gpu memory issue 2

[babbu3682]

I already posted my problem before. but I still have a problem with the cuda memory issue.

I tried to run your model called "CRNN" that is training CNN and RNN from scratch.

I didn't any torch your code and just running. In the first epoch there is no error.

But after a few epochs, the Cuda memory issue also shows up.

How can you run the CRNN from scratch no using torch.no_grad?

Every epoch increase, the memory increase until CUDA out of error .