Nguyen Duy Kien

Hi, I haven't faced this problem with the codebase. Can you try to resume the training with your checkpoint and see whether it happens again?

if you use fp16, it can explode but it should not happen when you resume it again

Hi, I will try to have some results with swin-transformer but it will take a bit of time since more experiments require more resources :D.

Hi, you should build with the command `python -m pip install -e ./` if you are inside the repo. If you are outside, then please use `python -m pip install...