| 1 | group: Model Executor |
| 2 | depends_on: |
| 3 | - image-build |
| 4 | steps: |
| 5 | - label: ":nvidia: (H200 MIG 35GB) Model Executor" |
| 6 | device: h200_35gb |
| 7 | key: model-executor |
| 8 | timeout_in_minutes: 60 |
| 9 | source_file_dependencies: |
| 10 | - vllm/engine/arg_utils.py |
| 11 | - vllm/config/model.py |
| 12 | - vllm/model_executor |
| 13 | - vllm/model_executor/warmup |
| 14 | - tests/model_executor |
| 15 | - tests/model_executor/test_jit_warmup.py |
| 16 | - tests/model_executor/test_jit_warmup_cutedsl_launcher.py |
| 17 | - tests/model_executor/test_jit_warmup_triton_launcher.py |
| 18 | - tests/entrypoints/openai/completion/test_tensorizer_entrypoint.py |
| 19 | commands: |
| 20 | - apt-get update && apt-get install -y curl libsodium23 |
| 21 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 22 | # Dump tracebacks of all threads if a test hangs, so a wedged GPU/CUDA |
| 23 | # init surfaces a stack instead of silently stalling. |
| 24 | - export PYTHONFAULTHANDLER=1 |
| 25 | # Per-test watchdog: a single hung test (e.g. stuck during engine/CUDA |
| 26 | # init) fails fast with a traceback instead of running until the global |
| 27 | # build timeout. The `thread` method also handles hangs inside C/CUDA |
| 28 | # calls that the signal method cannot interrupt. |
| 29 | - pytest -v -s model_executor -m '(not slow_test)' --timeout=900 --timeout-method=thread |
| 30 | - pytest -v -s entrypoints/openai/completion/test_tensorizer_entrypoint.py --timeout=900 --timeout-method=thread |
| 31 | mirror: |
| 32 | amd: |
| 33 | label: ":amd: (MI300) Model Executor" |
| 34 | dind: false |
| 35 | device: mi300_1 |
| 36 | timeout_in_minutes: 75 |
| 37 | depends_on: |
| 38 | - image-build-amd |
| 39 | source_file_dependencies: |
| 40 | - vllm/engine/arg_utils.py |
| 41 | - vllm/config/model.py |
| 42 | - vllm/model_executor |
| 43 | - vllm/model_executor/warmup |
| 44 | - tests/model_executor |
| 45 | - tests/model_executor/test_jit_warmup.py |
| 46 | - tests/model_executor/test_jit_warmup_cutedsl_launcher.py |
| 47 | - tests/model_executor/test_jit_warmup_triton_launcher.py |
| 48 | - tests/entrypoints/openai/completion/test_tensorizer_entrypoint.py |
| 49 | - vllm/_aiter_ops.py |
| 50 | - vllm/platforms/rocm.py |
| 51 | commands: |
| 52 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 53 | - export PYTHONFAULTHANDLER=1 |
| 54 | - pytest -v -s model_executor -m '(not slow_test)' --timeout=900 --timeout-method=thread |
| 55 | - pytest -v -s entrypoints/openai/completion/test_tensorizer_entrypoint.py --timeout=900 --timeout-method=thread |