I want to reproduce my experiment completely. Previously I used a server with RTX 3090, but now it has been moved now, so I bought a new one. And I configure the pytorch environment as before, but I can’t reproduce my experiment (The accuracy is different from before). I use this function to fix the seed as the picture shows. What else can I do to reproduce my experiment? Thanks!
From Reproducibility — PyTorch 1.12 documentation guide,
Completely reproducible results are not guaranteed across PyTorch releases, individual commits, or different platforms. Furthermore, results may not be reproducible between CPU and GPU executions, even when using identical seeds.
So, you may be unable to reproduce the exact training sequence/accuracy with a different CPU/GPU.
What’s the deviation in accuracy that you observe?