biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[ROCm][CI] Set VLLM_FLOAT32_MATMUL_PRECISION="tf32" For terratorch Tests In AMD CI (#31242)

Browse Source 
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
This commit is contained in:
Micah Williamson
2025-12-23 21:21:50 -06:00
committed by GitHub
parent e42894f5b5
commit 3ce791ac77
2 changed files with 12 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
.buildkite/test-amd.yaml
View File

@@ -162,7 +162,10 @@ steps:
- tests/entrypoints/test_chat_utils - tests/entrypoints/test_chat_utils
commands: commands:
- export VLLM_WORKER_MULTIPROC_METHOD=spawn - export VLLM_WORKER_MULTIPROC_METHOD=spawn
- 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/ - 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
# Need tf32 to avoid conflicting precision issue with terratorch on ROCm.
# TODO: Remove after next torch update
- VLLM_FLOAT32_MATMUL_PRECISION="tf32" pytest -v -s entrypoints/openai/test_vision_embeds.py
- pytest -v -s entrypoints/test_chat_utils.py - pytest -v -s entrypoints/test_chat_utils.py
- label: Entrypoints Integration Test (API Server 2) - label: Entrypoints Integration Test (API Server 2)
@@ -979,7 +982,10 @@ steps:
- export MIOPEN_DEBUG_CONV_GEMM=0 - export MIOPEN_DEBUG_CONV_GEMM=0
- pip install git+https://github.com/TIGER-AI-Lab/Mantis.git - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
- pip freeze | grep -E 'torch' - pip freeze | grep -E 'torch'
- pytest -v -s models/multimodal -m core_model --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/processing - 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
# Need tf32 to avoid conflicting precision issue with terratorch on ROCm.
# TODO: Remove after next torch update
- VLLM_FLOAT32_MATMUL_PRECISION="tf32" pytest -v -s models/multimodal/pooling/test_prithvi_mae.py -m core_model
- 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 - 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
- label: Multi-Modal Accuracy Eval (Small Models) # 5min - label: Multi-Modal Accuracy Eval (Small Models) # 5min
@@ -1341,7 +1347,9 @@ steps:
# end platform plugin tests # end platform plugin tests
# begin io_processor plugins test, all the code in between uses the prithvi_io_processor plugin # begin io_processor plugins test, all the code in between uses the prithvi_io_processor plugin
- pip install -e ./plugins/prithvi_io_processor_plugin - pip install -e ./plugins/prithvi_io_processor_plugin
- pytest -v -s plugins_tests/test_io_processor_plugins.py # Need tf32 to avoid conflicting precision issue with terratorch on ROCm.
# TODO: Remove after next torch update
- VLLM_FLOAT32_MATMUL_PRECISION="tf32" pytest -v -s plugins_tests/test_io_processor_plugins.py
- pip uninstall prithvi_io_processor_plugin -y - pip uninstall prithvi_io_processor_plugin -y
# end io_processor plugins test # end io_processor plugins test
# begin stat_logger plugins test # begin stat_logger plugins test

2
tests/models/test_terratorch.py
View File

@@ -38,7 +38,7 @@ def test_inference(
max_num_seqs=32, max_num_seqs=32,
default_torch_num_threads=1, default_torch_num_threads=1,
) as vllm_model: ) as vllm_model:
vllm_output = vllm_model.llm.encode(prompt) vllm_output = vllm_model.llm.encode(prompt, pooling_task="plugin")
assert torch.equal( assert torch.equal(
torch.isnan(vllm_output[0].outputs.data).any(), torch.tensor(False) torch.isnan(vllm_output[0].outputs.data).any(), torch.tensor(False)
) )