Closed HymEric closed 4 years ago
How to run with a single GPU? Or two GPUs in one server?
For one GPU, you may need to: (1) Split over the bsz dimension to mimic the effects of Shuffle BN. (2) Adjust the learning rate, e.g., linearly scale lr.
bsz
lr
For two GPU, you can just do (2).
How to run with a single GPU? Or two GPUs in one server?