What lr should I choose for imagenet training if I use 4 GPUs instead of 4 with batch size of 32 per gpu?
What lr should I choose for imagenet training if I use 4 GPUs instead of 4 with batch size of 32 per gpu?