修改mpirun启动脚本

run_test.sh

+# 设置默认值
+testmode="internode"
+profiling=""
+
 for para in $*
 do
     if [[ $para == --testmode* ]];then
@@ -15,7 +19,6 @@ LAUNCH_WITH_BINDING=${TEST_DIR}/launch_with_binding.sh # Please adjust the varia
 DTK_ENV="/opt/dtk/env.sh" # where env.sh of dtk
 TEST_ENV=${TEST_DIR}/test_env.sh
 
-
 #######################################################################################
 # Those variables no need to modify
 # HOSTFILE="hostfile_$(basename "$0" | sed -E 's/^run_(.+)\.sh$/\1/')"
@@ -40,7 +43,7 @@ mpirun -np ${GPUS}  --hostfile ${HOSTFILE} \
                     --testmode=${testmode} \
                     --profiling=${profiling}
                     "
-                    #> log-$((${GPUS} / 8))nodes-`date +%F-%H%M`.log 2>&1
+		            # > log-${testmode}-$((${GPUS} / 8))nodes-`date +%F-%H%M`.log 2>&1
 wait
 
 

tests_mpi/test_env.sh

@@ -4,6 +4,7 @@
 # 适用于: mpirun 启动的多节点训练
 # 网络: InfiniBand (SHCA) 或 RoCE
 # =============================================================================
+export PYTHONPATH=$(pwd)
 
 # rocSHMEM
 export ROCSHMEM_GDA_NUM_QPS_DEFAULT_CTX=288
@@ -16,6 +17,3 @@ export ROCSHMEM_TOPO_FILE_FORCE=$(pwd)/tests_mpi/topo.config
 export LD_LIBRARY_PATH=/opt/dtk/dushmem/lib:$LD_LIBRARY_PATH
 export DEEP_EP_DEVICE_TO_HCA_MAPPING=0:mlx5_2:1,1:mlx5_3:1,2:mlx5_4:1,3:mlx5_5:1,4:mlx5_6:1,5:mlx5_7:1,6:mlx5_8:1,7:mlx5_9:1
 export NVSHMEM_SYMMETRIC_SIZE=10737418240
-
-export HIP_VISIBLE_DEVICES=0,1,2,3,4,5,6,7  # 根据硬件拓扑调整
-export PYTHONPATH=$(pwd)