search

DUOMO / TransGPT

MIT License
714 stars 77 forks source link

How to load the TransGPT-MM-v0 model? #12

Open qingchunlizhi opened 1 year ago

qingchunlizhi commented 1 year ago

I have found that multimodal models cannot be used directly. Can you provide more detailed explanations to reproduce the performence of multimodality

image
threeColorFr commented 1 year ago

请说出您的具体问题，具体推理参考https://github.com/DUOMO/TransGPT/tree/main/multi_modal

threeColorFr commented 1 year ago

比如终端形式，你可以传参数你的模型路径--from_pretrained your-model-path，比如你的图中就是TransGPT-MM-v0或者TransGPT-MM-v1 image

kangxiao18 commented 10 months ago

tokenizer = AutoTokenizer.from_pretrained("THUDM/chatglm-6b", trust_remote_code=True) 到这一步时会出错，我将这里改为DUOMO-Lab/TransGPT-MM-v1，也就是在huggingface里下载的TransGPT-MM-v1文件夹，则会报出config.json找不到的问题

corkiyao commented 2 months ago

tokenizer = AutoTokenizer.from_pretrained("THUDM/chatglm-6b", trust_remote_code=True) 到这一步时会出错，我将这里改为DUOMO-Lab/TransGPT-MM-v1，也就是在huggingface里下载的TransGPT-MM-v1文件夹，则会报出config.json找不到的问题

你把你的config.json里面的chat模型路径修改成本地就好了。