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 9, 12:30 UTC
[ROCm][CI] Add TORCH_NCCL_BLOCKING_WAIT For Distributed Tests (A100) (#32891)
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
Micah Williamson committed
Jan 27, 2026 at 21:32 UTC
911b51b69f92b7cbccc54453991a64eac72f2366
1 file changed
+3
.buildkite/test-amd.yaml
+3
@@ -1509,6 +1509,9 @@ steps:
1509
source_file_dependencies:
1510
- vllm/
1511
commands:
1512
+ # Work around HIP bug tracked here: https://github.com/ROCm/hip/issues/3876
1513
+ # TODO: Remove when the bug is fixed in a future ROCm release
1514
+ - export TORCH_NCCL_BLOCKING_WAIT=1
1515
# NOTE: don't test llama model here, it seems hf implementation is buggy
1516
# see https://github.com/vllm-project/vllm/pull/5689 for details
1517
- pytest -v -s distributed/test_custom_all_reduce.py