File tree Expand file tree Collapse file tree 2 files changed +21
-2
lines changed
.buildkite/scripts/scheduled_integration_test Expand file tree Collapse file tree 2 files changed +21
-2
lines changed Original file line number Diff line number Diff line change @@ -17,7 +17,17 @@ wait_for_server() {
1717}
1818
1919MODEL=" deepseek-ai/DeepSeek-V2-lite"
20- BACKENDS=(" deepep_high_throughput" " deepep_low_latency" )
20+
21+ # Set BACKENDS based on platform
22+ if command -v rocm-smi & > /dev/null || [[ -d /opt/rocm ]] || [[ -n " ${ROCM_PATH:- } " ]]; then
23+ # ROCm platform
24+ BACKENDS=(" allgather_reducescatter" )
25+ # Disable MOE padding for ROCm since it is causing eplb to fail
26+ export VLLM_ROCM_MOE_PADDING=0
27+ else
28+ # Non-ROCm platform (CUDA/other)
29+ BACKENDS=(" deepep_high_throughput" " deepep_low_latency" )
30+ fi
2131
2232cleanup () {
2333 if [[ -n " ${SERVER_PID:- } " ]] && kill -0 " ${SERVER_PID} " 2> /dev/null; then
Original file line number Diff line number Diff line change @@ -17,7 +17,16 @@ wait_for_server() {
1717}
1818
1919MODEL=" QWen/Qwen3-30B-A3B-FP8"
20- BACKENDS=(" deepep_high_throughput" " deepep_low_latency" )
20+ # Set BACKENDS based on platform
21+ if command -v rocm-smi & > /dev/null || [[ -d /opt/rocm ]] || [[ -n " ${ROCM_PATH:- } " ]]; then
22+ # ROCm platform
23+ BACKENDS=(" allgather_reducescatter" )
24+ # Disable MOE padding for ROCm since it is causing eplb to fail
25+ export VLLM_ROCM_MOE_PADDING=0
26+ else
27+ # Non-ROCm platform (CUDA/other)
28+ BACKENDS=(" deepep_high_throughput" " deepep_low_latency" )
29+ fi
2130
2231cleanup () {
2332 if [[ -n " ${SERVER_PID:- } " ]] && kill -0 " ${SERVER_PID} " 2> /dev/null; then
You can’t perform that action at this time.
0 commit comments