?本地要改成这样:
export GPUS_PER_NODE=8
export MASTER_ADDR=localhost #127.0.0.1
export MASTER_PORT=9988
export NNODES=1
export NODE_RANK=0
export WORLD_SIZE=8
DISTRIBUTED_ARGS="
--use-dummy-data \
--nproc_per_node $GPUS_PER_NODE \
--nnodes $NNODES \
--node_rank $NODE_RANK \
--master_addr $MASTER_ADDR \
--master_port $MASTER_PORT
"