For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

[fhaji75]

I want to train on a custom dataset, but I got multiple errors I tried to solve that, one error is
For debugging, consider passing CUDA_LAUNCH_BLOCKING=1.
I couldn't solve this, I ran the repository in Colab.