A problem about Fine-tuning based on HuatuoGPT-Vision-7B-Qwen2.5VL
#1
by
Hunan-Tiger
- opened
Hi,
Thanks your the great work!
I want to training HuatuoGPT-Vision-7B-Qwen2.5VL on my dataset, but i found the loss has always been 0, which means the model can not be train. I use the training Code of offical QWen2.5VL.
Do you have any ideas?
Thank you again!
The model might have already seen the data in its training. Are you sure it's not a public dataset?
The model might have already seen the data in its training. Are you sure it's not a public dataset?
No, it is based on MedMNIST (Public). I design text(category)-image pairs to training large model. I do not sure HuatuoGPT-Vision-7B-Qwen2.5VL has seen the dataset.
It is worked suddenly! I dont konw what happened, but thank you very much.