wondervictor
commented
6 months ago nnodes 设置为机器的数量， node_rank 设置为每台机机器的rank，详情请参考：https://pytorch.org/tutorials/intermediate/ddp_series_multinode.html
Open zhongzee opened 6 months ago
wondervictor
commented
6 months ago nnodes 设置为机器的数量， node_rank 设置为每台机机器的rank，详情请参考：https://pytorch.org/tutorials/intermediate/ddp_series_multinode.html
chmod +x tools/dist_train.sh
sample command for pre-training, use AMP for mixed-precision training
./tools/dist_train.sh configs/pretrain/yolo_world_l_t2i_bn_2e-4_100e_4x8gpus_obj365v1_goldg_train_lvis_minival.py 8 --amp 这个命令只适合单服务器的，如和改成多服务器的形式呢？nodes和node_rank要怎么设置呢？