A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 5, 23:30 UTC
[ROCm][CI] Override PYTORCH_ROCM_ARCH with detected GPU arch in test containers (#38165)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Mar 26, 2026 at 13:33 UTC
bff9a1c2666de363d5b34f629e9fa61455a338ca
1 file changed
+2
.buildkite/scripts/hardware_ci/run-amd-test.sh
+2
@@ -496,6 +496,7 @@ if is_multi_node "$commands"; then
496
else
497
echo "--- Single-node job"
498
echo "Render devices: $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES"
499
+
500
docker run \
501
--device /dev/kfd $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES \
502
$RDMA_FLAGS \
@@ -511,6 +512,7 @@ else
512
-v "${HF_CACHE}:${HF_MOUNT}" \
513
-e "HF_HOME=${HF_MOUNT}" \
514
-e "PYTHONPATH=${MYPYTHONPATH}" \
515
+ -e "PYTORCH_ROCM_ARCH=" \
516
--name "${container_name}" \
517
"${image_name}" \
518
/bin/bash -c "${commands}"