|
| 1 | +group: Compile |
| 2 | +depends_on: |
| 3 | + - image-build |
| 4 | +steps: |
| 5 | +- label: Fusion and Compile Tests (B200) |
| 6 | + timeout_in_minutes: 40 |
| 7 | + working_dir: "/vllm-workspace/" |
| 8 | + gpu: b200 |
| 9 | + source_file_dependencies: |
| 10 | + - csrc/quantization/fp4/ |
| 11 | + - vllm/model_executor/layers/quantization/utils/flashinfer_utils.py |
| 12 | + - vllm/v1/attention/backends/flashinfer.py |
| 13 | + - vllm/v1/worker/ |
| 14 | + - vllm/v1/cudagraph_dispatcher.py |
| 15 | + - vllm/compilation/ |
| 16 | + # can affect pattern matching |
| 17 | + - vllm/model_executor/layers/layernorm.py |
| 18 | + - vllm/model_executor/layers/activation.py |
| 19 | + - vllm/model_executor/layers/quantization/input_quant_fp8.py |
| 20 | + - tests/compile/test_fusion_attn.py |
| 21 | + - tests/compile/test_silu_mul_quant_fusion.py |
| 22 | + - tests/compile/distributed/test_fusion_all_reduce.py |
| 23 | + - tests/compile/distributed/test_fusions_e2e.py |
| 24 | + - tests/compile/fullgraph/test_full_graph.py |
| 25 | + commands: |
| 26 | + - nvidia-smi |
| 27 | + - pytest -v -s tests/compile/test_fusion_attn.py |
| 28 | + - pytest -v -s tests/compile/test_silu_mul_quant_fusion.py |
| 29 | + # this runner has 2 GPUs available even though num_gpus=2 is not set |
| 30 | + - pytest -v -s tests/compile/distributed/test_fusion_all_reduce.py |
| 31 | + # Limit to Inductor partition, no custom ops, and allreduce & attn fusion to reduce running time |
| 32 | + # Wrap with quotes to escape yaml |
| 33 | + - "pytest -v -s tests/compile/distributed/test_fusions_e2e.py::test_tp2_attn_quant_allreduce_rmsnorm -k 'True and not +quant_fp8 and not +rms_norm'" |
| 34 | + # test_fp8_kv_scale_compile requires FlashAttention (not supported on default L4/L40) |
| 35 | + - pytest -v -s tests/compile/fullgraph/test_full_graph.py::test_fp8_kv_scale_compile |
| 36 | + |
| 37 | +- label: Fusion E2E (2 GPUs)(B200) |
| 38 | + timeout_in_minutes: 40 |
| 39 | + working_dir: "/vllm-workspace/" |
| 40 | + gpu: b200 |
| 41 | + optional: true |
| 42 | + num_gpus: 2 |
| 43 | + source_file_dependencies: |
| 44 | + - csrc/quantization/fp4/ |
| 45 | + - vllm/model_executor/layers/quantization/utils/flashinfer_utils.py |
| 46 | + - vllm/v1/attention/backends/flashinfer.py |
| 47 | + - vllm/compilation/ |
| 48 | + # can affect pattern matching |
| 49 | + - vllm/model_executor/layers/layernorm.py |
| 50 | + - vllm/model_executor/layers/activation.py |
| 51 | + - vllm/model_executor/layers/quantization/input_quant_fp8.py |
| 52 | + - tests/compile/distributed/test_fusions_e2e.py |
| 53 | + commands: |
| 54 | + - nvidia-smi |
| 55 | + # Run all e2e fusion tests |
| 56 | + - pytest -v -s tests/compile/distributed/test_fusions_e2e.py |
| 57 | + |
0 commit comments