| 1 | group: JIT Monitor |
| 2 | depends_on: |
| 3 | - image-build |
| 4 | steps: |
| 5 | - label: ":nvidia: (H200 MIG 35GB) No Runtime JITs E2E" |
| 6 | key: jit-monitor-no-runtime-jit |
| 7 | device: h200_35gb |
| 8 | timeout_in_minutes: 45 |
| 9 | source_file_dependencies: |
| 10 | - vllm/utils/jit_monitor.py |
| 11 | - vllm/v1/worker/gpu_worker.py |
| 12 | - vllm/model_executor/warmup/ |
| 13 | - vllm/config/observability.py |
| 14 | - tests/jit_monitor/test_no_runtime_jit.py |
| 15 | - tests/models/registry.py |
| 16 | commands: |
| 17 | # Boot a curated JIT-heavy model set with the JIT monitor in "error" mode |
| 18 | # and run generation; any post-warmup JIT compilation fails the test. |
| 19 | # Per-test watchdog so a wedged engine/CUDA init fails with a traceback |
| 20 | # instead of running until the build timeout. |
| 21 | - export PYTHONFAULTHANDLER=1 |
| 22 | - pytest -v -s jit_monitor/test_no_runtime_jit.py --timeout=900 --timeout-method=thread |
| 23 | mirror: |
| 24 | amd: |
| 25 | # Focused MRV2 regression; the broader model/backend matrix is pending. |
| 26 | label: ':amd: (MI355 DPX) MRV2 Sampler JIT Warmup' |
| 27 | dind: false |
| 28 | device: mi355_dpx |
| 29 | num_devices: 1 |
| 30 | timeout_in_minutes: 10 |
| 31 | working_dir: /vllm-workspace/ |
| 32 | depends_on: |
| 33 | - image-build-amd |
| 34 | source_file_dependencies: |
| 35 | - tests/jit_monitor/ |
| 36 | - tests/models/utils.py |
| 37 | - tests/utils.py |
| 38 | - vllm/v1/worker/gpu/ |
| 39 | - vllm/v1/sample/ops/topk_topp_sampler.py |
| 40 | - vllm/platforms/rocm.py |
| 41 | - vllm/envs.py |
| 42 | commands: |
| 43 | - pytest -v -s tests/jit_monitor/test_no_runtime_jit_rocm.py --timeout=300 |