| 1 | group: Models - Basic |
| 2 | depends_on: |
| 3 | - image-build |
| 4 | steps: |
| 5 | - label: ":nvidia: (H200 MIG 18GB) Basic Models (Initialization)" |
| 6 | key: basic-models-tests-initialization |
| 7 | timeout_in_minutes: 25 |
| 8 | device: h200_18gb |
| 9 | source_file_dependencies: |
| 10 | - vllm/ |
| 11 | - "!vllm/distributed/kv_transfer/" |
| 12 | - tests/models/test_initialization.py |
| 13 | - tests/models/registry.py |
| 14 | commands: |
| 15 | # Run a subset of model initialization tests |
| 16 | - pytest -v -s models/test_initialization.py::test_can_initialize_small_subset |
| 17 | mirror: |
| 18 | amd: |
| 19 | label: ":amd: (MI355 DPX) Basic Models (Initialization)" |
| 20 | dind: false |
| 21 | device: mi355_dpx |
| 22 | timeout_in_minutes: 30 |
| 23 | depends_on: |
| 24 | - image-build-amd |
| 25 | |
| 26 | - label: ":nvidia: (H200 MIG 35GB) Basic Models (Extra Initialization) Shard %N" |
| 27 | device: h200_35gb |
| 28 | key: basic-models-tests-extra-initialization |
| 29 | timeout_in_minutes: 30 |
| 30 | source_file_dependencies: |
| 31 | - vllm/model_executor/models/ |
| 32 | - tests/models/test_initialization.py |
| 33 | - tests/models/registry.py |
| 34 | commands: |
| 35 | # Only when vLLM model source is modified - test initialization of a large |
| 36 | # subset of supported models (the complement of the small subset in the above |
| 37 | # test.) Also run if model initialization test file is modified |
| 38 | - pytest -v -s models/test_initialization.py -k 'not test_can_initialize_small_subset' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB |
| 39 | parallelism: 14 |
| 40 | mirror: |
| 41 | amd: |
| 42 | label: ":amd: (MI355 DPX) Basic Models (Extra Initialization) Shard %N" |
| 43 | dind: false |
| 44 | device: mi355_dpx |
| 45 | timeout_in_minutes: 60 |
| 46 | parallelism: 10 |
| 47 | working_dir: "/vllm-workspace/tests" |
| 48 | depends_on: |
| 49 | - image-build-amd |
| 50 | source_file_dependencies: |
| 51 | - vllm/model_executor/models/ |
| 52 | - vllm/model_executor/layers/ |
| 53 | - tests/models/test_initialization.py |
| 54 | - tests/models/registry.py |
| 55 | - vllm/_aiter_ops.py |
| 56 | - vllm/platforms/rocm.py |
| 57 | |
| 58 | - label: ":nvidia: (H200 MIG 35GB) Basic Models (Other)" |
| 59 | device: h200_35gb |
| 60 | key: basic-models-tests-other |
| 61 | timeout_in_minutes: 38 |
| 62 | source_file_dependencies: |
| 63 | - vllm/ |
| 64 | - "!vllm/distributed/kv_transfer/" |
| 65 | - tests/models/test_terratorch.py |
| 66 | - tests/models/transformers/test_backend.py |
| 67 | - tests/models/test_registry.py |
| 68 | - tests/models/test_deepseek_v4_vl_rocm.py |
| 69 | - tests/models/test_deepseek_v41_decoder_replay_layers.py |
| 70 | - tests/models/test_deepseek_v41_replay_batch.py |
| 71 | commands: |
| 72 | - pytest -v -s models/test_terratorch.py models/transformers/test_backend.py models/test_registry.py models/test_deepseek_v4_vl_rocm.py models/test_deepseek_v41_decoder_replay_layers.py models/test_deepseek_v41_replay_batch.py |
| 73 | mirror: |
| 74 | amd: |
| 75 | label: ":amd: (MI355) Basic Models (Other)" |
| 76 | dind: false |
| 77 | device: mi355_1 |
| 78 | timeout_in_minutes: 65 |
| 79 | depends_on: |
| 80 | - image-build-amd |
| 81 | source_file_dependencies: |
| 82 | - tests/models/test_deepseek_v4_vl_rocm.py |
| 83 | - tests/models/test_hyv4_rocm.py |
| 84 | commands: |
| 85 | - pytest -v -s models/test_terratorch.py models/transformers/test_backend.py models/test_registry.py models/test_deepseek_v4_vl_rocm.py |
| 86 | - VLLM_ROCM_USE_AITER=1 pytest -v -s models/test_hyv4_rocm.py -m 'not distributed' |
| 87 | |
| 88 | - label: ":nvidia: (B200) Inkling" |
| 89 | key: inkling-unit-tests-b200 |
| 90 | timeout_in_minutes: 40 |
| 91 | device: b200-k8s |
| 92 | source_file_dependencies: |
| 93 | - vllm/models/inkling/ |
| 94 | - vllm/cute_utils/ |
| 95 | - cmake/external_projects/tml_fa4.cmake |
| 96 | - tests/models/inkling/ |
| 97 | commands: |
| 98 | # FA4 kernel tests require SM100; the suite skips them elsewhere. |
| 99 | - pytest -v -s models/inkling |
| 100 | mirror: |
| 101 | amd: |
| 102 | label: ":amd: (MI355) Inkling" |
| 103 | dind: false |
| 104 | device: mi355_1 |
| 105 | timeout_in_minutes: 65 |
| 106 | working_dir: "/vllm-workspace/tests" |
| 107 | depends_on: |
| 108 | - image-build-amd |
| 109 | source_file_dependencies: |
| 110 | - vllm/models/inkling/ |
| 111 | - vllm/cute_utils/ |
| 112 | - cmake/external_projects/tml_fa4.cmake |
| 113 | - tests/models/inkling/ |
| 114 | - vllm/platforms/rocm.py |
| 115 | commands: |
| 116 | - pytest -v -s models/inkling/rocm |
| 117 | |
| 118 | - label: ":nvidia: (B200) Kimi K3" |
| 119 | key: kimi-k3-unit-tests-b200 |
| 120 | timeout_in_minutes: 40 |
| 121 | device: b200-k8s |
| 122 | source_file_dependencies: |
| 123 | - vllm/models/kimi_k3/ |
| 124 | - csrc/libtorch_stable/kimi_k3/ |
| 125 | - tests/models/kimi_k3/ |
| 126 | - tests/kernels/attention/test_kimi_k3_mla_fused_epilogue.py |
| 127 | - tests/kernels/test_bf16_skinny_gemm.py |
| 128 | commands: |
| 129 | # The native NVIDIA Kimi K3 kernels require the SM100 family. |
| 130 | - pytest -v -s models/kimi_k3 --ignore=models/kimi_k3/test_prefix_cache.py kernels/attention/test_kimi_k3_mla_fused_epilogue.py kernels/test_bf16_skinny_gemm.py |
| 131 | mirror: |
| 132 | amd: |
| 133 | label: ":amd: (MI355) Kimi K3" |
| 134 | dind: false |
| 135 | device: mi355_1 |
| 136 | num_devices: 1 |
| 137 | timeout_in_minutes: 60 |
| 138 | working_dir: "/vllm-workspace/tests" |
| 139 | depends_on: |
| 140 | - image-build-amd |
| 141 | source_file_dependencies: |
| 142 | - vllm/platforms/rocm.py |
| 143 | commands: |
| 144 | # The ROCm KDA/AttnRes kernels are gated on gfx950, so this has to be MI355. |
| 145 | # The two kernels/ files the NVIDIA job adds are CUDA-only (they import |
| 146 | # cuda.bindings.driver), so they are deliberately left out here. |
| 147 | - pytest -v -s models/kimi_k3 |
| 148 | |
| 149 | - label: ":nvidia: (4xB200) Kimi K3 Prefix Cache Shard %N" |
| 150 | key: kimi-k3-prefix-cache-4xb200 |
| 151 | timeout_in_minutes: 75 |
| 152 | device: b200-k8s |
| 153 | num_devices: 4 |
| 154 | source_file_dependencies: |
| 155 | - vllm/models/kimi_k3/ |
| 156 | - vllm/v1/core/ |
| 157 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 158 | - vllm/distributed/kv_transfer/kv_connector/v1/simple_cpu_offload_connector.py |
| 159 | - vllm/v1/simple_kv_offload/ |
| 160 | - vllm/v1/spec_decode/ |
| 161 | - vllm/model_executor/layers/mamba/ |
| 162 | - tests/models/kimi_k3/test_prefix_cache.py |
| 163 | - .buildkite/scripts/install-kv-connectors.sh |
| 164 | commands: |
| 165 | - bash /vllm-workspace/.buildkite/scripts/install-kv-connectors.sh |
| 166 | - pytest -v -s models/kimi_k3/test_prefix_cache.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 167 | parallelism: 2 |
| 168 | |
| 169 | - label: ":nvidia: (H200) GLM5Next" |
| 170 | key: glm5next-unit-tests |
| 171 | timeout_in_minutes: 20 |
| 172 | device: h200_35gb |
| 173 | source_file_dependencies: |
| 174 | - vllm/models/glm5next/ |
| 175 | - tests/models/glm5next/ |
| 176 | commands: |
| 177 | - pytest -v -s models/glm5next |
| 178 | mirror: |
| 179 | amd: |
| 180 | label: ":amd: (MI355 DPX) GLM5Next" |
| 181 | dind: false |
| 182 | device: mi355_dpx |
| 183 | timeout_in_minutes: 35 |
| 184 | working_dir: "/vllm-workspace/tests" |
| 185 | depends_on: |
| 186 | - image-build-amd |
| 187 | source_file_dependencies: |
| 188 | - vllm/platforms/rocm.py |
| 189 | |
| 190 | - label: ":nvidia: (H200) Qwen4 Exp" |
| 191 | key: qwen4-exp-unit-tests |
| 192 | timeout_in_minutes: 40 |
| 193 | device: h200_35gb |
| 194 | source_file_dependencies: |
| 195 | - vllm/models/qwen4_exp/ |
| 196 | - vllm/transformers_utils/configs/qwen4_exp.py |
| 197 | - vllm/config/compilation.py |
| 198 | - vllm/v1/attention/backends/short_conv_attn.py |
| 199 | - tests/models/qwen4_exp/ |
| 200 | commands: |
| 201 | - pytest -v -s models/qwen4_exp/test_hc_ops.py models/qwen4_exp/test_qsa_prepare.py models/qwen4_exp/test_qsa_reference.py |
| 202 | mirror: |
| 203 | amd: |
| 204 | label: ":amd: (MI355 DPX) Qwen4 Exp" |
| 205 | dind: false |
| 206 | device: mi355_dpx |
| 207 | depends_on: |
| 208 | - image-build-amd |
| 209 | commands: |
| 210 | - pytest -v -s models/qwen4_exp/test_qsa_amd.py |
| 211 | # PLE and QSA import conflicting HC registrations in a shared process. |
| 212 | - pytest -v -s models/qwen4_exp/test_ple.py |
| 213 | |
| 214 | - label: ":computer: (CPU) Qwen4 Exp" |
| 215 | key: qwen4-exp-unit-tests-cpu |
| 216 | depends_on: |
| 217 | - image-build-cpu |
| 218 | timeout_in_minutes: 20 |
| 219 | device: cpu-small |
| 220 | source_file_dependencies: |
| 221 | - vllm/models/qwen4_exp/ |
| 222 | - vllm/transformers_utils/configs/qwen4_exp.py |
| 223 | - vllm/config/compilation.py |
| 224 | - vllm/v1/attention/backends/short_conv_attn.py |
| 225 | - tests/models/qwen4_exp/ |
| 226 | commands: |
| 227 | - pytest -v -s models/qwen4_exp/test_config.py models/qwen4_exp/test_ple.py |
| 228 | |
| 229 | - label: ":computer: (CPU) Basic Models Other" |
| 230 | key: basic-models-test-other-cpu |
| 231 | depends_on: |
| 232 | - image-build-cpu |
| 233 | timeout_in_minutes: 20 |
| 234 | source_file_dependencies: |
| 235 | - vllm/ |
| 236 | - "!vllm/distributed/kv_transfer/" |
| 237 | - tests/models/test_utils.py |
| 238 | - tests/models/test_vision.py |
| 239 | - tests/models/test_adapters.py |
| 240 | - tests/models/test_qwen3_5_mtp_config.py |
| 241 | - tests/models/transformers/fusers/ |
| 242 | - tests/model_executor/layers/test_activation.py |
| 243 | device: cpu-small |
| 244 | commands: |
| 245 | - pytest -v -s models/test_utils.py models/test_vision.py models/test_adapters.py models/test_qwen3_5_mtp_config.py models/transformers/fusers/ model_executor/layers/test_activation.py |