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 10, 20:15 UTC
[ROCm][CI] Fix failure in Language Models Tests (Extra Standard) by reducing agent pool size (#31553)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Jan 1, 2026 at 21:29 UTC
5cc4876630659b88c04c444a279556081a6ee82a
1 file changed
+2
-1
.buildkite/test-amd.yaml
+2
-1
@@ -859,7 +859,7 @@ steps:
859
- label: Language Models Tests (Extra Standard) %N
860
timeout_in_minutes: 45
861
mirror_hardwares: [amdexperimental]
862
- agent_pool: mi325_8
862
+ agent_pool: mi325_2
863
# grade: Blocking
864
torch_nightly: true
865
source_file_dependencies:
@@ -871,6 +871,7 @@ steps:
871
# Shard slow subset of standard language models tests. Only run when model
872
# source is modified, or when specified test files are modified
873
- pip freeze | grep -E 'torch'
874
+ - export TORCH_NCCL_BLOCKING_WAIT=1
875
- pytest -v -s models/language -m 'core_model and slow_test' \
876
--num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT \
877
--shard-id=$$BUILDKITE_PARALLEL_JOB