| 1 | group: Benchmarks |
| 2 | depends_on: |
| 3 | - image-build |
| 4 | steps: |
| 5 | - label: ":nvidia: (H200 MIG 18GB) Benchmarks CLI" |
| 6 | key: benchmarks-cli-test |
| 7 | timeout_in_minutes: 45 |
| 8 | device: h200_18gb |
| 9 | source_file_dependencies: |
| 10 | - vllm/ |
| 11 | - "!vllm/distributed/kv_transfer/" |
| 12 | - rust/src/bench/tests/python_serve_flags.txt |
| 13 | - tests/benchmarks/ |
| 14 | commands: |
| 15 | - pytest -v -s benchmarks/ |
| 16 | mirror: |
| 17 | amd: |
| 18 | label: ":amd: (MI355 DPX) Benchmarks CLI" |
| 19 | dind: false |
| 20 | device: mi355_dpx |
| 21 | timeout_in_minutes: 40 |
| 22 | depends_on: |
| 23 | - image-build-amd |
| 24 | |
| 25 | - label: ":nvidia: (B200) Attention Benchmark Smoke" |
| 26 | key: attention-benchmarks-smoke-test-b200 |
| 27 | device: b200-k8s |
| 28 | num_gpus: 2 |
| 29 | optional: true |
| 30 | working_dir: "/vllm-workspace/" |
| 31 | timeout_in_minutes: 20 |
| 32 | source_file_dependencies: |
| 33 | - benchmarks/attention_benchmarks/ |
| 34 | - vllm/v1/attention/ |
| 35 | commands: |
| 36 | - python3 benchmarks/attention_benchmarks/benchmark.py --backends flash flashinfer --batch-specs "8q1s1k" |
| 37 | mirror: |
| 38 | amd: |
| 39 | label: ":amd: (MI355) Attention Benchmark Smoke" |
| 40 | dind: false |
| 41 | device: mi355_2 |
| 42 | timeout_in_minutes: 25 |
| 43 | depends_on: |
| 44 | - image-build-amd |
| 45 | source_file_dependencies: |
| 46 | - benchmarks/attention_benchmarks/ |
| 47 | - vllm/v1/attention/ |
| 48 | - vllm/_aiter_ops.py |
| 49 | - vllm/platforms/rocm.py |
| 50 | commands: |
| 51 | - python3 benchmarks/attention_benchmarks/benchmark.py --backends ROCM_ATTN ROCM_AITER_FA ROCM_AITER_UNIFIED_ATTN --batch-specs "8q1s1k" |