lemniscate.pytorch
lemniscate.pytorch copied to clipboard
training about DDP
In distributed data parallel scenario, how can you guarantee the synchronization of memory bank in each process?