When lauched a multi-machine-multi-card training program with the help of horovod module, I encountered the error “RuntimeError: no deterministic convolution algorithms available in CuDNN”. My pytorch version is pytorch 1.4. Can anyone know how to fix this error? Thanks
Could you update PyTorch to the latest stable release and rerun the code, please?
In case you are still hitting the issue, could you then post a minimal, executable code snippet so that we could try to reproduce it?
ok, thanks, I will try it