diff --git a/python/llm/dev/benchmark/all-in-one/run-deepspeed-arc.sh b/python/llm/dev/benchmark/all-in-one/run-deepspeed-arc.sh index 849c4504..42354272 100644 --- a/python/llm/dev/benchmark/all-in-one/run-deepspeed-arc.sh +++ b/python/llm/dev/benchmark/all-in-one/run-deepspeed-arc.sh @@ -10,7 +10,9 @@ source $basekit_root/ccl/latest/env/vars.sh --force NUM_GPUS=2 # number of used GPU export USE_XETLA=OFF -export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=2 +if grep -q "Core" /proc/cpuinfo; then + export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=2 +fi export TORCH_LLM_ALLREDUCE=0 # Different from PVC mpirun -np $NUM_GPUS --prepend-rank python run.py diff --git a/python/llm/example/GPU/Deepspeed-AutoTP/run_vicuna_33b_arc_2_card.sh b/python/llm/example/GPU/Deepspeed-AutoTP/run_vicuna_33b_arc_2_card.sh index 39b060ae..1e23668f 100644 --- a/python/llm/example/GPU/Deepspeed-AutoTP/run_vicuna_33b_arc_2_card.sh +++ b/python/llm/example/GPU/Deepspeed-AutoTP/run_vicuna_33b_arc_2_card.sh @@ -26,7 +26,9 @@ source $basekit_root/ccl/latest/env/vars.sh --force NUM_GPUS=2 # number of used GPU export USE_XETLA=OFF -export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=2 +if grep -q "Core" /proc/cpuinfo; then + export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=2 +fi export TORCH_LLM_ALLREDUCE=0 # Different from PVC mpirun -np $NUM_GPUS --prepend-rank \