Different machines with different results?

vision
1

HI Im currently do some research by using multi-gpu.
My concern is, if I use different machines even though I set torch.manual_seed(seed), will the results be different?

cf)all the other settings(i.e torch version, machine kinds etc…)

2

I think using torch.manual_seed is necessary to ensure that the results are the same, but not necessarily sufficient. For example, you might need to also run in deterministic mode (see this page for more details: Reproducibility — PyTorch 1.9.1 documentation).

Additionally, mixing other random number generators (e.g., numpy) might be an issue if they do not use the manual seed set with torch.manual_seed.

3

Thanks for ur reply.
Actually it was sufficient for me to use only torch.manual_seed in my machine, but Im not sure if my model gets different results when I use other machines .