training, the gradient disappears

Open mrswang1 opened this issue 2 years ago • 2 comments

When I finetune the model for my own dataset, the model has not yet started training, and the output of stage 3 and stage 4 of the model is NaN.

How can I solve this problem?

May 22 '23 11:05 mrswang1

@mrswang1 Do you have the training code for efficientvit that may be you have written.I am a UG student and I wanted to study it.If possible please provide me with it.Your help would really help me a lot for my research.

May 27 '23 17:05 arnvsnigi

Hi mrswang1,

The training guide for ImageNet classification is available at https://github.com/mit-han-lab/efficientvit/blob/master/TRAINING.md.

Does the issue still exist? If so, can you give me an example for debugging?

Thanks, Han

Jul 21 '23 15:07 han-cai