The PyTorch docs say that you need to set torch.backends.cudnn.deterministic = True
for reproducible runs.
What’s the equivalent setting in the C++ front-end?
The PyTorch docs say that you need to set torch.backends.cudnn.deterministic = True
for reproducible runs.
What’s the equivalent setting in the C++ front-end?
Hi Thomas,
This is part of the globalContext
(ATen/Context.h
): Would at::globalContext().setDeterministicCuDNN(true)
help?
Best regards
Thomas