Pytorch CuDNN error

Hi All,

Anyone know what this error is all about?

Traceback (most recent call last):
  File "train.py", line 135, in <module>
    loss.backward()
  File "/home/ubuntu/anaconda3/lib/python3.6/site-packages/torch/tensor.py", line 107, in backward
    torch.autograd.backward(self, gradient, retain_graph, create_graph)
  File "/home/ubuntu/anaconda3/lib/python3.6/site-packages/torch/autograd/__init__.py", line 93, in backward
    allow_unreachable=True)  # allow_unreachable flag
RuntimeError: cuDNN error: CUDNN_STATUS_NOT_INITIALIZED

Any and all help is very much appreciated

Thanks

This error might point to a faulty driver installation, a failure in initializing the CUDA context or an unsupported GPU.
Was your setup working before or are you using a new environment/system?
In the former case, could you disable cudnn (torch.backends.cudnn.enabled=False), and run the code again?
Maybe some other error is hidden with this error message.