Closed WindSearcher closed 11 months ago
看到了,run_ptpcg.sh可以参考train_multi.sh中的分布式训练
`# CUDA_VISIBLE_DEVICES=${GPUS} python -u run_dee_task.py \
CUDA_VISIBLE_DEVICES=${GPUS} python -m torch.distributed.launch --master_port=25662 --nproc_per_node ${NUM_GPUS} run_dee_task.py ` 把原先的注释加上新的这一行即可
最后还要加个 --parallel_decorate
flag
您好,想请问下关于分布式训练ptpcg,如下图,在0和7卡上运行,但实际好像只在0卡上训练