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, 15:15 UTC
[ROCm][CI] Fix plugin tests (2 GPUs) failures on ROCm and removing `VLLM_FLOAT32_MATMUL_PRECISION` from all ROCm tests (#31829)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Jan 6, 2026 at 19:12 UTC
364a8bc6dc7d8fc344f07f01adc6a2336887e9bd
2 files changed
+5
-9
.buildkite/test-amd.yaml
+3
-9
@@ -163,9 +163,7 @@ steps:
163
commands:
164
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
165
- pytest -v -s entrypoints/openai --ignore=entrypoints/openai/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/test_oot_registration.py --ignore=entrypoints/openai/test_tensorizer_entrypoint.py --ignore=entrypoints/openai/correctness/ --ignore=entrypoints/openai/tool_parsers/ --ignore=entrypoints/openai/test_vision_embeds.py
166
- # Need tf32 to avoid conflicting precision issue with terratorch on ROCm.
167
- # TODO: Remove after next torch update
168
- - VLLM_FLOAT32_MATMUL_PRECISION="tf32" pytest -v -s entrypoints/openai/test_vision_embeds.py
166
+ - pytest -v -s entrypoints/openai/test_vision_embeds.py
167
- pytest -v -s entrypoints/test_chat_utils.py
168
169
- label: Entrypoints Integration Test (API Server 2)
@@ -989,9 +987,7 @@ steps:
987
- pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
988
- pip freeze | grep -E 'torch'
989
- pytest -v -s models/multimodal -m core_model --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/processing --ignore models/multimodal/pooling/test_prithvi_mae.py
992
- # Need tf32 to avoid conflicting precision issue with terratorch on ROCm.
993
- # TODO: Remove after next torch update
994
- - VLLM_FLOAT32_MATMUL_PRECISION="tf32" pytest -v -s models/multimodal/pooling/test_prithvi_mae.py -m core_model
990
+ - pytest -v -s models/multimodal/pooling/test_prithvi_mae.py -m core_model
991
- cd .. && VLLM_WORKER_MULTIPROC_METHOD=spawn pytest -v -s tests/models/multimodal/generation/test_whisper.py -m core_model # Otherwise, mp_method="spawn" doesn't work
992
993
- label: Multi-Modal Accuracy Eval (Small Models) # 5min
@@ -1356,9 +1352,7 @@ steps:
1352
# end platform plugin tests
1353
# begin io_processor plugins test, all the code in between uses the prithvi_io_processor plugin
1354
- pip install -e ./plugins/prithvi_io_processor_plugin
1359
- # Need tf32 to avoid conflicting precision issue with terratorch on ROCm.
1360
- # TODO: Remove after next torch update
1361
- - VLLM_FLOAT32_MATMUL_PRECISION="tf32" pytest -v -s plugins_tests/test_io_processor_plugins.py
1355
+ - pytest -v -s plugins_tests/test_io_processor_plugins.py
1356
- pip uninstall prithvi_io_processor_plugin -y
1357
# end io_processor plugins test
1358
# begin stat_logger plugins test
requirements/rocm-test.txt
+2
@@ -88,3 +88,5 @@ open-clip-torch==2.32.0
88
perceptron==0.1.4
89
# Required for the multi-modal models test
90
timm==1.0.17
91
+# Required for plugins test
92
+albumentations==1.4.6