_pretrained("huggingface/model001") 只能通过AutoProcessor加载，不能通过llavaProcessor加载