| 1 | group: Engine |
| 2 | depends_on: |
| 3 | - image-build |
| 4 | steps: |
| 5 | - label: ":nvidia: (H200 MIG 18GB) Engine Core" |
| 6 | key: engine |
| 7 | timeout_in_minutes: 30 |
| 8 | device: h200_18gb |
| 9 | source_file_dependencies: |
| 10 | - vllm/compilation/ |
| 11 | - vllm/config/ |
| 12 | - vllm/engine/ |
| 13 | - vllm/entrypoints/logger.py |
| 14 | - vllm/envs.py |
| 15 | - vllm/logger.py |
| 16 | - vllm/logging_utils/ |
| 17 | - vllm/platforms/ |
| 18 | - vllm/sequence.py |
| 19 | - vllm/triton_utils/ |
| 20 | - vllm/utils/ |
| 21 | - tests/engine |
| 22 | - tests/test_sequence |
| 23 | - tests/test_config |
| 24 | - tests/test_logger |
| 25 | - tests/test_vllm_port |
| 26 | - tests/jit_monitor/test_hooks.py |
| 27 | - tests/jit_monitor/test_hooks_gpu.py |
| 28 | commands: |
| 29 | - pytest -v -s engine test_sequence.py test_config.py test_logger.py test_vllm_port.py jit_monitor/test_hooks.py jit_monitor/test_hooks_gpu.py |
| 30 | mirror: |
| 31 | amd: |
| 32 | label: ":amd: (MI355 DPX) Engine Core" |
| 33 | dind: false |
| 34 | device: mi355_dpx |
| 35 | timeout_in_minutes: 40 |
| 36 | depends_on: |
| 37 | - image-build-amd |
| 38 | |
| 39 | - label: ":nvidia: (L4) V1 Engine" |
| 40 | key: engine-1-gpu |
| 41 | device: l4 |
| 42 | num_devices: 1 |
| 43 | timeout_in_minutes: 45 |
| 44 | source_file_dependencies: |
| 45 | - vllm/v1/engine/ |
| 46 | - tests/v1/engine/ |
| 47 | - tests/v1/test_tensor_ipc_queue.py |
| 48 | commands: |
| 49 | - pytest -v -s v1/engine/test_preprocess_error_handling.py |
| 50 | - pytest -v -s v1/engine --ignore v1/engine/test_preprocess_error_handling.py |
| 51 | - pytest -v -s v1/test_tensor_ipc_queue.py |
| 52 | mirror: |
| 53 | amd: |
| 54 | label: ":amd: (MI250) V1 Engine" |
| 55 | dind: false |
| 56 | device: mi250_1 |
| 57 | timeout_in_minutes: 45 |
| 58 | depends_on: |
| 59 | - image-build-amd |
| 60 | source_file_dependencies: |
| 61 | - vllm/platforms/rocm.py |
| 62 | commands: |
| 63 | - pytest -v -s v1/engine/test_preprocess_error_handling.py |
| 64 | - pytest -v -s v1/engine --ignore v1/engine/test_preprocess_error_handling.py |
| 65 | - pytest -v -s v1/test_tensor_ipc_queue.py |
| 66 | - pytest -v -s config/test_mp_reducer.py |
| 67 | |
| 68 | - label: ":nvidia: (H200 MIG 18GB) E2E Scheduling" |
| 69 | key: e2e-scheduling-1-gpu |
| 70 | timeout_in_minutes: 53 |
| 71 | device: h200_18gb |
| 72 | source_file_dependencies: |
| 73 | - vllm/v1/ |
| 74 | - tests/v1/e2e/general/ |
| 75 | commands: |
| 76 | - pytest -v -s v1/e2e/general/test_async_scheduling.py |
| 77 | mirror: |
| 78 | amd: |
| 79 | label: ":amd: (MI250) E2E Scheduling" |
| 80 | dind: false |
| 81 | device: mi250_1 |
| 82 | timeout_in_minutes: 55 |
| 83 | depends_on: |
| 84 | - image-build-amd |
| 85 | source_file_dependencies: |
| 86 | - vllm/platforms/rocm.py |
| 87 | |
| 88 | - label: ":nvidia: (H200 MIG 18GB) E2E Scheduling Accuracy" |
| 89 | key: e2e-scheduling-accuracy-1-gpu |
| 90 | timeout_in_minutes: 53 |
| 91 | device: h200_18gb |
| 92 | source_file_dependencies: |
| 93 | - vllm/v1/ |
| 94 | - tests/v1/e2e/general/ |
| 95 | commands: |
| 96 | - pytest -v -s v1/e2e/general/test_async_scheduling_accuracy.py |
| 97 | mirror: |
| 98 | amd: |
| 99 | label: ":amd: (MI250) E2E Scheduling Accuracy" |
| 100 | dind: false |
| 101 | device: mi250_1 |
| 102 | timeout_in_minutes: 55 |
| 103 | depends_on: |
| 104 | - image-build-amd |
| 105 | source_file_dependencies: |
| 106 | - vllm/platforms/rocm.py |
| 107 | |
| 108 | - label: ":nvidia: (H200 MIG 18GB) E2E Core" |
| 109 | device: h200_18gb |
| 110 | key: e2e-core-1-gpu |
| 111 | timeout_in_minutes: 65 |
| 112 | source_file_dependencies: |
| 113 | - vllm/v1/ |
| 114 | - tests/v1/e2e/general/ |
| 115 | commands: |
| 116 | - pytest -v -s v1/e2e/general --ignore v1/e2e/general/test_async_scheduling.py --ignore v1/e2e/general/test_async_scheduling_accuracy.py --ignore v1/e2e/general/test_kv_sharing_fast_prefill.py -k "not test_mamba_prefix_cache_mrv2" |
| 117 | mirror: |
| 118 | amd: |
| 119 | label: ":amd: (MI250) E2E Core" |
| 120 | dind: false |
| 121 | commands: |
| 122 | - pytest -v -s v1/e2e/general --ignore v1/e2e/general/test_async_scheduling.py --ignore v1/e2e/general/test_async_scheduling_accuracy.py --ignore v1/e2e/general/test_kv_sharing_fast_prefill.py -k "not test_mamba_prefix_cache_mrv2" |
| 123 | device: mi250_1 |
| 124 | timeout_in_minutes: 50 |
| 125 | depends_on: |
| 126 | - image-build-amd |
| 127 | source_file_dependencies: |
| 128 | - vllm/v1/ |
| 129 | - tests/v1/e2e/general/ |
| 130 | - vllm/platforms/rocm.py |
| 131 | |
| 132 | # Keep the large Gemma-3n and Mamba MRV2 warmup allocations on 35GB slices. |
| 133 | - label: ":nvidia: (H200 MIG 35GB) E2E Core Large Memory" |
| 134 | device: h200_35gb |
| 135 | key: e2e-core-large-memory |
| 136 | timeout_in_minutes: 50 |
| 137 | source_file_dependencies: |
| 138 | - vllm/v1/ |
| 139 | - tests/v1/e2e/general/ |
| 140 | commands: |
| 141 | - pytest -v -s v1/e2e/general/test_kv_sharing_fast_prefill.py |
| 142 | - pytest -v -s v1/e2e/general/test_mamba_prefix_cache.py::test_mamba_prefix_cache_mrv2 v1/e2e/general/test_mamba_prefix_cache.py::test_mamba_prefix_cache_mrv2_async |
| 143 | mirror: |
| 144 | amd: |
| 145 | label: ":amd: (MI355 DPX) E2E Core Large Memory" |
| 146 | dind: false |
| 147 | device: mi355_dpx |
| 148 | num_devices: 1 |
| 149 | timeout_in_minutes: 65 |
| 150 | working_dir: /vllm-workspace/tests |
| 151 | depends_on: |
| 152 | - image-build-amd |
| 153 | source_file_dependencies: |
| 154 | - vllm/v1/ |
| 155 | - tests/v1/e2e/general/ |
| 156 | - vllm/platforms/rocm.py |
| 157 | commands: |
| 158 | - pytest -v -s v1/e2e/general/test_kv_sharing_fast_prefill.py |
| 159 | - pytest -v -s v1/e2e/general/test_mamba_prefix_cache.py::test_mamba_prefix_cache_mrv2 v1/e2e/general/test_mamba_prefix_cache.py::test_mamba_prefix_cache_mrv2_async |
| 160 | |
| 161 | - label: ":nvidia: (H100) ReplaySSM E2E" |
| 162 | device: h100 |
| 163 | num_devices: 2 |
| 164 | key: replayssm-e2e |
| 165 | timeout_in_minutes: 90 |
| 166 | working_dir: "/vllm-workspace/tests" |
| 167 | source_file_dependencies: |
| 168 | - tests/v1/e2e/test_replayssm_decode.py |
| 169 | - vllm/config/cache.py |
| 170 | - vllm/config/vllm.py |
| 171 | - vllm/model_executor/layers/mamba/ |
| 172 | - vllm/model_executor/models/nemotron_h.py |
| 173 | - vllm/model_executor/warmup/replayssm_warmup.py |
| 174 | - vllm/v1/attention/backends/mamba_attn.py |
| 175 | - vllm/v1/spec_decode/ |
| 176 | - vllm/v1/worker/gpu/ |
| 177 | - vllm/v1/worker/gpu_model_runner.py |
| 178 | - vllm/v1/worker/gpu_worker.py |
| 179 | commands: |
| 180 | - pytest -v -s v1/e2e/test_replayssm_decode.py |
| 181 | |
| 182 | - label: ":nvidia: (L4) V1 E2E" |
| 183 | key: v1-e2e-2-gpus |
| 184 | timeout_in_minutes: 25 # TODO: Fix timeout after we have more confidence in the test stability |
| 185 | optional: true |
| 186 | device: l4 |
| 187 | num_devices: 2 |
| 188 | source_file_dependencies: |
| 189 | - vllm/compilation/ |
| 190 | - vllm/config/ |
| 191 | - vllm/distributed/ |
| 192 | - "!vllm/distributed/kv_transfer/" |
| 193 | - vllm/engine/ |
| 194 | - vllm/envs.py |
| 195 | - vllm/forward_context.py |
| 196 | - vllm/inputs/ |
| 197 | - vllm/logger.py |
| 198 | - vllm/logging_utils/ |
| 199 | - vllm/model_executor/ |
| 200 | - vllm/multimodal/ |
| 201 | - vllm/platforms/ |
| 202 | - vllm/sampling_params.py |
| 203 | - vllm/transformers_utils/ |
| 204 | - vllm/triton_utils/ |
| 205 | - vllm/utils/ |
| 206 | - vllm/v1/ |
| 207 | - tests/v1/e2e/spec_decode/ |
| 208 | commands: |
| 209 | # Only run tests that need exactly 2 GPUs |
| 210 | - >- |
| 211 | pytest -v -s |
| 212 | v1/e2e/spec_decode/draft_model/test_draft_model.py::test_draft_model_tensor_parallelism |
| 213 | v1/e2e/spec_decode/draft_model/test_draft_model.py::test_draft_model_engine_args_tensor_parallelism |
| 214 | mirror: |
| 215 | amd: |
| 216 | label: ":amd: (MI300) V1 E2E" |
| 217 | dind: false |
| 218 | device: mi300_2 |
| 219 | timeout_in_minutes: 30 |
| 220 | depends_on: |
| 221 | - image-build-amd |
| 222 | |
| 223 | - label: ":nvidia: (B200) V1 E2E Hybrid Chunked Prefill" |
| 224 | key: v1-e2e-hybrid-chunked-prefill-b200 |
| 225 | timeout_in_minutes: 35 |
| 226 | device: b200-k8s |
| 227 | num_devices: 1 |
| 228 | optional: true |
| 229 | source_file_dependencies: |
| 230 | - vllm/v1/attention/backends/utils.py |
| 231 | - vllm/v1/core/ |
| 232 | - vllm/v1/worker/gpu_model_runner.py |
| 233 | - tests/v1/e2e/test_hybrid_chunked_prefill.py |
| 234 | commands: |
| 235 | - pytest -v -s v1/e2e/test_hybrid_chunked_prefill.py |
| 236 | mirror: |
| 237 | amd: |
| 238 | label: ":amd: (MI300) V1 E2E Hybrid Chunked Prefill" |
| 239 | dind: false |
| 240 | device: mi300_1 |
| 241 | timeout_in_minutes: 35 |
| 242 | depends_on: |
| 243 | - image-build-amd |