cuDNN error: CUDNN_STATUS_EXECUTION_FAILED
skye95git opened this issue · comments
When I run repr_code.py, there is an error in line 44 reprs = model.code_encoding(*batch_gpu).data.cpu().numpy(): RuntimeError: cuDNN error: CUDNN_STATUS_EXECUTION_FAILED.
My environment is as follows:
cudatoolkit 10.0.130, cudnn 7.6.5, python 3.6.12, torch 1.2.0
I thought it was out of memory. But when I set memory=128G, This error still exists.
Can you tell me why?
I have solved the problem. My new environment is
Python 3.6.13, torch 1.9.0, cuda 11.1
It works.