四gpu训练
Closed this issue · 1 comments
longmalongma commented
如果用4个gpu训练会不会快一点？4个gpu训练的话需要更改哪些训练代码？
CUDA_VISIBLE_DEVICES=0,1,2,3 OMP_NUM_THREADS=4 python -m torch.distributed.launch --master_port 1236 --nproc_per_node=2 train.py --id 4gputrian --stage 2
hkchengrex commented