| 1 | # In this file, you can add more tests to run either by adding a new step or |
| 2 | # adding a new command to an existing step. See different options here for examples. |
| 3 | |
| 4 | # This script will be feed into Jinja template in `test-template-amd.j2` at |
| 5 | # https://github.com/vllm-project/buildkite-ci/blob/main/scripts/test-template-amd.j2 |
| 6 | # to generate the final pipeline yaml file. |
| 7 | |
| 8 | # Documentation |
| 9 | # label(str): the name of the test. emojis allowed. |
| 10 | # fast_check(bool): whether to run this on each commit on the fastcheck pipeline. |
| 11 | # fast_check_only(bool): run this test on the fastcheck pipeline only |
| 12 | # optional(bool): never run this test by default (i.e. need to unblock manually) unless it's a scheduled nightly run. |
| 13 | # soft_fail(bool): allow this step to fail without failing the entire pipeline (useful for flaky or experimental tests). |
| 14 | # command(str): the single command to run for tests. incompatible with commands. |
| 15 | # commands(list): the list of commands to run for the test. incompatible with command. |
| 16 | # mirror_hardwares(list): selector tags for AMD pipelines that should include the test. |
| 17 | # dind(bool): when false, run the job directly in the AMD Kubernetes pod. |
| 18 | # When true or omitted, use the legacy Docker-in-Docker path. |
| 19 | # num_gpus(int): override the number of GPUs for the test. defaults to 1 GPU. currently supports 2,4. |
| 20 | # num_nodes(int): whether to simulate multi-node setup by launching multiple containers on one host, |
| 21 | # in this case, commands must be specified. the first command runs on the first host, the second |
| 22 | # command runs on the second host. |
| 23 | # timeout_in_minutes(int): sets a timeout for the step in minutes. if not specified, uses the default timeout. |
| 24 | # parallelism(int): number of parallel jobs to run for this step. enables test sharding using $$BUILDKITE_PARALLEL_JOB |
| 25 | # and $$BUILDKITE_PARALLEL_JOB_COUNT environment variables. |
| 26 | # working_dir(str): specify the place where the command should execute, default to /vllm-workspace/tests |
| 27 | # source_file_dependencies(list): the list of prefixes to opt-in the test for, if empty, the test will always run. |
| 28 | |
| 29 | # When adding a test |
| 30 | # - If the test belongs to an existing group, add it there |
| 31 | # - If the test is short, add to any existing step |
| 32 | # - If the test takes more than 10min, then it is okay to create a new step. |
| 33 | # Note that all steps execute in parallel. |
| 34 | |
| 35 | |
| 36 | ##################################################################################################################################### |
| 37 | # # |
| 38 | # README # |
| 39 | # # |
| 40 | ##################################################################################################################################### |
| 41 | # # |
| 42 | # IMPORTANT: # |
| 43 | # * Currently AMD CI has MI250 agents, MI300 agents, and MI355 agents. All upcoming feature improvements are # |
| 44 | # tracked in: https://github.com/vllm-project/vllm/issues/34994 # |
| 45 | # # |
| 46 | #-----------------------------------------------------------------------------------------------------------------------------------# |
| 47 | # # |
| 48 | # NOTES: # |
| 49 | # * [Pytorch Nightly Dependency Override Check]: if this test fails, it means the nightly torch version is not compatible with # |
| 50 | # some of the dependencies. Please check the error message and add the package to # |
| 51 | # whitelist in `/vllm/tools/pre_commit/generate_nightly_torch_test.py`. # |
| 52 | # * [Entrypoints Integration (LLM)]: # |
| 53 | # - {`pytest -v -s entrypoints/llm/test_generate.py`}: It needs a clean process # |
| 54 | # - {`pytest -v -s entrypoints/offline_mode`}: Needs to avoid interference with other tests # |
| 55 | # * [Engine / Engine (1 GPU) / e2e Scheduling / e2e Core / V1 e2e / Spec Decode / V1 Sample + Logits / V1 Core + KV + Metrics]: # |
| 56 | # - Previously a single "V1 Test e2e + engine" step, now split across multiple groups. # |
| 57 | # - V1 e2e (2/4 GPUs) uses 4 GPUs but is scheduled on 8-GPU machines for stability. See: # |
| 58 | # https://github.com/vllm-project/vllm/pull/31040 # |
| 59 | # * [V1 Sample + Logits / V1 Core + KV + Metrics / V1 others (CPU)]: # |
| 60 | # - Previously a single "V1 others" step, now split to avoid interference. # |
| 61 | # - Integration test for streaming correctness (requires special branch for __harness__ lib). # |
| 62 | # * [V1 others (CPU)]: Split the tests to avoid interference # |
| 63 | # * [PyTorch Compilation Unit Tests]: Run unit tests defined directly under `compile/`, not including subdirectories, which # |
| 64 | # are usually heavier tests covered elsewhere. Use `find` to launch multiple instances # |
| 65 | # of pytest so that they do not suffer from: # |
| 66 | # https://github.com/vllm-project/vllm/issues/28965 # |
| 67 | # * [PyTorch Fullgraph Smoke Test]: Run smoke tests under fullgraph directory, except `test_full_graph.py` as it is a heavy # |
| 68 | # test that is covered in other steps. Use `find` to launch multiple instances of pytest # |
| 69 | # so that they do not suffer from: https://github.com/vllm-project/vllm/issues/28965 # |
| 70 | # * [PyTorch Fullgraph]: # |
| 71 | # - Limit to no custom ops to reduce running time. Wrap with quotes to escape yaml and avoid starting `-k` string # |
| 72 | # with a `-` # |
| 73 | # - Old E2E tests such as: # |
| 74 | # ```bash # |
| 75 | # pytest -v -s compile/distributed/test_fusions_e2e.py -k 'TRITON and not +quant_fp8 and not Llama-4' # |
| 76 | # ``` # |
| 77 | # were removed in https://github.com/vllm-project/vllm/pull/33293 in favor of new tests in `fusions_e2e`. We # |
| 78 | # avoid replicating the new jobs in this file as it's deprecated. # |
| 79 | # * [Basic Models Tests (Extra Initialization) %N]: Only when vLLM model source is modified - test initialization of a # |
| 80 | # large subset of supported models (the complement of the small subset in # |
| 81 | # the above test.) Also run if model initialization test file is modified. # |
| 82 | # * [Language Models Tests (Extra Standard) %N]: Shard slow subset of standard language models tests. Only run when model # |
| 83 | # source is modified, or when specified test files are modified. # |
| 84 | # * [Language Models Tests (Hybrid) %N]: Install fast path packages for testing against transformers (mamba, conv1d). # |
| 85 | # * [Language Models Test (Extended Generation)]: Install fast path packages for testing against transformers (mamba, conv1d). # |
| 86 | # * [Multi-Modal Models (Standard) 1-4]: # |
| 87 | # - Do NOT remove `VLLM_WORKER_MULTIPROC_METHOD=spawn` setting as ROCm requires this for certain models to function. # |
| 88 | # * [Transformers Nightly Models]: Whisper needs `VLLM_WORKER_MULTIPROC_METHOD=spawn` to avoid deadlock. # |
| 89 | # * [Plugin Tests (2 GPUs)]: # |
| 90 | # - {`pytest -v -s plugins_tests/test_oot_registration_online.py`}: It needs a clean process # |
| 91 | # - {`pytest -v -s plugins_tests/test_oot_registration_offline.py`}: It needs a clean process # |
| 92 | # - {`pytest -v -s plugins_tests/lora_resolvers`}: Unit tests for in-tree lora resolver plugins # |
| 93 | # * [LoRA TP (Distributed)]: # |
| 94 | # - There is some Tensor Parallelism related processing logic in LoRA that requires multi-GPU testing for validation. # |
| 95 | # - {`pytest -v -s -x lora/test_gptoss_tp.py`}: Disabled for now because MXFP4 backend on non-cuda platform doesn't support # |
| 96 | # LoRA yet. # |
| 97 | # * [Distributed Tests (NxGPUs)(HW-TAG)]: Don't test llama model here, it seems hf implementation is buggy. See: # |
| 98 | # https://github.com/vllm-project/vllm/pull/5689 # |
| 99 | # * [Distributed Tests (NxGPUs)(HW-TAG)]: Some old E2E tests were removed in https://github.com/vllm-project/vllm/pull/33293 # |
| 100 | # in favor of new tests in fusions_e2e. We avoid replicating the new jobs in # |
| 101 | # this file as it's deprecated. # |
| 102 | # # |
| 103 | ##################################################################################################################################### |
| 104 | |
| 105 | |
| 106 | steps: |
| 107 | |
| 108 | ######################################################################################################################################### |
| 109 | # # |
| 110 | # AMD CPU tests # |
| 111 | # # |
| 112 | ######################################################################################################################################### |
| 113 | |
| 114 | - label: ":computer: (CPU) Rust Frontend Cargo Style + Clippy" # TBD |
| 115 | timeout_in_minutes: 35 |
| 116 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 117 | dind: false |
| 118 | agent_pool: mi250_1 |
| 119 | no_gpu: true |
| 120 | optional: true |
| 121 | working_dir: "/vllm-workspace" |
| 122 | source_file_dependencies: |
| 123 | - rust/ |
| 124 | - rust-toolchain.toml |
| 125 | - .buildkite/test_areas/rust_frontend_cargo.yaml |
| 126 | - .buildkite/scripts/run-rust-frontend-cargo-ci.sh |
| 127 | commands: |
| 128 | - .buildkite/scripts/run-rust-frontend-cargo-ci.sh style-clippy |
| 129 | |
| 130 | - label: ":computer: (CPU) Rust Frontend Cargo" # TBD |
| 131 | timeout_in_minutes: 35 |
| 132 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 133 | dind: false |
| 134 | agent_pool: mi250_1 |
| 135 | no_gpu: true |
| 136 | optional: true |
| 137 | working_dir: "/vllm-workspace" |
| 138 | source_file_dependencies: |
| 139 | - rust/ |
| 140 | - rust-toolchain.toml |
| 141 | - .buildkite/test_areas/rust_frontend_cargo.yaml |
| 142 | - .buildkite/scripts/run-rust-frontend-cargo-ci.sh |
| 143 | commands: |
| 144 | - .buildkite/scripts/run-rust-frontend-cargo-ci.sh test |
| 145 | |
| 146 | - label: ":computer: (CPU) Docker Build Metadata" # TBD |
| 147 | timeout_in_minutes: 35 |
| 148 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 149 | dind: false |
| 150 | agent_pool: mi250_1 |
| 151 | no_gpu: true |
| 152 | optional: true |
| 153 | working_dir: "/vllm-workspace/tests" |
| 154 | source_file_dependencies: |
| 155 | - .buildkite/scripts/docker-build-metadata-args.sh |
| 156 | - .buildkite/scripts/ci-bake-rocm.sh |
| 157 | - .buildkite/release-pipeline.yaml |
| 158 | - docker/Dockerfile |
| 159 | - docker/Dockerfile.cpu |
| 160 | - docker/Dockerfile.rocm |
| 161 | - docker/Dockerfile.rocm_base |
| 162 | - docker/ci-rocm.hcl |
| 163 | - docker/docker-bake.hcl |
| 164 | - docker/docker-bake-rocm.hcl |
| 165 | - tests/tools/test_docker_build_metadata_args.py |
| 166 | commands: |
| 167 | - pytest -v -s tools/test_docker_build_metadata_args.py |
| 168 | |
| 169 | - label: ":amd: (MI250) Torch Stable ABI Audit" # TBD |
| 170 | timeout_in_minutes: 35 |
| 171 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 172 | dind: false |
| 173 | agent_pool: mi300_1 |
| 174 | no_gpu: true |
| 175 | optional: true |
| 176 | working_dir: "/vllm-workspace/" |
| 177 | source_file_dependencies: |
| 178 | - .buildkite/check-torch-abi.py |
| 179 | - csrc/ |
| 180 | - cmake/ |
| 181 | - setup.py |
| 182 | - vllm/platforms/rocm.py |
| 183 | commands: |
| 184 | - uv pip install --system torch-abi-audit==0.0.1 |
| 185 | - python3 /vllm-workspace/.buildkite/check-torch-abi.py |
| 186 | |
| 187 | ######################################################################################################################################### |
| 188 | # # |
| 189 | # MI250 (gfx90a) tests # |
| 190 | # # |
| 191 | ######################################################################################################################################### |
| 192 | |
| 193 | #---------------------------------------------------------- mi250 · compile ----------------------------------------------------------# |
| 194 | |
| 195 | - label: ":amd: (MI250) PyTorch Fullgraph Smoke" # TBD |
| 196 | timeout_in_minutes: 65 |
| 197 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 198 | dind: false |
| 199 | agent_pool: mi250_1 |
| 200 | optional: true |
| 201 | working_dir: "/vllm-workspace/tests" |
| 202 | source_file_dependencies: |
| 203 | - vllm/compilation/ |
| 204 | - vllm/model_executor/ |
| 205 | - vllm/v1/attention/ |
| 206 | - vllm/config/compilation.py |
| 207 | - csrc/ |
| 208 | - tests/compile |
| 209 | - vllm/platforms/rocm.py |
| 210 | commands: |
| 211 | - "find compile/fullgraph/ -name 'test_*.py' -not -name 'test_full_graph.py' -exec pytest -s -v {} \\\\;" |
| 212 | |
| 213 | #-------------------------------------------------------- mi250 · distributed --------------------------------------------------------# |
| 214 | |
| 215 | - label: ":amd: (MI250) Pipeline + Context Parallelism" # TBD |
| 216 | timeout_in_minutes: 90 |
| 217 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 218 | dind: false |
| 219 | agent_pool: mi250_4 |
| 220 | num_gpus: 4 |
| 221 | working_dir: "/vllm-workspace/tests" |
| 222 | source_file_dependencies: |
| 223 | - vllm/distributed/ |
| 224 | - vllm/engine/ |
| 225 | - vllm/executor/ |
| 226 | - vllm/model_executor/models/ |
| 227 | - vllm/model_executor/layers/ |
| 228 | - vllm/v1/attention/backends/ |
| 229 | - vllm/v1/attention/selector.py |
| 230 | - tests/distributed/ |
| 231 | - vllm/_aiter_ops.py |
| 232 | - vllm/platforms/rocm.py |
| 233 | - tests/v1/distributed/test_pp_dp_v2.py |
| 234 | optional: true |
| 235 | commands: |
| 236 | - pytest -v -s distributed/test_pp_cudagraph.py |
| 237 | - pytest -v -s distributed/test_pipeline_parallel.py |
| 238 | - pytest -v -s v1/distributed/test_pp_dp_v2.py |
| 239 | |
| 240 | #---------------------------------------------------------- mi250 · kernels ----------------------------------------------------------# |
| 241 | |
| 242 | - label: ":amd: (MI250) Helion Kernels Shard %N" |
| 243 | timeout_in_minutes: 45 |
| 244 | parallelism: 2 |
| 245 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 246 | dind: false |
| 247 | agent_pool: mi250_1 |
| 248 | optional: true |
| 249 | working_dir: "/vllm-workspace/tests" |
| 250 | source_file_dependencies: |
| 251 | - vllm/utils/import_utils.py |
| 252 | - tools/benchmark_helion_kernels.py |
| 253 | - tests/kernels/helion/ |
| 254 | - vllm/platforms/rocm.py |
| 255 | commands: |
| 256 | - pip install helion==1.4.0 |
| 257 | - pytest -v -s kernels/helion/ --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 258 | |
| 259 | #----------------------------------------------------- mi250 · models / language -----------------------------------------------------# |
| 260 | |
| 261 | - label: ":amd: (MI250) Language Models (PPL)" # TBD |
| 262 | timeout_in_minutes: 65 |
| 263 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 264 | dind: false |
| 265 | agent_pool: mi250_1 |
| 266 | working_dir: "/vllm-workspace/tests" |
| 267 | source_file_dependencies: |
| 268 | - vllm/ |
| 269 | - tests/models/language/generation_ppl_test |
| 270 | - vllm/model_executor/models/qwen3_5.py |
| 271 | - vllm/model_executor/models/qwen3_5_mtp.py |
| 272 | - vllm/transformers_utils/configs/qwen3_5.py |
| 273 | - vllm/transformers_utils/configs/qwen3_5_moe.py |
| 274 | - vllm/model_executor/models/qwen2.py |
| 275 | - vllm/model_executor/models/qwen3.py |
| 276 | - vllm/model_executor/models/qwen3_next.py |
| 277 | - vllm/model_executor/models/qwen3_next_mtp.py |
| 278 | - vllm/model_executor/layers/fla/ops/ |
| 279 | - vllm/_aiter_ops.py |
| 280 | - vllm/v1/attention/backends/triton_attn.py |
| 281 | - vllm/v1/attention/backends/rocm_attn.py |
| 282 | - vllm/v1/attention/backends/rocm_aiter_unified_attn.py |
| 283 | - vllm/v1/attention/backends/rocm_aiter_fa.py |
| 284 | - vllm/v1/attention/backends/flex_attention.py |
| 285 | - vllm/v1/attention/ops/ |
| 286 | - vllm/platforms/rocm.py |
| 287 | optional: true |
| 288 | commands: |
| 289 | - pytest -v -s models/language/generation_ppl_test |
| 290 | |
| 291 | #---------------------------------------------------- mi250 · models / multimodal ----------------------------------------------------# |
| 292 | |
| 293 | - label: ":amd: (MI250) Multimodal Models (Standard) 3: llava + qwen2_vl" # TBD |
| 294 | timeout_in_minutes: 55 |
| 295 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 296 | dind: false |
| 297 | agent_pool: mi250_1 |
| 298 | optional: true |
| 299 | working_dir: "/vllm-workspace/tests" |
| 300 | source_file_dependencies: |
| 301 | - vllm/ |
| 302 | - tests/models/multimodal |
| 303 | commands: |
| 304 | - pytest -v -s models/multimodal/generation/test_common.py -m core_model -k "not qwen2 and not qwen3 and not gemma" |
| 305 | - pytest -v -s models/multimodal/generation/test_qwen2_vl.py -m core_model |
| 306 | |
| 307 | #------------------------------------------------------------ mi250 · v1 -------------------------------------------------------------# |
| 308 | |
| 309 | - label: ":amd: (MI250) Batch Invariance" # TBD |
| 310 | timeout_in_minutes: 35 |
| 311 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 312 | dind: false |
| 313 | agent_pool: mi250_1 |
| 314 | optional: true |
| 315 | working_dir: "/vllm-workspace/tests" |
| 316 | source_file_dependencies: |
| 317 | - vllm/v1/attention |
| 318 | - vllm/model_executor/layers |
| 319 | - tests/v1/determinism/ |
| 320 | - vllm/_aiter_ops.py |
| 321 | - vllm/platforms/rocm.py |
| 322 | commands: |
| 323 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 324 | - pip install pytest-timeout pytest-forked |
| 325 | - pytest -v -s v1/determinism/test_batch_invariance.py |
| 326 | - pytest -v -s v1/determinism/test_rms_norm_batch_invariant.py |
| 327 | |
| 328 | - label: ":amd: (MI250) CUDAGraph" # TBD |
| 329 | timeout_in_minutes: 55 |
| 330 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 331 | dind: false |
| 332 | agent_pool: mi250_1 |
| 333 | optional: true |
| 334 | working_dir: "/vllm-workspace/tests" |
| 335 | source_file_dependencies: |
| 336 | - tests/v1/cudagraph |
| 337 | - vllm/v1/cudagraph_dispatcher.py |
| 338 | - vllm/config/compilation.py |
| 339 | - vllm/compilation |
| 340 | - vllm/v1/worker/encoder_cudagraph.py |
| 341 | - vllm/v1/worker/encoder_cudagraph_defs.py |
| 342 | - vllm/platforms/rocm.py |
| 343 | commands: |
| 344 | - pytest -v -s v1/cudagraph/test_cudagraph_dispatch.py |
| 345 | - pytest -v -s v1/cudagraph/test_cudagraph_mode.py |
| 346 | - pytest -v -s v1/cudagraph/test_breakable_cudagraph.py |
| 347 | - pytest -v -s v1/cudagraph/test_encoder_cudagraph.py |
| 348 | |
| 349 | - label: ":amd: (MI250) E2E Core" # TBD |
| 350 | timeout_in_minutes: 50 |
| 351 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 352 | dind: false |
| 353 | agent_pool: mi250_1 |
| 354 | optional: true |
| 355 | working_dir: "/vllm-workspace/tests" |
| 356 | source_file_dependencies: |
| 357 | - vllm/v1/ |
| 358 | - tests/v1/e2e/general/ |
| 359 | - vllm/platforms/rocm.py |
| 360 | commands: |
| 361 | - 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" |
| 362 | |
| 363 | - label: ":amd: (MI250) E2E Scheduling" # TBD |
| 364 | timeout_in_minutes: 55 |
| 365 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 366 | dind: false |
| 367 | agent_pool: mi250_1 |
| 368 | optional: true |
| 369 | working_dir: "/vllm-workspace/tests" |
| 370 | source_file_dependencies: |
| 371 | - vllm/v1/ |
| 372 | - tests/v1/e2e/general/ |
| 373 | - vllm/platforms/rocm.py |
| 374 | commands: |
| 375 | - pytest -v -s v1/e2e/general/test_async_scheduling.py |
| 376 | |
| 377 | - label: ":amd: (MI250) E2E Scheduling Accuracy" # TBD |
| 378 | timeout_in_minutes: 55 |
| 379 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 380 | dind: false |
| 381 | agent_pool: mi250_1 |
| 382 | optional: true |
| 383 | working_dir: "/vllm-workspace/tests" |
| 384 | source_file_dependencies: |
| 385 | - vllm/v1/ |
| 386 | - tests/v1/e2e/general/ |
| 387 | - vllm/platforms/rocm.py |
| 388 | commands: |
| 389 | - pytest -v -s v1/e2e/general/test_async_scheduling_accuracy.py |
| 390 | |
| 391 | - label: ":amd: (MI250) V1 Engine" # TBD |
| 392 | timeout_in_minutes: 45 |
| 393 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 394 | dind: false |
| 395 | agent_pool: mi250_1 |
| 396 | optional: true |
| 397 | working_dir: "/vllm-workspace/tests" |
| 398 | source_file_dependencies: |
| 399 | - vllm/v1/ |
| 400 | - tests/v1/engine/ |
| 401 | - tests/v1/test_tensor_ipc_queue.py |
| 402 | - tests/config/test_mp_reducer.py |
| 403 | - vllm/config/ |
| 404 | - vllm/engine/arg_utils.py |
| 405 | - vllm/transformers_utils/config.py |
| 406 | - vllm/platforms/rocm.py |
| 407 | - vllm/v1/engine/ |
| 408 | commands: |
| 409 | - pytest -v -s v1/engine/test_preprocess_error_handling.py |
| 410 | - pytest -v -s v1/engine --ignore v1/engine/test_preprocess_error_handling.py |
| 411 | - pytest -v -s v1/test_tensor_ipc_queue.py |
| 412 | - pytest -v -s config/test_mp_reducer.py |
| 413 | |
| 414 | - label: ":amd: (MI250) Spec Decode Draft Model" # TBD |
| 415 | timeout_in_minutes: 60 |
| 416 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 417 | dind: false |
| 418 | agent_pool: mi250_1 |
| 419 | optional: true |
| 420 | working_dir: "/vllm-workspace/tests" |
| 421 | source_file_dependencies: |
| 422 | - vllm/v1/spec_decode/ |
| 423 | - vllm/v1/worker/gpu/spec_decode/ |
| 424 | - vllm/model_executor/model_loader/ |
| 425 | - vllm/v1/sample/ |
| 426 | - vllm/model_executor/layers/ |
| 427 | - tests/v1/e2e/spec_decode/ |
| 428 | - vllm/platforms/rocm.py |
| 429 | commands: |
| 430 | - pytest -v -s v1/e2e/spec_decode/draft_model/ |
| 431 | |
| 432 | - label: ":amd: (MI250) Spec Decode Speculators + MTP" # TBD |
| 433 | timeout_in_minutes: 75 |
| 434 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250] |
| 435 | dind: false |
| 436 | agent_pool: mi250_1 |
| 437 | optional: true |
| 438 | working_dir: "/vllm-workspace/tests" |
| 439 | source_file_dependencies: |
| 440 | - vllm/v1/spec_decode/ |
| 441 | - vllm/v1/worker/gpu/spec_decode/ |
| 442 | - vllm/model_executor/model_loader/ |
| 443 | - vllm/v1/sample/ |
| 444 | - vllm/model_executor/layers/ |
| 445 | - vllm/transformers_utils/configs/speculators/ |
| 446 | - tests/v1/e2e/spec_decode/ |
| 447 | - vllm/platforms/rocm.py |
| 448 | - vllm/v1/attention/backends/ |
| 449 | commands: |
| 450 | - pytest -v -s v1/e2e/spec_decode/speculators/ |
| 451 | - pytest -v -s v1/e2e/spec_decode/mtp/ |
| 452 | |
| 453 | #------------------------------------------------------------- mi250 · misc ------------------------------------------------------------# |
| 454 | |
| 455 | - label: ":amd: (MI300) Python-only Installation" # TBD |
| 456 | timeout_in_minutes: 55 |
| 457 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 458 | dind: false |
| 459 | agent_pool: mi300_1 |
| 460 | optional: true |
| 461 | working_dir: "/vllm-workspace/tests" |
| 462 | source_file_dependencies: |
| 463 | - tests/standalone_tests/python_only_compile.sh |
| 464 | - setup.py |
| 465 | - vllm/platforms/rocm.py |
| 466 | commands: |
| 467 | - bash standalone_tests/python_only_compile.sh |
| 468 | |
| 469 | ######################################################################################################################################### |
| 470 | # # |
| 471 | # MI300 (gfx942) tests # |
| 472 | # # |
| 473 | ######################################################################################################################################### |
| 474 | |
| 475 | #----------------------------------------------------- mi300 · basic_correctness -----------------------------------------------------# |
| 476 | |
| 477 | - label: ":amd: (MI300) Distributed Models Shard %N" |
| 478 | timeout_in_minutes: 80 |
| 479 | parallelism: 3 |
| 480 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 481 | dind: false |
| 482 | agent_pool: mi300_2 |
| 483 | num_gpus: 2 |
| 484 | optional: true |
| 485 | working_dir: "/vllm-workspace/tests" |
| 486 | source_file_dependencies: |
| 487 | - vllm/models/hy_v4/ |
| 488 | - vllm/config/ |
| 489 | - vllm/v1/worker/gpu/ |
| 490 | - vllm/v1/attention/ops/ |
| 491 | - vllm/v1/attention/backend.py |
| 492 | - vllm/model_executor/model_loader/sharded_state_loader.py |
| 493 | - vllm/model_executor/models/ |
| 494 | - vllm/model_executor/layers/ |
| 495 | - vllm/v1/attention/backends/ |
| 496 | - vllm/v1/attention/selector.py |
| 497 | - vllm/_aiter_ops.py |
| 498 | - vllm/platforms/rocm.py |
| 499 | - tests/basic_correctness/ |
| 500 | - tests/model_executor/model_loader/test_sharded_state_loader.py |
| 501 | - tests/models/ |
| 502 | - vllm/v1/worker/ |
| 503 | - tests/conftest.py |
| 504 | - tests/utils.py |
| 505 | - tests/models/test_vision.py |
| 506 | commands: |
| 507 | - case "$$BUILDKITE_PARALLEL_JOB" in 0) TARGET_TEST_SUITE=MI300 pytest basic_correctness/ -v -s -m 'distributed(num_gpus=2)' && HIP_VISIBLE_DEVICES=0,1 pytest -v -s model_executor/model_loader/test_sharded_state_loader.py -m '(not slow_test)' ;; 1) pytest models/transformers/test_backend.py -v -s -m 'distributed(num_gpus=2)' && pytest models/language -v -s -m 'distributed(num_gpus=2)' && VLLM_ROCM_USE_AITER=1 pytest -v -s models/test_hyv4_rocm.py -m 'distributed(num_gpus=2)' ;; 2) pytest models/multimodal -v -s -m 'distributed(num_gpus=2)' --ignore models/multimodal/generation/test_whisper.py && VLLM_WORKER_MULTIPROC_METHOD=spawn pytest models/multimodal/generation/test_whisper.py -v -s -m 'distributed(num_gpus=2)' && pytest -v -s models/test_vision.py -m 'distributed(num_gpus=2)' ;; *) exit 2 ;; esac |
| 508 | |
| 509 | #---------------------------------------------------------- mi300 · compile ----------------------------------------------------------# |
| 510 | |
| 511 | - label: ":amd: (MI300) PyTorch Compilation" # TBD |
| 512 | timeout_in_minutes: 90 |
| 513 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 514 | dind: false |
| 515 | agent_pool: mi300_1 |
| 516 | optional: true |
| 517 | working_dir: "/vllm-workspace/tests" |
| 518 | source_file_dependencies: |
| 519 | - vllm/compilation/ |
| 520 | - vllm/model_executor/layers/ |
| 521 | - vllm/v1/worker/ |
| 522 | - vllm/v1/attention/ |
| 523 | - vllm/v1/cudagraph_dispatcher.py |
| 524 | - vllm/config/compilation.py |
| 525 | - csrc/ |
| 526 | - tests/compile |
| 527 | - vllm/platforms/rocm.py |
| 528 | - vllm/__init__.py |
| 529 | - vllm/_aiter_ops.py |
| 530 | - vllm/_custom_ops.py |
| 531 | - vllm/config/ |
| 532 | - vllm/distributed/ |
| 533 | - vllm/engine/ |
| 534 | - vllm/env_override.py |
| 535 | - vllm/envs.py |
| 536 | - vllm/forward_context.py |
| 537 | - vllm/inputs/ |
| 538 | - vllm/ir/ |
| 539 | - vllm/kernels/ |
| 540 | - vllm/logger.py |
| 541 | - vllm/model_executor/ |
| 542 | - vllm/multimodal/ |
| 543 | - vllm/platforms/ |
| 544 | - vllm/plugins/ |
| 545 | - vllm/sampling_params.py |
| 546 | - vllm/sequence.py |
| 547 | - vllm/transformers_utils/ |
| 548 | - vllm/triton_utils/ |
| 549 | - vllm/utils/ |
| 550 | - vllm/v1/ |
| 551 | commands: |
| 552 | - "find compile/ -maxdepth 1 -name 'test_*.py' -print0 | xargs -0 -n1 -I{} pytest -s -v '{}'" |
| 553 | - pytest -s -v compile/dynamic_shapes/ |
| 554 | |
| 555 | - label: ":amd: (MI300) Fusion E2E Config Sweep" # TBD |
| 556 | timeout_in_minutes: 40 |
| 557 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 558 | dind: false |
| 559 | agent_pool: mi300_1 |
| 560 | num_gpus: 1 |
| 561 | working_dir: "/vllm-workspace/" |
| 562 | source_file_dependencies: |
| 563 | - csrc/quantization/ |
| 564 | - vllm/compilation/ |
| 565 | - vllm/model_executor/layers/layernorm.py |
| 566 | - vllm/model_executor/layers/activation.py |
| 567 | - vllm/model_executor/layers/attention/attention.py |
| 568 | - vllm/model_executor/layers/quantization/input_quant_fp8.py |
| 569 | - tests/compile/fusions_e2e/ |
| 570 | - vllm/_aiter_ops.py |
| 571 | - vllm/platforms/rocm.py |
| 572 | optional: true |
| 573 | commands: |
| 574 | - rocm-smi |
| 575 | - pytest -v -s tests/compile/fusions_e2e/test_tp1_quant.py -k "llama-3" |
| 576 | |
| 577 | - label: ":amd: (MI300) Fusion E2E Quick" # TBD |
| 578 | timeout_in_minutes: 30 |
| 579 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 580 | dind: false |
| 581 | agent_pool: mi300_1 |
| 582 | num_gpus: 1 |
| 583 | working_dir: "/vllm-workspace/" |
| 584 | source_file_dependencies: |
| 585 | - csrc/quantization/ |
| 586 | - vllm/model_executor/ |
| 587 | - vllm/v1/attention/ |
| 588 | - vllm/compilation/ |
| 589 | - tests/compile/fusions_e2e/ |
| 590 | - vllm/_aiter_ops.py |
| 591 | - vllm/platforms/rocm.py |
| 592 | optional: true |
| 593 | commands: |
| 594 | - rocm-smi |
| 595 | # Run all models and attn backends but only Inductor partition and native custom ops |
| 596 | - "pytest -v -s tests/compile/fusions_e2e/test_tp1_quant.py -k 'inductor_partition and not +rms_norm and not +quant_fp8'" |
| 597 | # Different from CUDA, Qwen requires +rms_norm and +quant_fp8 as rms+quant fusion is only supported on AITER |
| 598 | - "pytest -v -s tests/compile/fusions_e2e/test_tp1_quant.py -k 'inductor_partition and +rms_norm and +quant_fp8 and qwen3'" |
| 599 | # DeepSeek-Coder uses the portable static-FP8 MLA fusion path on ROCm. |
| 600 | - "pytest -v -s tests/compile/fusions_e2e/test_tp1_quant.py -k 'inductor_partition and not +rms_norm and +quant_fp8 and DeepSeek-Coder and TRITON_MLA and not True'" |
| 601 | |
| 602 | - label: ":amd: (MI300) PyTorch Compilation Passes" # TBD |
| 603 | timeout_in_minutes: 80 |
| 604 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 605 | dind: false |
| 606 | agent_pool: mi300_1 |
| 607 | working_dir: "/vllm-workspace/tests" |
| 608 | source_file_dependencies: |
| 609 | - vllm/ |
| 610 | - tests/compile/passes |
| 611 | - vllm/platforms/rocm.py |
| 612 | - vllm/__init__.py |
| 613 | - vllm/_aiter_ops.py |
| 614 | - vllm/_custom_ops.py |
| 615 | - vllm/compilation/ |
| 616 | - vllm/config/ |
| 617 | - vllm/distributed/ |
| 618 | - vllm/engine/ |
| 619 | - vllm/env_override.py |
| 620 | - vllm/envs.py |
| 621 | - vllm/forward_context.py |
| 622 | - vllm/inputs/ |
| 623 | - vllm/ir/ |
| 624 | - vllm/kernels/ |
| 625 | - vllm/logger.py |
| 626 | - vllm/model_executor/ |
| 627 | - vllm/multimodal/ |
| 628 | - vllm/platforms/ |
| 629 | - vllm/plugins/ |
| 630 | - vllm/sampling_params.py |
| 631 | - vllm/sequence.py |
| 632 | - vllm/transformers_utils/ |
| 633 | - vllm/triton_utils/ |
| 634 | - vllm/utils/ |
| 635 | - vllm/v1/ |
| 636 | optional: true |
| 637 | commands: |
| 638 | - pytest -s -v compile/passes --ignore compile/passes/distributed |
| 639 | |
| 640 | - label: ":amd: (MI300) PyTorch Fullgraph" # TBD |
| 641 | timeout_in_minutes: 90 |
| 642 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 643 | dind: false |
| 644 | agent_pool: mi300_1 |
| 645 | optional: true |
| 646 | working_dir: "/vllm-workspace/tests" |
| 647 | source_file_dependencies: |
| 648 | - vllm/compilation/ |
| 649 | - vllm/model_executor/ |
| 650 | - vllm/v1/attention/ |
| 651 | - vllm/config/compilation.py |
| 652 | - csrc/ |
| 653 | - tests/compile |
| 654 | - vllm/platforms/rocm.py |
| 655 | - vllm/__init__.py |
| 656 | - vllm/_aiter_ops.py |
| 657 | - vllm/_custom_ops.py |
| 658 | - vllm/config/ |
| 659 | - vllm/distributed/ |
| 660 | - vllm/engine/ |
| 661 | - vllm/env_override.py |
| 662 | - vllm/envs.py |
| 663 | - vllm/forward_context.py |
| 664 | - vllm/inputs/ |
| 665 | - vllm/ir/ |
| 666 | - vllm/kernels/ |
| 667 | - vllm/logger.py |
| 668 | - vllm/multimodal/ |
| 669 | - vllm/platforms/ |
| 670 | - vllm/plugins/ |
| 671 | - vllm/sampling_params.py |
| 672 | - vllm/sequence.py |
| 673 | - vllm/transformers_utils/ |
| 674 | - vllm/triton_utils/ |
| 675 | - vllm/utils/ |
| 676 | - vllm/v1/ |
| 677 | commands: |
| 678 | - find compile/fullgraph/ -name 'test_*.py' -not -name 'test_full_cudagraph.py' -print0 | xargs -0 -n1 -I{} pytest -s -v '{}' |
| 679 | |
| 680 | - label: ":amd: (MI300) PyTorch Nightly Dependency Override Check" # TBD |
| 681 | timeout_in_minutes: 30 |
| 682 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 683 | dind: false |
| 684 | agent_pool: mi300_1 |
| 685 | optional: true |
| 686 | working_dir: "/vllm-workspace/tests" |
| 687 | source_file_dependencies: |
| 688 | - requirements/test/nightly-torch.txt |
| 689 | - vllm/platforms/rocm.py |
| 690 | commands: |
| 691 | - bash standalone_tests/pytorch_nightly_dependency.sh |
| 692 | |
| 693 | - label: ":amd: (MI300) Distributed Compile (Inductor Partition)" # TBD |
| 694 | timeout_in_minutes: 75 |
| 695 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 696 | dind: false |
| 697 | agent_pool: mi300_2 |
| 698 | num_gpus: 2 |
| 699 | working_dir: "/vllm-workspace/" |
| 700 | source_file_dependencies: |
| 701 | - vllm/compilation/ |
| 702 | - vllm/model_executor/layers |
| 703 | - tests/compile/passes/distributed/ |
| 704 | - tests/compile/fusions_e2e/ |
| 705 | - vllm/_aiter_ops.py |
| 706 | - vllm/platforms/rocm.py |
| 707 | commands: |
| 708 | - export VLLM_TEST_CLEAN_GPU_MEMORY=1 |
| 709 | - pytest -s -v tests/compile/passes/distributed |
| 710 | - pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py::test_tp2_ar_rms_fusions -k "inductor_partition" |
| 711 | |
| 712 | - label: ":amd: (MI300) Distributed Compile (Dynamo Partition)" # TBD |
| 713 | timeout_in_minutes: 65 |
| 714 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 715 | dind: false |
| 716 | agent_pool: mi300_2 |
| 717 | num_gpus: 2 |
| 718 | working_dir: "/vllm-workspace/" |
| 719 | source_file_dependencies: |
| 720 | - vllm/compilation/ |
| 721 | - vllm/model_executor/layers |
| 722 | - tests/compile/passes/distributed/ |
| 723 | - tests/compile/fusions_e2e/ |
| 724 | - vllm/_aiter_ops.py |
| 725 | - vllm/platforms/rocm.py |
| 726 | commands: |
| 727 | - export VLLM_TEST_CLEAN_GPU_MEMORY=1 |
| 728 | - VLLM_TEST_CLEAN_GPU_MEMORY=1 pytest -v -s tests/compile/passes/distributed/test_async_tp.py |
| 729 | - pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py::test_tp2_ar_rms_fusions -k "dynamo_partition" |
| 730 | |
| 731 | # - label: ":amd: (MI300) Sequence Parallel Correctness" # TBD |
| 732 | # timeout_in_minutes: 180 |
| 733 | # mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 734 | # dind: false |
| 735 | # agent_pool: mi300_2 |
| 736 | # num_gpus: 2 |
| 737 | # optional: true |
| 738 | # working_dir: "/vllm-workspace/" |
| 739 | # source_file_dependencies: |
| 740 | # - vllm/model_executor/layers/ |
| 741 | # - vllm/compilation/ |
| 742 | # - vllm/v1/worker/ |
| 743 | # - vllm/v1/cudagraph_dispatcher.py |
| 744 | # - tests/compile/correctness_e2e/test_sequence_parallel.py |
| 745 | # - vllm/platforms/rocm.py |
| 746 | # commands: |
| 747 | # - export VLLM_TEST_CLEAN_GPU_MEMORY=1 |
| 748 | # - pytest -v -s tests/compile/correctness_e2e/test_sequence_parallel.py |
| 749 | |
| 750 | - label: ":amd: (MI300) Distributed Compile + RPC" # TBD |
| 751 | timeout_in_minutes: 65 |
| 752 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 753 | dind: false |
| 754 | agent_pool: mi300_2 |
| 755 | num_gpus: 2 |
| 756 | working_dir: "/vllm-workspace/tests" |
| 757 | source_file_dependencies: |
| 758 | - vllm/compilation/ |
| 759 | - vllm/distributed/ |
| 760 | - vllm/engine/ |
| 761 | - vllm/executor/ |
| 762 | - vllm/worker/worker_base.py |
| 763 | - vllm/v1/engine/ |
| 764 | - vllm/v1/worker/ |
| 765 | - tests/compile/fullgraph/test_basic_correctness.py |
| 766 | - tests/compile/test_wrapper.py |
| 767 | - tests/entrypoints/llm/test_collective_rpc.py |
| 768 | - vllm/platforms/rocm.py |
| 769 | optional: true |
| 770 | commands: |
| 771 | - pytest -v -s entrypoints/llm/test_collective_rpc.py |
| 772 | - pytest -v -s ./compile/test_wrapper.py |
| 773 | |
| 774 | #----------------------------------------------------------- mi300 · cuda ------------------------------------------------------------# |
| 775 | |
| 776 | - label: ":amd: (MI300) CUDA Platform" # TBD |
| 777 | timeout_in_minutes: 25 |
| 778 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 779 | dind: false |
| 780 | agent_pool: mi300_1 |
| 781 | working_dir: "/vllm-workspace/tests" |
| 782 | source_file_dependencies: |
| 783 | - vllm/ |
| 784 | - tests/cuda |
| 785 | - vllm/envs.py |
| 786 | - vllm/logger.py |
| 787 | - vllm/platforms/ |
| 788 | - vllm/plugins/ |
| 789 | - vllm/utils/ |
| 790 | optional: true |
| 791 | commands: |
| 792 | - pytest -v -s cuda/test_cuda_context.py |
| 793 | - pytest -v -s cuda/test_platform_no_cuda_init.py |
| 794 | - pytest -v -s cuda/test_cuda_compatibility_path.py |
| 795 | |
| 796 | #-------------------------------------------------------- mi300 · detokenizer --------------------------------------------------------# |
| 797 | |
| 798 | - label: ":amd: (MI300) Async Engine, Inputs, Utils, Worker" # TBD |
| 799 | timeout_in_minutes: 50 |
| 800 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 801 | dind: false |
| 802 | agent_pool: mi300_1 |
| 803 | optional: true |
| 804 | working_dir: "/vllm-workspace/tests" |
| 805 | source_file_dependencies: |
| 806 | - vllm/ |
| 807 | - tests/detokenizer |
| 808 | - tests/multimodal |
| 809 | - tests/utils_ |
| 810 | - vllm/assets/ |
| 811 | - vllm/config/ |
| 812 | - vllm/distributed/ |
| 813 | - vllm/engine/ |
| 814 | - vllm/inputs/ |
| 815 | - vllm/model_executor/ |
| 816 | - vllm/multimodal/ |
| 817 | - vllm/platforms/ |
| 818 | - vllm/sampling_params.py |
| 819 | - vllm/tokenizers/ |
| 820 | - vllm/transformers_utils/ |
| 821 | - vllm/utils/ |
| 822 | - vllm/v1/ |
| 823 | commands: |
| 824 | - pytest -v -s detokenizer |
| 825 | - pytest -v -s -m 'not cpu_test' multimodal |
| 826 | - pytest -v -s multimodal/test_cache.py::test_sleep_wake_preserves_mm_cache_consistency |
| 827 | - pytest -v -s utils_ |
| 828 | |
| 829 | #-------------------------------------------------------- mi300 · distributed --------------------------------------------------------# |
| 830 | |
| 831 | - label: ":amd: (MI300) Distributed Comm Ops" # TBD |
| 832 | timeout_in_minutes: 30 |
| 833 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 834 | dind: false |
| 835 | agent_pool: mi300_2 |
| 836 | num_gpus: 2 |
| 837 | working_dir: "/vllm-workspace/tests" |
| 838 | source_file_dependencies: |
| 839 | - vllm/distributed |
| 840 | - tests/distributed |
| 841 | - vllm/platforms/rocm.py |
| 842 | commands: |
| 843 | - pytest -v -s distributed/test_comm_ops.py |
| 844 | - pytest -v -s distributed/test_shm_broadcast.py |
| 845 | - pytest -v -s distributed/test_shm_buffer.py |
| 846 | - pytest -v -s distributed/test_shm_storage.py |
| 847 | |
| 848 | - label: ":amd: (MI300) EPLB Execution" # TBD |
| 849 | timeout_in_minutes: 35 |
| 850 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 851 | dind: false |
| 852 | agent_pool: mi300_4 |
| 853 | num_gpus: 4 |
| 854 | working_dir: "/vllm-workspace/tests" |
| 855 | source_file_dependencies: |
| 856 | - vllm/distributed/eplb |
| 857 | - tests/distributed/test_eplb_execute.py |
| 858 | - tests/distributed/test_eplb_spec_decode.py |
| 859 | - vllm/platforms/rocm.py |
| 860 | commands: |
| 861 | - pytest -v -s distributed/test_eplb_execute.py |
| 862 | - pytest -v -s distributed/test_eplb_spec_decode.py |
| 863 | |
| 864 | - label: ":amd: (MI300) Distributed Features" # TBD |
| 865 | timeout_in_minutes: 90 |
| 866 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 867 | dind: false |
| 868 | agent_pool: mi300_2 |
| 869 | num_gpus: 2 |
| 870 | working_dir: "/vllm-workspace/" |
| 871 | source_file_dependencies: |
| 872 | - vllm/distributed/ |
| 873 | - vllm/v1/distributed/ |
| 874 | - vllm/model_executor/layers/fused_moe/ |
| 875 | - vllm/v1/attention/backends/ |
| 876 | - vllm/v1/attention/selector.py |
| 877 | - tests/v1/distributed/test_dbo.py |
| 878 | - tests/distributed/test_context_parallel.py |
| 879 | - examples/features/data_parallel/data_parallel_offline.py |
| 880 | - vllm/_aiter_ops.py |
| 881 | - vllm/platforms/rocm.py |
| 882 | - csrc/custom_quickreduce.cu |
| 883 | - csrc/ops.h |
| 884 | - csrc/torch_bindings.cpp |
| 885 | - vllm/model_executor/layers/ |
| 886 | - vllm/entrypoints/llm.py |
| 887 | - vllm/config/parallel.py |
| 888 | - vllm/v1/engine/ |
| 889 | - vllm/v1/executor/ |
| 890 | - vllm/v1/worker/ |
| 891 | - vllm/_custom_ops.py |
| 892 | - vllm/envs.py |
| 893 | - tests/distributed/test_rocm_aiter_custom_ar.py |
| 894 | - tests/distributed/test_rocm_quick_reduce.py |
| 895 | - tests/distributed/test_quick_all_reduce.py |
| 896 | - tests/v1/e2e/general/test_rocm_aiter_custom_ar.py |
| 897 | - tests/utils.py |
| 898 | - examples/rl/rlhf_async_new_apis.py |
| 899 | - tests/distributed/test_weight_transfer.py |
| 900 | - tests/distributed/test_packed_tensor.py |
| 901 | optional: true |
| 902 | commands: |
| 903 | - pytest -v -s tests/distributed/test_context_parallel.py |
| 904 | - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 examples/rl/rlhf_async_new_apis.py |
| 905 | - VLLM_LOGGING_LEVEL=DEBUG python3 examples/features/data_parallel/data_parallel_offline.py --model=Qwen/Qwen1.5-MoE-A2.7B -tp=1 -dp=2 --max-model-len=2048 --all2all-backend=deepep_high_throughput |
| 906 | - VLLM_LOGGING_LEVEL=DEBUG python3 examples/features/data_parallel/data_parallel_offline.py --model=Qwen/Qwen1.5-MoE-A2.7B -tp=1 -dp=2 --max-model-len=2048 --all2all-backend=allgather_reducescatter --disable-nccl-for-dp-synchronization |
| 907 | - pytest -v -s tests/v1/distributed/test_dbo.py |
| 908 | - VLLM_ALLOW_INSECURE_SERIALIZATION=1 pytest -v -s tests/distributed/test_weight_transfer.py |
| 909 | - pytest -v -s tests/distributed/test_packed_tensor.py |
| 910 | - pytest -v -s tests/distributed/test_rocm_aiter_custom_ar.py |
| 911 | - pytest -v -s tests/v1/e2e/general/test_rocm_aiter_custom_ar.py |
| 912 | - pytest -v -s tests/distributed/test_rocm_quick_reduce.py |
| 913 | - pytest -v -s tests/distributed/test_quick_all_reduce.py |
| 914 | |
| 915 | - label: ":amd: (MI300) Distributed" # TBD |
| 916 | timeout_in_minutes: 40 |
| 917 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 918 | dind: false |
| 919 | agent_pool: mi300_4 |
| 920 | num_gpus: 4 |
| 921 | optional: true |
| 922 | working_dir: "/vllm-workspace/tests" |
| 923 | source_file_dependencies: |
| 924 | - vllm/ |
| 925 | - tests/models/test_vision.py |
| 926 | commands: |
| 927 | - pytest -v -s distributed/test_custom_all_reduce.py |
| 928 | - torchrun --nproc_per_node=2 distributed/test_ca_buffer_sharing.py |
| 929 | - TARGET_TEST_SUITE=MI300 pytest basic_correctness/ -v -s -m 'distributed(num_gpus=2)' |
| 930 | - pytest -v -s -x lora/test_mixtral.py |
| 931 | - pytest -v -s models/test_vision.py -m 'distributed(num_gpus=4)' |
| 932 | |
| 933 | - label: ":amd: (MI300) Distributed Torchrun + Examples" # TBD |
| 934 | timeout_in_minutes: 55 |
| 935 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 936 | dind: false |
| 937 | agent_pool: mi300_4 |
| 938 | num_gpus: 4 |
| 939 | working_dir: "/vllm-workspace/tests" |
| 940 | source_file_dependencies: |
| 941 | - vllm/distributed/ |
| 942 | - tests/distributed/test_torchrun_example.py |
| 943 | - tests/distributed/test_torchrun_example_moe.py |
| 944 | - examples/rl/ |
| 945 | - examples/features/data_parallel/data_parallel_offline.py |
| 946 | - vllm/platforms/rocm.py |
| 947 | commands: |
| 948 | - torchrun --nproc-per-node=4 distributed/test_torchrun_example.py |
| 949 | - PP_SIZE=2 torchrun --nproc-per-node=4 distributed/test_torchrun_example.py |
| 950 | - TP_SIZE=4 torchrun --nproc-per-node=4 distributed/test_torchrun_example_moe.py |
| 951 | - PP_SIZE=2 TP_SIZE=2 torchrun --nproc-per-node=4 distributed/test_torchrun_example_moe.py |
| 952 | - DP_SIZE=4 ENABLE_EP=1 torchrun --nproc-per-node=4 distributed/test_torchrun_example_moe.py |
| 953 | - TP_SIZE=2 DP_SIZE=2 ENABLE_EP=1 torchrun --nproc-per-node=4 distributed/test_torchrun_example_moe.py |
| 954 | - python3 ../examples/features/data_parallel/data_parallel_offline.py --enforce-eager |
| 955 | # rlhf examples |
| 956 | - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 ../examples/rl/rlhf_http_nccl.py |
| 957 | - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 ../examples/rl/rlhf_http_ipc.py |
| 958 | |
| 959 | - label: ":amd: (MI300) Elastic EP Scaling" # TBD |
| 960 | timeout_in_minutes: 55 |
| 961 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 962 | dind: false |
| 963 | agent_pool: mi300_4 |
| 964 | num_gpus: 4 |
| 965 | optional: true |
| 966 | working_dir: "/vllm-workspace/tests" |
| 967 | source_file_dependencies: |
| 968 | - vllm/distributed/ |
| 969 | - vllm/engine/ |
| 970 | - vllm/executor/ |
| 971 | - vllm/compilation/ |
| 972 | - tests/distributed/ |
| 973 | - vllm/platforms/rocm.py |
| 974 | commands: |
| 975 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 976 | - pytest -v -s distributed/test_elastic_ep.py |
| 977 | |
| 978 | - label: ":amd: (MI300) RayExecutorV2" # TBD |
| 979 | timeout_in_minutes: 75 |
| 980 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 981 | dind: false |
| 982 | agent_pool: mi300_4 |
| 983 | num_gpus: 4 |
| 984 | working_dir: "/vllm-workspace/tests" |
| 985 | source_file_dependencies: |
| 986 | - vllm/v1/executor/ray_executor_v2.py |
| 987 | - vllm/v1/executor/abstract.py |
| 988 | - vllm/v1/executor/multiproc_executor.py |
| 989 | - tests/distributed/test_ray_v2_executor.py |
| 990 | - tests/distributed/test_ray_v2_executor_e2e.py |
| 991 | - tests/distributed/test_pipeline_parallel.py |
| 992 | - tests/basic_correctness/models/test_basic_correctness.py |
| 993 | - vllm/platforms/rocm.py |
| 994 | commands: |
| 995 | - export VLLM_USE_RAY_V2_EXECUTOR_BACKEND=1 |
| 996 | - pytest -v -s distributed/test_ray_v2_executor.py |
| 997 | - pytest -v -s distributed/test_ray_v2_executor_e2e.py |
| 998 | - pytest -v -s distributed/test_pipeline_parallel.py -k "ray" |
| 999 | - TARGET_TEST_SUITE=L4 pytest -v -s basic_correctness/models/test_basic_correctness.py -k "ray" |
| 1000 | |
| 1001 | - label: ":amd: (MI300) Distributed DP + EP" # TBD |
| 1002 | timeout_in_minutes: 50 |
| 1003 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1004 | dind: false |
| 1005 | agent_pool: mi300_8 |
| 1006 | num_gpus: 8 |
| 1007 | optional: true |
| 1008 | working_dir: "/vllm-workspace/tests" |
| 1009 | source_file_dependencies: |
| 1010 | - examples/features/torchrun/torchrun_dp_example_offline.py |
| 1011 | - vllm/config/parallel.py |
| 1012 | - vllm/distributed/ |
| 1013 | - vllm/v1/engine/llm_engine.py |
| 1014 | - vllm/v1/executor/uniproc_executor.py |
| 1015 | - vllm/v1/worker/gpu_worker.py |
| 1016 | - vllm/platforms/rocm.py |
| 1017 | commands: |
| 1018 | - torchrun --nproc-per-node=8 ../examples/features/torchrun/torchrun_dp_example_offline.py --tp-size=2 --pp-size=1 --dp-size=4 --enable-ep |
| 1019 | |
| 1020 | - label: ":amd: (MI300) Distributed Torchrun + Shutdown" # TBD |
| 1021 | timeout_in_minutes: 35 |
| 1022 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1023 | dind: false |
| 1024 | agent_pool: mi300_2 |
| 1025 | num_gpus: 2 |
| 1026 | working_dir: "/vllm-workspace/tests" |
| 1027 | source_file_dependencies: |
| 1028 | - vllm/distributed/ |
| 1029 | - vllm/engine/ |
| 1030 | - vllm/executor/ |
| 1031 | - vllm/worker/worker_base.py |
| 1032 | - vllm/v1/engine/ |
| 1033 | - vllm/v1/worker/ |
| 1034 | - tests/distributed/ |
| 1035 | - tests/v1/shutdown |
| 1036 | - tests/v1/worker/test_worker_memory_snapshot.py |
| 1037 | - vllm/platforms/rocm.py |
| 1038 | commands: |
| 1039 | - VLLM_TEST_SAME_HOST=1 torchrun --nproc-per-node=4 distributed/test_same_node.py | grep 'Same node test passed' |
| 1040 | - VLLM_TEST_SAME_HOST=1 VLLM_TEST_WITH_DEFAULT_DEVICE_SET=1 torchrun --nproc-per-node=4 distributed/test_same_node.py | grep 'Same node test passed' |
| 1041 | - HIP_VISIBLE_DEVICES=0,1 pytest -v -s v1/shutdown |
| 1042 | - pytest -v -s v1/worker/test_worker_memory_snapshot.py |
| 1043 | |
| 1044 | - label: ":amd: (MI300) Distributed Compile + Comm" # TBD |
| 1045 | timeout_in_minutes: 65 |
| 1046 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1047 | dind: false |
| 1048 | agent_pool: mi300_4 |
| 1049 | num_gpus: 4 |
| 1050 | working_dir: "/vllm-workspace/tests" |
| 1051 | source_file_dependencies: |
| 1052 | - vllm/distributed/ |
| 1053 | - tests/distributed/test_pynccl |
| 1054 | - tests/distributed/test_events |
| 1055 | - tests/compile/fullgraph/test_basic_correctness.py |
| 1056 | - tests/distributed/test_symm_mem_allreduce.py |
| 1057 | - tests/distributed/test_multiproc_executor.py |
| 1058 | - vllm/platforms/rocm.py |
| 1059 | commands: |
| 1060 | - pytest -v -s compile/fullgraph/test_basic_correctness.py |
| 1061 | - pytest -v -s distributed/test_pynccl.py |
| 1062 | - pytest -v -s distributed/test_events.py |
| 1063 | - pytest -v -s distributed/test_symm_mem_allreduce.py |
| 1064 | - pytest -v -s distributed/test_multiproc_executor.py::test_multiproc_executor_multi_node |
| 1065 | |
| 1066 | #-------------------------------------------------------- mi300 · entrypoints --------------------------------------------------------# |
| 1067 | |
| 1068 | - label: ":amd: (MI300) Entrypoints Unit" # TBD |
| 1069 | timeout_in_minutes: 60 |
| 1070 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1071 | dind: false |
| 1072 | agent_pool: mi300_1 |
| 1073 | fast_check: true |
| 1074 | optional: true |
| 1075 | working_dir: "/vllm-workspace/tests" |
| 1076 | source_file_dependencies: |
| 1077 | - vllm/entrypoints |
| 1078 | - tests/entrypoints/unit_tests |
| 1079 | - tests/entrypoints/weight_transfer |
| 1080 | - tests/entrypoints/launchers |
| 1081 | - vllm/platforms/rocm.py |
| 1082 | commands: |
| 1083 | - pytest -v -s entrypoints/unit_tests |
| 1084 | - pytest -v -s entrypoints/weight_transfer |
| 1085 | - pytest -v -s entrypoints/launchers |
| 1086 | |
| 1087 | - label: ":amd: (MI300) Entrypoints Integration (Responses API)" # TBD |
| 1088 | timeout_in_minutes: 50 |
| 1089 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1090 | dind: false |
| 1091 | agent_pool: mi300_1 |
| 1092 | optional: true |
| 1093 | fast_check: true |
| 1094 | working_dir: "/vllm-workspace/tests" |
| 1095 | source_file_dependencies: |
| 1096 | - vllm/ |
| 1097 | - tests/entrypoints/openai/responses |
| 1098 | commands: |
| 1099 | - pytest -v -s entrypoints/openai/responses |
| 1100 | |
| 1101 | - label: ":amd: (MI300) Entrypoints Integration (Speech to Text)" # TBD |
| 1102 | timeout_in_minutes: 60 |
| 1103 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1104 | dind: false |
| 1105 | agent_pool: mi300_1 |
| 1106 | fast_check: true |
| 1107 | working_dir: "/vllm-workspace/tests" |
| 1108 | source_file_dependencies: |
| 1109 | - vllm/ |
| 1110 | - tests/entrypoints/speech_to_text |
| 1111 | optional: true |
| 1112 | commands: |
| 1113 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 1114 | - pytest -v -s entrypoints/speech_to_text |
| 1115 | |
| 1116 | - label: ":amd: (MI300) Entrypoints Integration (Multimodal)" |
| 1117 | timeout_in_minutes: 55 |
| 1118 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1119 | dind: false |
| 1120 | agent_pool: mi300_1 |
| 1121 | optional: true |
| 1122 | fast_check: true |
| 1123 | working_dir: "/vllm-workspace/tests" |
| 1124 | source_file_dependencies: |
| 1125 | - vllm/ |
| 1126 | - tests/entrypoints/multimodal |
| 1127 | commands: |
| 1128 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 1129 | - pytest -v -s entrypoints/multimodal |
| 1130 | |
| 1131 | - label: ":amd: (MI300) Scale-out EC Connector E2E" |
| 1132 | timeout_in_minutes: 60 |
| 1133 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1134 | dind: false |
| 1135 | agent_pool: mi300_2 |
| 1136 | num_gpus: 2 |
| 1137 | optional: true |
| 1138 | working_dir: "/vllm-workspace/tests" |
| 1139 | source_file_dependencies: |
| 1140 | - vllm/distributed/ec_transfer/ |
| 1141 | - vllm/entrypoints/scale_out/ |
| 1142 | - tests/entrypoints/scale_out/ec_integration/ |
| 1143 | - vllm/platforms/rocm.py |
| 1144 | commands: |
| 1145 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 1146 | - bash entrypoints/scale_out/ec_integration/run_scale_out_ec_e2e_test.sh |
| 1147 | |
| 1148 | #----------------------------------------------------------- mi300 · evals -----------------------------------------------------------# |
| 1149 | |
| 1150 | - label: ":amd: (MI300) LM Eval Small Models" # TBD |
| 1151 | timeout_in_minutes: 45 |
| 1152 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1153 | dind: false |
| 1154 | agent_pool: mi300_1 |
| 1155 | working_dir: "/vllm-workspace/tests" |
| 1156 | source_file_dependencies: |
| 1157 | - csrc/ |
| 1158 | - vllm/model_executor/layers/quantization |
| 1159 | - vllm/model_executor/models/ |
| 1160 | - vllm/model_executor/model_loader/ |
| 1161 | - vllm/v1/attention/backends/ |
| 1162 | - vllm/v1/attention/selector.py |
| 1163 | - vllm/_aiter_ops.py |
| 1164 | - vllm/platforms/rocm.py |
| 1165 | optional: true |
| 1166 | commands: |
| 1167 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-small.txt |
| 1168 | |
| 1169 | - label: ":amd: (MI300) DeepSeek V2-Lite Prefetch Offload Accuracy" # TBD |
| 1170 | timeout_in_minutes: 30 |
| 1171 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1172 | dind: false |
| 1173 | agent_pool: mi300_1 |
| 1174 | num_gpus: 1 |
| 1175 | working_dir: "/vllm-workspace" |
| 1176 | source_file_dependencies: |
| 1177 | - vllm/model_executor/models/ |
| 1178 | - vllm/model_executor/model_loader/ |
| 1179 | - vllm/model_executor/layers/fused_moe/ |
| 1180 | - vllm/model_executor/layers/quantization/ |
| 1181 | - vllm/v1/attention/backends/ |
| 1182 | - vllm/v1/attention/backends/mla/ |
| 1183 | - vllm/v1/attention/selector.py |
| 1184 | - .buildkite/scripts/scheduled_integration_test/ |
| 1185 | - vllm/_aiter_ops.py |
| 1186 | - vllm/platforms/rocm.py |
| 1187 | optional: true |
| 1188 | commands: |
| 1189 | - bash .buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_prefetch_offload.sh 0.25 200 8030 |
| 1190 | |
| 1191 | - label: ":amd: (MI300) MRCR Eval Small Models" # TBD |
| 1192 | timeout_in_minutes: 40 |
| 1193 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1194 | dind: false |
| 1195 | agent_pool: mi300_1 |
| 1196 | optional: true |
| 1197 | working_dir: "/vllm-workspace/tests" |
| 1198 | source_file_dependencies: |
| 1199 | - vllm/model_executor/models/ |
| 1200 | - vllm/model_executor/model_loader/ |
| 1201 | - vllm/v1/attention/backends/ |
| 1202 | - vllm/v1/attention/selector.py |
| 1203 | - vllm/_aiter_ops.py |
| 1204 | - vllm/platforms/rocm.py |
| 1205 | - tests/evals/mrcr/ |
| 1206 | commands: |
| 1207 | - pytest -s -v evals/mrcr/test_mrcr_correctness.py --config-list-file=evals/mrcr/configs/models-small.txt |
| 1208 | |
| 1209 | - label: ":amd: (MI300) LM Eval KV-Offload" # TBD |
| 1210 | timeout_in_minutes: 55 |
| 1211 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1212 | dind: false |
| 1213 | agent_pool: mi300_1 |
| 1214 | optional: true |
| 1215 | working_dir: "/vllm-workspace/tests" |
| 1216 | source_file_dependencies: |
| 1217 | - vllm/distributed/kv_transfer/kv_connector/v1/offloading/ |
| 1218 | - vllm/distributed/kv_transfer/kv_connector/v1/simple_cpu_offload_connector.py |
| 1219 | - vllm/v1/kv_offload/ |
| 1220 | - vllm/v1/simple_kv_offload/ |
| 1221 | - tests/evals/gsm8k/test_gsm8k_offloading.py |
| 1222 | - vllm/platforms/rocm.py |
| 1223 | commands: |
| 1224 | - pytest -s -v evals/gsm8k/test_gsm8k_offloading.py -k "nemotron-h-8b or gemma-4-e4b-it" |
| 1225 | |
| 1226 | - label: ":amd: (MI300) LM Eval KV-Offload Medium" # TBD |
| 1227 | timeout_in_minutes: 70 |
| 1228 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1229 | dind: false |
| 1230 | agent_pool: mi300_2 |
| 1231 | num_gpus: 2 |
| 1232 | optional: true |
| 1233 | working_dir: "/vllm-workspace/tests" |
| 1234 | source_file_dependencies: |
| 1235 | - vllm/distributed/kv_transfer/kv_connector/v1/offloading/ |
| 1236 | - vllm/distributed/kv_transfer/kv_connector/v1/simple_cpu_offload_connector.py |
| 1237 | - vllm/v1/kv_offload/ |
| 1238 | - vllm/v1/simple_kv_offload/ |
| 1239 | - tests/evals/gsm8k/test_gsm8k_offloading.py |
| 1240 | - vllm/platforms/rocm.py |
| 1241 | commands: |
| 1242 | - pytest -s -v evals/gsm8k/test_gsm8k_offloading.py -k "qwen3.5-35b or deepseek-v2-lite" |
| 1243 | |
| 1244 | - label: ":amd: (MI300) LM Eval KV-Offload Large" # TBD |
| 1245 | timeout_in_minutes: 65 |
| 1246 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1247 | dind: false |
| 1248 | agent_pool: mi300_4 |
| 1249 | num_gpus: 4 |
| 1250 | optional: true |
| 1251 | working_dir: "/vllm-workspace/tests" |
| 1252 | source_file_dependencies: |
| 1253 | - vllm/distributed/kv_transfer/kv_connector/v1/offloading/ |
| 1254 | - vllm/distributed/kv_transfer/kv_connector/v1/simple_cpu_offload_connector.py |
| 1255 | - vllm/v1/kv_offload/ |
| 1256 | - vllm/v1/simple_kv_offload/ |
| 1257 | - tests/evals/gsm8k/test_gsm8k_offloading.py |
| 1258 | - vllm/platforms/rocm.py |
| 1259 | commands: |
| 1260 | - VLLM_ROCM_USE_AITER=1 pytest -s -v evals/gsm8k/test_gsm8k_offloading.py -k "deepseek-v4-flash" |
| 1261 | |
| 1262 | - label: ":amd: (MI300) LM Eval Small Models Harness" # TBD |
| 1263 | timeout_in_minutes: 35 |
| 1264 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1265 | dind: false |
| 1266 | agent_pool: mi300_1 |
| 1267 | optional: true |
| 1268 | working_dir: "/vllm-workspace/.buildkite/lm-eval-harness" |
| 1269 | source_file_dependencies: |
| 1270 | - csrc/ |
| 1271 | - vllm/model_executor/layers/quantization |
| 1272 | - vllm/model_executor/models/ |
| 1273 | - vllm/model_executor/model_loader/ |
| 1274 | - vllm/v1/attention/backends/ |
| 1275 | - vllm/v1/attention/selector.py |
| 1276 | - vllm/_aiter_ops.py |
| 1277 | - vllm/platforms/rocm.py |
| 1278 | - .buildkite/lm-eval-harness/ |
| 1279 | commands: |
| 1280 | - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-small-rocm.txt |
| 1281 | |
| 1282 | - label: ":amd: (MI300) GPQA Eval (GPT-OSS)" # TBD |
| 1283 | timeout_in_minutes: 40 |
| 1284 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1285 | dind: false |
| 1286 | agent_pool: mi300_2 |
| 1287 | num_gpus: 2 |
| 1288 | optional: true |
| 1289 | working_dir: "/vllm-workspace/tests" |
| 1290 | source_file_dependencies: |
| 1291 | - csrc/ |
| 1292 | - vllm/model_executor/layers/quantization |
| 1293 | - vllm/model_executor/models/ |
| 1294 | - vllm/model_executor/model_loader/ |
| 1295 | - vllm/v1/attention/backends/ |
| 1296 | - vllm/v1/attention/selector.py |
| 1297 | - vllm/model_executor/layers/fused_moe/ |
| 1298 | - vllm/_aiter_ops.py |
| 1299 | - vllm/platforms/rocm.py |
| 1300 | - tests/evals/gpt_oss/ |
| 1301 | commands: |
| 1302 | - pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-gfx942.txt |
| 1303 | |
| 1304 | - label: ":amd: (MI300) LM Eval Small Models Distributed" |
| 1305 | timeout_in_minutes: 55 |
| 1306 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1307 | dind: false |
| 1308 | agent_pool: mi300_2 |
| 1309 | num_gpus: 2 |
| 1310 | working_dir: "/vllm-workspace/tests" |
| 1311 | source_file_dependencies: |
| 1312 | - csrc/ |
| 1313 | - vllm/model_executor/layers/quantization |
| 1314 | - vllm/model_executor/models/ |
| 1315 | - vllm/model_executor/model_loader/ |
| 1316 | - vllm/v1/attention/backends/ |
| 1317 | - vllm/v1/attention/selector.py |
| 1318 | - vllm/_aiter_ops.py |
| 1319 | - vllm/platforms/rocm.py |
| 1320 | - tests/evals/gsm8k/configs/models-mi3xx-fp8-and-mixed.txt |
| 1321 | optional: true |
| 1322 | commands: |
| 1323 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-mi3xx-fp8-and-mixed.txt |
| 1324 | |
| 1325 | - label: ":amd: (MI300) DeepSeek V2-Lite Sync EPLB Accuracy" # TBD |
| 1326 | timeout_in_minutes: 40 |
| 1327 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1328 | dind: false |
| 1329 | agent_pool: mi300_4 |
| 1330 | num_gpus: 4 |
| 1331 | optional: true |
| 1332 | working_dir: "/vllm-workspace" |
| 1333 | source_file_dependencies: |
| 1334 | - vllm/model_executor/models/ |
| 1335 | - vllm/model_executor/model_loader/ |
| 1336 | - vllm/distributed/eplb |
| 1337 | - vllm/model_executor/layers/fused_moe/ |
| 1338 | - vllm/model_executor/layers/quantization/ |
| 1339 | - vllm/v1/attention/backends/ |
| 1340 | - vllm/v1/attention/backends/mla/ |
| 1341 | - vllm/v1/attention/selector.py |
| 1342 | - .buildkite/scripts/scheduled_integration_test/ |
| 1343 | - vllm/_aiter_ops.py |
| 1344 | - vllm/platforms/rocm.py |
| 1345 | commands: |
| 1346 | - VLLM_ENGINE_READY_TIMEOUT_S=1800 bash .buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_ep_eplb.sh 0.25 200 8010 |
| 1347 | |
| 1348 | - label: ":amd: (MI300) LM Eval Large Models Harness" # TBD |
| 1349 | timeout_in_minutes: 85 |
| 1350 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1351 | dind: false |
| 1352 | agent_pool: mi300_4 |
| 1353 | num_gpus: 4 |
| 1354 | optional: true |
| 1355 | working_dir: "/vllm-workspace/.buildkite/lm-eval-harness" |
| 1356 | source_file_dependencies: |
| 1357 | - csrc/ |
| 1358 | - vllm/model_executor/layers/quantization |
| 1359 | - vllm/model_executor/models/ |
| 1360 | - vllm/model_executor/model_loader/ |
| 1361 | - vllm/v1/attention/backends/ |
| 1362 | - vllm/v1/attention/selector.py |
| 1363 | - vllm/_aiter_ops.py |
| 1364 | - vllm/platforms/rocm.py |
| 1365 | commands: |
| 1366 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 1367 | - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large.txt --tp-size=4 |
| 1368 | |
| 1369 | - label: ":amd: (MI300) Qwen3-30B-A3B-FP8-block Sync EPLB Accuracy" # TBD |
| 1370 | timeout_in_minutes: 30 |
| 1371 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1372 | dind: false |
| 1373 | agent_pool: mi300_4 |
| 1374 | num_gpus: 4 |
| 1375 | optional: true |
| 1376 | working_dir: "/vllm-workspace" |
| 1377 | source_file_dependencies: |
| 1378 | - vllm/model_executor/models/ |
| 1379 | - vllm/model_executor/model_loader/ |
| 1380 | - vllm/model_executor/layers/quantization/ |
| 1381 | - vllm/distributed/eplb |
| 1382 | - vllm/model_executor/layers/fused_moe/ |
| 1383 | - vllm/v1/attention/backends/ |
| 1384 | - vllm/v1/attention/selector.py |
| 1385 | - .buildkite/scripts/scheduled_integration_test/ |
| 1386 | - vllm/_aiter_ops.py |
| 1387 | - vllm/platforms/rocm.py |
| 1388 | commands: |
| 1389 | - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_block_ep_eplb.sh 0.8 200 8020 |
| 1390 | |
| 1391 | - label: ":amd: (MI300) Qwen3-30B-A3B-FP8 DP4 Async EPLB Accuracy" # TBD |
| 1392 | timeout_in_minutes: 30 |
| 1393 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1394 | dind: false |
| 1395 | agent_pool: mi300_4 |
| 1396 | num_gpus: 4 |
| 1397 | optional: true |
| 1398 | working_dir: "/vllm-workspace" |
| 1399 | source_file_dependencies: |
| 1400 | - vllm/model_executor/models/ |
| 1401 | - vllm/model_executor/model_loader/ |
| 1402 | - vllm/model_executor/layers/quantization/ |
| 1403 | - vllm/distributed/eplb |
| 1404 | - vllm/model_executor/layers/fused_moe/ |
| 1405 | - vllm/v1/attention/backends/ |
| 1406 | - vllm/v1/attention/selector.py |
| 1407 | - .buildkite/scripts/scheduled_integration_test/ |
| 1408 | - vllm/_aiter_ops.py |
| 1409 | - vllm/platforms/rocm.py |
| 1410 | commands: |
| 1411 | - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_dp4_async_eplb.sh 0.8 200 8050 |
| 1412 | |
| 1413 | # TODO(akaratza): Retire this standalone ROCm coverage after |
| 1414 | # tests/distributed/test_eplb_spec_decode.py is enabled and validated on ROCm. |
| 1415 | # The upstream parity job currently skips that entire module on ROCm. |
| 1416 | - label: ":amd: (MI300) Qwen3-Next-80B-A3B-Instruct MTP Async EPLB Accuracy" # TBD |
| 1417 | timeout_in_minutes: 45 |
| 1418 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1419 | dind: false |
| 1420 | agent_pool: mi300_4 |
| 1421 | num_gpus: 4 |
| 1422 | optional: true |
| 1423 | working_dir: "/vllm-workspace" |
| 1424 | source_file_dependencies: |
| 1425 | - vllm/model_executor/models/ |
| 1426 | - vllm/model_executor/model_loader/ |
| 1427 | - vllm/v1/spec_decode/ |
| 1428 | - vllm/distributed/eplb |
| 1429 | - vllm/model_executor/layers/fused_moe/ |
| 1430 | - vllm/model_executor/layers/quantization/ |
| 1431 | - vllm/v1/attention/backends/ |
| 1432 | - vllm/v1/attention/selector.py |
| 1433 | - .buildkite/scripts/scheduled_integration_test/ |
| 1434 | - vllm/_aiter_ops.py |
| 1435 | - vllm/platforms/rocm.py |
| 1436 | commands: |
| 1437 | - VLLM_ENGINE_READY_TIMEOUT_S=1800 bash .buildkite/scripts/scheduled_integration_test/qwen3_next_mtp_async_eplb.sh 0.8 1319 8040 |
| 1438 | |
| 1439 | - label: ":amd: (MI300) LM Eval Large Models" # TBD |
| 1440 | timeout_in_minutes: 65 |
| 1441 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1442 | dind: false |
| 1443 | agent_pool: mi300_8 |
| 1444 | optional: true |
| 1445 | num_gpus: 8 |
| 1446 | working_dir: "/vllm-workspace/tests" |
| 1447 | source_file_dependencies: |
| 1448 | - vllm/model_executor/models/ |
| 1449 | - vllm/model_executor/model_loader/ |
| 1450 | - vllm/model_executor/layers/quantization/ |
| 1451 | - vllm/v1/attention/backends/ |
| 1452 | - vllm/v1/attention/selector.py |
| 1453 | - vllm/model_executor/layers/layernorm.py |
| 1454 | - csrc/ |
| 1455 | - vllm/_aiter_ops.py |
| 1456 | - vllm/platforms/rocm.py |
| 1457 | - tests/evals/ |
| 1458 | commands: |
| 1459 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 1460 | - export PYTORCH_ROCM_ARCH=gfx942 |
| 1461 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-mi3xx.txt |
| 1462 | |
| 1463 | - label: ":amd: (MI300) LM Eval Large Models ROCm Harness" # TBD |
| 1464 | timeout_in_minutes: 40 |
| 1465 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1466 | dind: false |
| 1467 | agent_pool: mi300_8 |
| 1468 | optional: true |
| 1469 | num_gpus: 8 |
| 1470 | working_dir: "/vllm-workspace/.buildkite/lm-eval-harness" |
| 1471 | source_file_dependencies: |
| 1472 | - vllm/model_executor/models/ |
| 1473 | - vllm/model_executor/model_loader/ |
| 1474 | - vllm/model_executor/layers/quantization/ |
| 1475 | - vllm/v1/attention/backends/ |
| 1476 | - vllm/v1/attention/selector.py |
| 1477 | - vllm/model_executor/layers/layernorm.py |
| 1478 | - csrc/ |
| 1479 | - vllm/_aiter_ops.py |
| 1480 | - vllm/platforms/rocm.py |
| 1481 | commands: |
| 1482 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 1483 | - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large-rocm.txt --tp-size=8 --timeout=5400 |
| 1484 | |
| 1485 | - label: ":amd: (MI300) LM Eval Large Models FP8" # TBD |
| 1486 | timeout_in_minutes: 40 |
| 1487 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1488 | dind: false |
| 1489 | agent_pool: mi300_4 |
| 1490 | num_gpus: 4 |
| 1491 | optional: true |
| 1492 | working_dir: "/vllm-workspace/.buildkite/lm-eval-harness" |
| 1493 | source_file_dependencies: |
| 1494 | - csrc/ |
| 1495 | - vllm/model_executor/layers/quantization |
| 1496 | - vllm/model_executor/models/ |
| 1497 | - vllm/model_executor/model_loader/ |
| 1498 | - vllm/v1/attention/backends/ |
| 1499 | - vllm/v1/attention/selector.py |
| 1500 | - vllm/_aiter_ops.py |
| 1501 | - vllm/platforms/rocm.py |
| 1502 | commands: |
| 1503 | - export VLLM_USE_DEEP_GEMM=0 |
| 1504 | - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large-rocm-tp4.txt --tp-size=4 |
| 1505 | |
| 1506 | #---------------------------------------------------------- mi300 · kernels ----------------------------------------------------------# |
| 1507 | |
| 1508 | - label: ":amd: (MI300) Core Operation Kernels Shard %N" # TBD |
| 1509 | timeout_in_minutes: 45 |
| 1510 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1511 | dind: false |
| 1512 | agent_pool: mi300_1 |
| 1513 | parallelism: 3 |
| 1514 | working_dir: "/vllm-workspace/tests" |
| 1515 | source_file_dependencies: |
| 1516 | - csrc/ |
| 1517 | - tests/kernels/core |
| 1518 | - tests/kernels/test_top_k_per_row.py |
| 1519 | - tests/kernels/test_concat_mla_q.py |
| 1520 | - tests/kernels/test_rocm_fp32_router_gemm.py |
| 1521 | - vllm/model_executor/layers/fused_moe/router/gate_linear.py |
| 1522 | - vllm/model_executor/layers/fused_moe/router/rocm_fp32_router_gemm.py |
| 1523 | - vllm/model_executor/layers/rotary_embedding/ |
| 1524 | - vllm/_aiter_ops.py |
| 1525 | - vllm/platforms/rocm.py |
| 1526 | - tests/kernels/test_fused_qk_norm_rope_gate.py |
| 1527 | optional: true |
| 1528 | commands: |
| 1529 | - pytest -v -s kernels/core --ignore=kernels/core/test_minimax_reduce_rms.py kernels/test_fused_qk_norm_rope_gate.py kernels/test_concat_mla_q.py kernels/test_top_k_per_row.py kernels/test_rocm_fp32_router_gemm.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 1530 | |
| 1531 | - label: ":amd: (MI300) Quantization Kernels Shard %N" |
| 1532 | timeout_in_minutes: 120 |
| 1533 | parallelism: 6 |
| 1534 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1535 | dind: false |
| 1536 | agent_pool: mi300_1 |
| 1537 | optional: true |
| 1538 | working_dir: "/vllm-workspace/tests" |
| 1539 | source_file_dependencies: |
| 1540 | - csrc/quantization/ |
| 1541 | - csrc/rocm/ |
| 1542 | - vllm/model_executor/layers/quantization |
| 1543 | - tests/kernels/quantization |
| 1544 | - tests/kernels/quant_utils.py |
| 1545 | - tests/kernels/utils.py |
| 1546 | - vllm/_aiter_ops.py |
| 1547 | - vllm/kernels/aiter_ops.py |
| 1548 | - vllm/_custom_ops.py |
| 1549 | - vllm/envs.py |
| 1550 | - vllm/platforms/rocm.py |
| 1551 | - vllm/model_executor/kernels/ |
| 1552 | - vllm/v1/attention/backends/rocm_aiter_fa.py |
| 1553 | - vllm/config/ |
| 1554 | - tests/kernels/quantization/test_rocm_skinny_gemms.py |
| 1555 | commands: |
| 1556 | - pytest -v -s kernels/quantization -k "not test_nvfp4_dense_emulation" --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 1557 | |
| 1558 | - label: ":amd: (MI300) MoE Kernels Shard %N" # TBD |
| 1559 | timeout_in_minutes: 55 |
| 1560 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1561 | dind: false |
| 1562 | agent_pool: mi300_1 |
| 1563 | optional: true |
| 1564 | parallelism: 5 |
| 1565 | working_dir: "/vllm-workspace/tests" |
| 1566 | source_file_dependencies: |
| 1567 | - csrc/quantization/cutlass_w8a8/moe/ |
| 1568 | - csrc/moe/ |
| 1569 | - csrc/rocm/ |
| 1570 | - tests/kernels/moe |
| 1571 | - tests/kernels/utils.py |
| 1572 | - vllm/model_executor/layers/fused_moe/ |
| 1573 | - vllm/model_executor/layers/quantization/mxfp4.py |
| 1574 | - vllm/model_executor/layers/quantization/utils/quant_utils.py |
| 1575 | - vllm/distributed/device_communicators/ |
| 1576 | - vllm/envs.py |
| 1577 | - vllm/config |
| 1578 | - vllm/_aiter_ops.py |
| 1579 | - vllm/platforms/rocm.py |
| 1580 | commands: |
| 1581 | - pytest -v -s kernels/moe --ignore=kernels/moe/test_modular_oai_triton_moe.py --ignore=kernels/moe/test_modular_kernel_combinations.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 1582 | - pytest -v -s kernels/moe/test_modular_oai_triton_moe.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 1583 | - VLLM_ROCM_USE_AITER=1 VLLM_ROCM_USE_AITER_MOE=1 pytest -v -s kernels/moe/test_modular_kernel_combinations.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 1584 | |
| 1585 | - label: ":amd: (MI300) vLLM IR" # TBD |
| 1586 | timeout_in_minutes: 50 |
| 1587 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1588 | dind: false |
| 1589 | agent_pool: mi300_1 |
| 1590 | optional: true |
| 1591 | working_dir: "/vllm-workspace/" |
| 1592 | source_file_dependencies: |
| 1593 | - vllm/ir |
| 1594 | - vllm/kernels |
| 1595 | - vllm/_aiter_ops.py |
| 1596 | - vllm/platforms/rocm.py |
| 1597 | - tests/ir/ |
| 1598 | - tests/kernels/ir/ |
| 1599 | commands: |
| 1600 | - pytest -v -s tests/ir |
| 1601 | - pytest -v -s tests/kernels/ir |
| 1602 | |
| 1603 | - label: ":amd: (MI300) Attention Kernels Shard %N" # TBD |
| 1604 | timeout_in_minutes: 90 |
| 1605 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1606 | dind: false |
| 1607 | agent_pool: mi300_1 |
| 1608 | optional: true |
| 1609 | parallelism: 2 |
| 1610 | working_dir: "/vllm-workspace/tests" |
| 1611 | source_file_dependencies: |
| 1612 | - csrc/attention/ |
| 1613 | - vllm/v1/attention |
| 1614 | - vllm/model_executor/layers/attention |
| 1615 | - tests/kernels/attention |
| 1616 | - vllm/_aiter_ops.py |
| 1617 | - vllm/envs.py |
| 1618 | - vllm/platforms/rocm.py |
| 1619 | - vllm/utils/flashinfer.py |
| 1620 | commands: |
| 1621 | - pytest -v -s kernels/attention --ignore=kernels/attention/test_triton_unified_attention_diffkv.py --ignore=kernels/attention/test_rocm_aiter_mla_decode.py --ignore=kernels/attention/test_rocm_aiter_mla_decode_metadata.py --ignore=kernels/attention/test_rocm_aiter_mla_fp8_support.py --ignore=kernels/attention/test_rocm_aiter_mla_op_registration.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 1622 | |
| 1623 | - label: ":amd: (MI300) MiniMax Reduce RMS Kernels" # TBD |
| 1624 | timeout_in_minutes: 45 |
| 1625 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1626 | dind: false |
| 1627 | agent_pool: mi300_2 |
| 1628 | num_gpus: 2 |
| 1629 | working_dir: "/vllm-workspace/tests" |
| 1630 | source_file_dependencies: |
| 1631 | - vllm/model_executor/layers/minimax_rms_norm/ |
| 1632 | - vllm/distributed/ |
| 1633 | - vllm/_aiter_ops.py |
| 1634 | - vllm/envs.py |
| 1635 | - vllm/platforms/rocm.py |
| 1636 | - tests/kernels/core/test_minimax_reduce_rms.py |
| 1637 | - tests/utils.py |
| 1638 | - csrc/libtorch_stable/minimax_reduce_rms_kernel.cu |
| 1639 | - csrc/libtorch_stable/minimax_reduce_rms_kernel.h |
| 1640 | - csrc/libtorch_stable/ops.h |
| 1641 | - csrc/libtorch_stable/torch_bindings.cpp |
| 1642 | - vllm/model_executor/layers/mamba/linear/minimax_linear_attn.py |
| 1643 | optional: true |
| 1644 | commands: |
| 1645 | - pytest -v -s kernels/core/test_minimax_reduce_rms.py |
| 1646 | |
| 1647 | - label: ":amd: (MI300) Kimi K3" # TBD |
| 1648 | timeout_in_minutes: 65 |
| 1649 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1650 | dind: false |
| 1651 | agent_pool: mi300_1 |
| 1652 | optional: true |
| 1653 | working_dir: "/vllm-workspace/tests" |
| 1654 | source_file_dependencies: |
| 1655 | - vllm/third_party/flash_linear_attention/ops/kda.py |
| 1656 | - vllm/third_party/flash_linear_attention/ops/chunk_delta_h.py |
| 1657 | - vllm/third_party/flash_linear_attention/ops/l2norm.py |
| 1658 | - vllm/models/kimi_k3/nvidia/kda.py |
| 1659 | - vllm/models/kimi_k3/nvidia/kda_metadata.py |
| 1660 | - vllm/models/kimi_k3/nvidia/ops/third_party/kda/ |
| 1661 | - vllm/models/kimi_k3/amd/ |
| 1662 | - CMakeLists.txt |
| 1663 | - csrc/libtorch_stable/kimi_k3/fused_kda_decode_kernel_rocm.cu |
| 1664 | - csrc/libtorch_stable/ops.h |
| 1665 | - csrc/libtorch_stable/torch_bindings.cpp |
| 1666 | - tests/models/kimi_k3/test_amd_attn_res.py |
| 1667 | - tests/models/kimi_k3/test_amd_kda_decode.py |
| 1668 | - tests/models/kimi_k3/test_kda.py |
| 1669 | - tests/models/kimi_k3/test_kda_metadata.py |
| 1670 | - vllm/_custom_ops.py |
| 1671 | - vllm/platforms/rocm.py |
| 1672 | commands: |
| 1673 | # Run the ROCm AttnRes/KDA decode counterparts plus existing KDA coverage. |
| 1674 | - pytest -v -s models/kimi_k3/test_amd_attn_res.py models/kimi_k3/test_amd_kda_decode.py models/kimi_k3/test_kda.py models/kimi_k3/test_kda_metadata.py |
| 1675 | |
| 1676 | - label: ":amd: (MI355) Kimi K3" # TBD |
| 1677 | timeout_in_minutes: 60 |
| 1678 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 1679 | dind: false |
| 1680 | agent_pool: mi355_1 |
| 1681 | num_gpus: 1 |
| 1682 | optional: true |
| 1683 | working_dir: "/vllm-workspace/tests" |
| 1684 | source_file_dependencies: |
| 1685 | - vllm/models/kimi_k3/ |
| 1686 | - csrc/libtorch_stable/kimi_k3/ |
| 1687 | - tests/models/kimi_k3/ |
| 1688 | - vllm/platforms/rocm.py |
| 1689 | commands: |
| 1690 | # Mirrors the ":amd: (MI355) Kimi K3" step generated from |
| 1691 | # .buildkite/test_areas/models_basic.yaml. The ROCm KDA/AttnRes kernels are |
| 1692 | # gated on gfx950, so this has to be MI355. The two kernels/ files the NVIDIA |
| 1693 | # step runs are CUDA-only (they import cuda.bindings.driver), so they are |
| 1694 | # deliberately left out here. |
| 1695 | - pytest -v -s models/kimi_k3 |
| 1696 | |
| 1697 | - label: ":amd: (MI355) Kimi K3 Multi-GPU" # TBD |
| 1698 | timeout_in_minutes: 30 |
| 1699 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 1700 | dind: false |
| 1701 | agent_pool: mi355_4 |
| 1702 | num_gpus: 4 |
| 1703 | optional: true |
| 1704 | working_dir: "/vllm-workspace/tests" |
| 1705 | source_file_dependencies: |
| 1706 | - vllm/models/kimi_k3/ |
| 1707 | - tests/models/kimi_k3/ |
| 1708 | - vllm/platforms/rocm.py |
| 1709 | commands: |
| 1710 | # The single-GPU step above skips every @multi_gpu_test case. `-m "distributed"` |
| 1711 | # selects exactly those: multi_gpu_test() applies pytest.mark.distributed |
| 1712 | # alongside its skipif (tests/utils.py:2080), so this picks up a new |
| 1713 | # multi-GPU test anywhere in the directory without editing this list. |
| 1714 | # Four GPUs run the two tp4 cases in test_amd_latent_moe_runner.py; the tp8 |
| 1715 | # case and the 8/16-GPU cases in test_latent_moe_tail.py still skip. |
| 1716 | - pytest -v -s -m "distributed" models/kimi_k3 |
| 1717 | |
| 1718 | - label: ":amd: (MI300) Mamba Kernels" # TBD |
| 1719 | timeout_in_minutes: 55 |
| 1720 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1721 | dind: false |
| 1722 | agent_pool: mi300_1 |
| 1723 | optional: true |
| 1724 | working_dir: "/vllm-workspace/tests" |
| 1725 | source_file_dependencies: |
| 1726 | - csrc/mamba/ |
| 1727 | - tests/kernels/mamba |
| 1728 | - vllm/model_executor/layers/mamba/gdn |
| 1729 | - vllm/model_executor/layers/mamba/ops |
| 1730 | - vllm/third_party/flash_linear_attention/ |
| 1731 | - vllm/platforms/rocm.py |
| 1732 | commands: |
| 1733 | - pytest -v -s kernels/mamba |
| 1734 | |
| 1735 | - label: ":amd: (MI300) DeepEP FP8 MoE Kernels" # TBD |
| 1736 | timeout_in_minutes: 55 |
| 1737 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1738 | dind: false |
| 1739 | agent_pool: mi300_2 |
| 1740 | num_gpus: 2 |
| 1741 | optional: true |
| 1742 | working_dir: "/vllm-workspace/tests" |
| 1743 | source_file_dependencies: |
| 1744 | - csrc/moe/ |
| 1745 | - csrc/quantization/w8a8/cutlass/moe/ |
| 1746 | - vllm/model_executor/layers/fused_moe/ |
| 1747 | - tests/kernels/moe/test_deepep_moe.py |
| 1748 | - vllm/_aiter_ops.py |
| 1749 | - vllm/platforms/rocm.py |
| 1750 | - vllm/envs.py |
| 1751 | commands: |
| 1752 | - pytest -v -s kernels/moe/test_deepep_moe.py |
| 1753 | |
| 1754 | #----------------------------------------------------------- mi300 · lora ------------------------------------------------------------# |
| 1755 | |
| 1756 | - label: ":amd: (MI300) LoRA TP (Distributed) %N" |
| 1757 | timeout_in_minutes: 65 |
| 1758 | parallelism: 4 |
| 1759 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1760 | dind: false |
| 1761 | agent_pool: mi300_4 |
| 1762 | num_gpus: 4 |
| 1763 | working_dir: "/vllm-workspace/tests" |
| 1764 | source_file_dependencies: |
| 1765 | - vllm/lora |
| 1766 | - vllm/model_executor/layers/fused_moe/ |
| 1767 | - tests/lora |
| 1768 | - vllm/platforms/rocm.py |
| 1769 | optional: true |
| 1770 | commands: |
| 1771 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 1772 | - case "$$BUILDKITE_PARALLEL_JOB" in 0|1|2|3) ;; *) echo "unexpected BUILDKITE_PARALLEL_JOB=$$BUILDKITE_PARALLEL_JOB" && exit 1;; esac |
| 1773 | - if [ "$$BUILDKITE_PARALLEL_JOB" = "3" ]; then pytest -v -s -x lora/test_chatglm3_tp.py; fi |
| 1774 | - if [ "$$BUILDKITE_PARALLEL_JOB" = "1" ]; then pytest -v -s -x lora/test_llama_tp.py; fi |
| 1775 | - if [ "$$BUILDKITE_PARALLEL_JOB" = "3" ]; then pytest -v -s -x lora/test_qwen3_with_multi_loras.py; fi |
| 1776 | - if [ "$$BUILDKITE_PARALLEL_JOB" = "2" ]; then pytest -v -s -x lora/test_olmoe_tp.py; fi |
| 1777 | - if [ "$$BUILDKITE_PARALLEL_JOB" = "0" ]; then pytest -v -s -x lora/test_gptoss_tp.py; fi |
| 1778 | - if [ "$$BUILDKITE_PARALLEL_JOB" = "1" ]; then pytest -v -s -x lora/test_qwen35_densemodel_lora.py; fi |
| 1779 | - if [ "$$BUILDKITE_PARALLEL_JOB" = "2" ]; then pytest -v -s -x lora/test_gemma4_tp.py; fi |
| 1780 | - if [ "$$BUILDKITE_PARALLEL_JOB" = "0" ]; then pytest -v -s -x lora/test_sequence_classification.py::test_batched_loras_tp; fi |
| 1781 | |
| 1782 | #------------------------------------------------------ mi300 · model_executor -------------------------------------------------------# |
| 1783 | |
| 1784 | - label: ":amd: (MI300) Model Executor" # TBD |
| 1785 | timeout_in_minutes: 75 |
| 1786 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1787 | dind: false |
| 1788 | agent_pool: mi300_1 |
| 1789 | optional: true |
| 1790 | working_dir: "/vllm-workspace/tests" |
| 1791 | source_file_dependencies: |
| 1792 | - vllm/engine/arg_utils.py |
| 1793 | - vllm/config/model.py |
| 1794 | - vllm/model_executor |
| 1795 | - tests/model_executor |
| 1796 | - tests/entrypoints/openai/completion/test_tensorizer_entrypoint.py |
| 1797 | - vllm/_aiter_ops.py |
| 1798 | - vllm/platforms/rocm.py |
| 1799 | - vllm/model_executor/warmup |
| 1800 | - tests/model_executor/test_jit_warmup.py |
| 1801 | - tests/model_executor/test_jit_warmup_cutedsl_launcher.py |
| 1802 | - tests/model_executor/test_jit_warmup_triton_launcher.py |
| 1803 | commands: |
| 1804 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 1805 | - export PYTHONFAULTHANDLER=1 |
| 1806 | - pytest -v -s model_executor -m '(not slow_test)' --timeout=900 --timeout-method=thread |
| 1807 | - pytest -v -s entrypoints/openai/completion/test_tensorizer_entrypoint.py --timeout=900 --timeout-method=thread |
| 1808 | |
| 1809 | #---------------------------------------------------- mi300 · model_runner_v2 -------------------------------------------------------# |
| 1810 | |
| 1811 | - label: ":amd: (MI300) Model Runner V2 Core" # TBD |
| 1812 | timeout_in_minutes: 55 |
| 1813 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1814 | dind: false |
| 1815 | agent_pool: mi300_1 |
| 1816 | optional: true |
| 1817 | working_dir: "/vllm-workspace/tests" |
| 1818 | source_file_dependencies: |
| 1819 | - vllm/v1/worker/gpu/ |
| 1820 | - vllm/v1/worker/gpu_worker.py |
| 1821 | - vllm/v1/core/sched/ |
| 1822 | - vllm/v1/attention/ |
| 1823 | - tests/v1/engine/test_llm_engine.py |
| 1824 | - tests/v1/e2e/ |
| 1825 | - tests/entrypoints/llm/test_struct_output_generate.py |
| 1826 | - vllm/platforms/rocm.py |
| 1827 | commands: |
| 1828 | - set -x |
| 1829 | - export VLLM_USE_V2_MODEL_RUNNER=1 |
| 1830 | - pytest -v -s v1/engine/test_llm_engine.py -k "not test_engine_metrics" |
| 1831 | - pytest -v -s v1/e2e/general/test_async_scheduling.py -k "not ngram" |
| 1832 | - pytest -v -s v1/e2e/general/test_context_length.py |
| 1833 | - pytest -v -s v1/e2e/general/test_min_tokens.py |
| 1834 | - pytest -v -s entrypoints/llm/test_struct_output_generate.py -k "xgrammar and not speculative_config6 and not speculative_config7 and not speculative_config8 and not speculative_config0" |
| 1835 | |
| 1836 | - label: ":amd: (MI300) Model Runner V2 Examples" # TBD |
| 1837 | timeout_in_minutes: 40 |
| 1838 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1839 | dind: false |
| 1840 | agent_pool: mi300_1 |
| 1841 | optional: true |
| 1842 | working_dir: "/vllm-workspace/examples" |
| 1843 | source_file_dependencies: |
| 1844 | - vllm/v1/worker/gpu/ |
| 1845 | - vllm/v1/core/sched/ |
| 1846 | - vllm/v1/worker/gpu_worker.py |
| 1847 | - examples/basic/offline_inference/ |
| 1848 | - examples/generate/multimodal/ |
| 1849 | - examples/features/ |
| 1850 | - examples/pooling/embed/vision_embedding_offline.py |
| 1851 | - examples/features/tensorize_vllm_model.py |
| 1852 | - examples/deployment/llm_engine_example.py |
| 1853 | - vllm/platforms/rocm.py |
| 1854 | commands: |
| 1855 | - set -x |
| 1856 | - export VLLM_USE_V2_MODEL_RUNNER=1 |
| 1857 | - pip install tensorizer |
| 1858 | - python3 basic/offline_inference/chat.py |
| 1859 | - python3 basic/offline_inference/generate.py --model facebook/opt-125m |
| 1860 | - python3 generate/multimodal/audio_language_offline.py --seed 0 |
| 1861 | - python3 generate/multimodal/vision_language_offline.py --seed 0 |
| 1862 | - python3 generate/multimodal/vision_language_multi_image_offline.py --seed 0 |
| 1863 | - python3 generate/multimodal/encoder_decoder_multimodal_offline.py --model-type whisper --seed 0 |
| 1864 | - python3 pooling/embed/vision_embedding_offline.py --seed 0 |
| 1865 | - python3 features/automatic_prefix_caching/prefix_caching_offline.py |
| 1866 | - python3 deployment/llm_engine_example.py |
| 1867 | - python3 features/tensorize_vllm_model.py --model facebook/opt-125m serialize --serialized-directory /tmp/ --suffix v1 && python3 features/tensorize_vllm_model.py --model facebook/opt-125m deserialize --path-to-tensors /tmp/vllm/facebook/opt-125m/v1/model.tensors |
| 1868 | - python3 features/speculative_decoding/spec_decode_offline.py --test --method eagle --num_spec_tokens 3 --dataset-name hf --dataset-path philschmid/mt-bench --num-prompts 80 --temp 0 --top-p 1.0 --top-k -1 --tp 1 --enable-chunked-prefill --max-model-len 2048 |
| 1869 | - python3 features/speculative_decoding/spec_decode_offline.py --test --method eagle3 --num_spec_tokens 3 --dataset-name hf --dataset-path philschmid/mt-bench --num-prompts 80 --temp 0 --top-p 1.0 --top-k -1 --tp 1 --enable-chunked-prefill --max-model-len 1536 |
| 1870 | |
| 1871 | - label: ":amd: (MI300) Model Runner V2 Distributed" # TBD |
| 1872 | timeout_in_minutes: 45 |
| 1873 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1874 | dind: false |
| 1875 | agent_pool: mi300_2 |
| 1876 | num_gpus: 2 |
| 1877 | optional: true |
| 1878 | working_dir: "/vllm-workspace/tests" |
| 1879 | source_file_dependencies: |
| 1880 | - vllm/v1/worker/gpu/ |
| 1881 | - vllm/v1/worker/gpu_worker.py |
| 1882 | - tests/basic_correctness/models/test_basic_correctness.py |
| 1883 | - tests/v1/distributed/test_async_llm_dp.py |
| 1884 | - tests/v1/distributed/test_eagle_dp.py |
| 1885 | - vllm/platforms/rocm.py |
| 1886 | commands: |
| 1887 | - set -x |
| 1888 | - export VLLM_USE_V2_MODEL_RUNNER=1 |
| 1889 | - TARGET_TEST_SUITE=MI300 pytest -v -s basic_correctness/models/test_basic_correctness.py -m 'distributed(num_gpus=2)' -k "not ray and not True" |
| 1890 | - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_async_llm_dp.py -k "not ray" |
| 1891 | - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_eagle_dp.py |
| 1892 | |
| 1893 | - label: ":amd: (MI300) Model Runner V2 Pipeline Parallelism" # TBD |
| 1894 | timeout_in_minutes: 65 |
| 1895 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1896 | dind: false |
| 1897 | agent_pool: mi300_4 |
| 1898 | num_gpus: 4 |
| 1899 | optional: true |
| 1900 | working_dir: "/vllm-workspace/tests" |
| 1901 | source_file_dependencies: |
| 1902 | - vllm/v1/worker/gpu/ |
| 1903 | - vllm/v1/worker/gpu_worker.py |
| 1904 | - tests/distributed/test_pipeline_parallel.py |
| 1905 | - tests/distributed/test_pp_cudagraph.py |
| 1906 | - vllm/platforms/rocm.py |
| 1907 | commands: |
| 1908 | - set -x |
| 1909 | - export VLLM_USE_V2_MODEL_RUNNER=1 |
| 1910 | - pytest -v -s distributed/test_pipeline_parallel.py -k "not ray" |
| 1911 | - pytest -v -s distributed/test_pp_cudagraph.py -k "not ray" |
| 1912 | |
| 1913 | - label: ":amd: (MI300) Model Runner V2 Spec Decode" # TBD |
| 1914 | timeout_in_minutes: 90 |
| 1915 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1916 | dind: false |
| 1917 | agent_pool: mi300_1 |
| 1918 | optional: true |
| 1919 | working_dir: "/vllm-workspace/tests" |
| 1920 | source_file_dependencies: |
| 1921 | - vllm/v1/worker/gpu/ |
| 1922 | - vllm/v1/worker/gpu_worker.py |
| 1923 | - tests/v1/spec_decode/test_max_len.py |
| 1924 | - tests/v1/spec_decode/test_rejection_sampler_utils.py |
| 1925 | - tests/v1/spec_decode/test_synthetic_rejection_sampler_utils.py |
| 1926 | - tests/v1/e2e/spec_decode/ |
| 1927 | - vllm/platforms/rocm.py |
| 1928 | commands: |
| 1929 | - set -x |
| 1930 | - export VLLM_USE_V2_MODEL_RUNNER=1 |
| 1931 | - pytest -v -s v1/spec_decode/test_max_len.py -k "eagle or mtp" |
| 1932 | - pytest -v -s v1/spec_decode/test_rejection_sampler_utils.py |
| 1933 | - pytest -v -s v1/spec_decode/test_synthetic_rejection_sampler_utils.py |
| 1934 | - pytest -v -s v1/e2e/spec_decode/eagle/ |
| 1935 | - pytest -v -s v1/e2e/spec_decode/speculators/ |
| 1936 | - pytest -v -s v1/e2e/spec_decode/mtp/ |
| 1937 | |
| 1938 | #------------------------------------------------------ mi300 · models / basic -------------------------------------------------------# |
| 1939 | |
| 1940 | - label: ":amd: (MI300) Basic Models (Initialization)" # TBD |
| 1941 | timeout_in_minutes: 30 |
| 1942 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1943 | dind: false |
| 1944 | agent_pool: mi300_1 |
| 1945 | working_dir: "/vllm-workspace/tests" |
| 1946 | source_file_dependencies: |
| 1947 | - vllm/ |
| 1948 | - tests/models/test_initialization.py |
| 1949 | - tests/models/registry.py |
| 1950 | optional: true |
| 1951 | commands: |
| 1952 | - pytest -v -s models/test_initialization.py::test_can_initialize_small_subset |
| 1953 | |
| 1954 | #----------------------------------------------------- mi300 · models / language -----------------------------------------------------# |
| 1955 | |
| 1956 | - label: ":amd: (MI300) Language Models (MTEB)" # TBD |
| 1957 | timeout_in_minutes: 70 |
| 1958 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1959 | dind: false |
| 1960 | agent_pool: mi300_1 |
| 1961 | optional: true |
| 1962 | working_dir: "/vllm-workspace/tests" |
| 1963 | source_file_dependencies: |
| 1964 | - vllm/ |
| 1965 | - tests/models/language/pooling_mteb_test |
| 1966 | commands: |
| 1967 | - pytest -v -s models/language/pooling_mteb_test |
| 1968 | |
| 1969 | - label: ":amd: (MI300) Language Models (Extended Pooling) Shard %N" |
| 1970 | timeout_in_minutes: 95 |
| 1971 | parallelism: 4 |
| 1972 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1973 | dind: false |
| 1974 | agent_pool: mi300_1 |
| 1975 | optional: true |
| 1976 | working_dir: "/vllm-workspace/tests" |
| 1977 | source_file_dependencies: |
| 1978 | - vllm/ |
| 1979 | - tests/models/language/pooling |
| 1980 | commands: |
| 1981 | - pytest -v -s models/language/pooling -m 'not core_model' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB |
| 1982 | |
| 1983 | - label: ":amd: (MI300) Language Models (Extended Generation)" # TBD |
| 1984 | timeout_in_minutes: 105 |
| 1985 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 1986 | dind: false |
| 1987 | agent_pool: mi300_1 |
| 1988 | optional: true |
| 1989 | working_dir: "/vllm-workspace/tests" |
| 1990 | source_file_dependencies: |
| 1991 | - vllm/ |
| 1992 | - tests/models/language/generation |
| 1993 | commands: |
| 1994 | - MAMBA_FORCE_BUILD=TRUE uv pip install --system --no-build-isolation 'git+https://github.com/AndreasKaratzas/mamba@fix-rocm-7.0-warp-size-constexpr' |
| 1995 | - CAUSAL_CONV1D_FORCE_BUILD=TRUE uv pip install --system --no-build-isolation 'git+https://github.com/Dao-AILab/causal-conv1d@v1.6.0' |
| 1996 | - pytest -v -s models/language/generation -m '(not core_model) and (not hybrid_model)' |
| 1997 | |
| 1998 | #---------------------------------------------------- mi300 · models / multimodal ----------------------------------------------------# |
| 1999 | |
| 2000 | - label: ":amd: (MI300) Multimodal Models (Extended Generation 1)" # TBD |
| 2001 | timeout_in_minutes: 100 |
| 2002 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2003 | dind: false |
| 2004 | agent_pool: mi300_1 |
| 2005 | optional: true |
| 2006 | working_dir: "/vllm-workspace/tests" |
| 2007 | source_file_dependencies: |
| 2008 | - vllm/ |
| 2009 | - tests/models/multimodal/generation |
| 2010 | - tests/models/multimodal/test_mapping.py |
| 2011 | commands: |
| 2012 | - pytest -v -s models/multimodal/generation -m 'not core_model' --ignore models/multimodal/generation/test_common.py |
| 2013 | - pytest -v -s models/multimodal/test_mapping.py |
| 2014 | |
| 2015 | - label: ":amd: (MI300) Multimodal Models (Extended Generation 2) Shard %N" |
| 2016 | timeout_in_minutes: 100 |
| 2017 | parallelism: 4 |
| 2018 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2019 | dind: false |
| 2020 | agent_pool: mi300_1 |
| 2021 | optional: true |
| 2022 | working_dir: "/vllm-workspace/tests" |
| 2023 | source_file_dependencies: |
| 2024 | - vllm/ |
| 2025 | - tests/models/multimodal/generation |
| 2026 | commands: |
| 2027 | - pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=0) and not core_model' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB |
| 2028 | |
| 2029 | |
| 2030 | - label: ":amd: (MI300) Multimodal Models (Extended Generation 3)" # TBD |
| 2031 | timeout_in_minutes: 90 |
| 2032 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2033 | dind: false |
| 2034 | agent_pool: mi300_1 |
| 2035 | optional: true |
| 2036 | working_dir: "/vllm-workspace/tests" |
| 2037 | source_file_dependencies: |
| 2038 | - vllm/ |
| 2039 | - tests/models/multimodal/generation |
| 2040 | commands: |
| 2041 | - pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=1) and not core_model' |
| 2042 | |
| 2043 | - label: ":amd: (MI300) Multimodal Models (Standard) 3: llava + qwen2_vl" # TBD |
| 2044 | timeout_in_minutes: 55 |
| 2045 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2046 | dind: false |
| 2047 | agent_pool: mi300_1 |
| 2048 | optional: true |
| 2049 | working_dir: "/vllm-workspace/tests" |
| 2050 | source_file_dependencies: |
| 2051 | - vllm/ |
| 2052 | - tests/models/multimodal/generation |
| 2053 | - tests/models/multimodal/test_mapping.py |
| 2054 | - tests/models/multimodal |
| 2055 | commands: |
| 2056 | - pytest -v -s models/multimodal/generation/test_common.py -m core_model -k "not qwen2 and not qwen3 and not gemma" |
| 2057 | - pytest -v -s models/multimodal/generation/test_qwen2_vl.py -m core_model |
| 2058 | |
| 2059 | - label: ":amd: (MI300) Multimodal Processor Shard %N" # TBD |
| 2060 | timeout_in_minutes: 115 |
| 2061 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2062 | dind: false |
| 2063 | agent_pool: mi300_1 |
| 2064 | optional: true |
| 2065 | parallelism: 4 |
| 2066 | working_dir: "/vllm-workspace/tests" |
| 2067 | source_file_dependencies: |
| 2068 | - vllm/ |
| 2069 | - tests/models/multimodal |
| 2070 | - tests/models/registry.py |
| 2071 | commands: |
| 2072 | - pytest -v -s models/multimodal/processing/test_tensor_schema.py --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB |
| 2073 | |
| 2074 | - label: ":amd: (MI300) Multimodal Models (Extended Pooling)" # TBD |
| 2075 | timeout_in_minutes: 75 |
| 2076 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2077 | dind: false |
| 2078 | agent_pool: mi300_1 |
| 2079 | optional: true |
| 2080 | working_dir: "/vllm-workspace/tests" |
| 2081 | source_file_dependencies: |
| 2082 | - vllm/ |
| 2083 | - tests/models/multimodal/pooling |
| 2084 | commands: |
| 2085 | - pytest -v -s models/multimodal/pooling -m 'not core_model' |
| 2086 | |
| 2087 | #----------------------------------------------------- mi300 · models / quantized -----------------------------------------------------# |
| 2088 | |
| 2089 | - label: ":amd: (MI300) Quantized Models" # TBD |
| 2090 | timeout_in_minutes: 90 |
| 2091 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2092 | dind: false |
| 2093 | agent_pool: mi300_1 |
| 2094 | working_dir: "/vllm-workspace/tests" |
| 2095 | source_file_dependencies: |
| 2096 | - vllm/model_executor/layers/quantization |
| 2097 | - vllm/_aiter_ops.py |
| 2098 | - vllm/platforms/rocm.py |
| 2099 | - tests/models/quantization |
| 2100 | - vllm/model_executor/model_loader/ |
| 2101 | optional: true |
| 2102 | commands: |
| 2103 | - unset VLLM_USE_V2_MODEL_RUNNER |
| 2104 | - pytest -v -s models/quantization |
| 2105 | |
| 2106 | #-------------------------------------------------- mi300 · models / transformers ---------------------------------------------------# |
| 2107 | |
| 2108 | # - label: ":amd: (MI300) Transformers Nightly Models (Initialization) Shard %N" # TBD |
| 2109 | # timeout_in_minutes: 65 |
| 2110 | # mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2111 | # dind: false |
| 2112 | # agent_pool: mi300_1 |
| 2113 | # parallelism: 6 |
| 2114 | # optional: true |
| 2115 | # working_dir: "/vllm-workspace/" |
| 2116 | # source_file_dependencies: |
| 2117 | # - vllm/model_executor/models/ |
| 2118 | # - vllm/model_executor/model_loader/ |
| 2119 | # - vllm/multimodal/ |
| 2120 | # - vllm/model_executor/layers/ |
| 2121 | # - vllm/v1/attention/backends/ |
| 2122 | # - vllm/v1/attention/selector.py |
| 2123 | # - vllm/_aiter_ops.py |
| 2124 | # - vllm/platforms/rocm.py |
| 2125 | # - tests/models/ |
| 2126 | # commands: |
| 2127 | # - pip install --upgrade git+https://github.com/huggingface/transformers |
| 2128 | # - pytest -v -s tests/models/test_initialization.py --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB |
| 2129 | |
| 2130 | # - label: ":amd: (MI300) Transformers Nightly Models (Processing) Shard %N" # TBD |
| 2131 | # timeout_in_minutes: 145 |
| 2132 | # mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2133 | # dind: false |
| 2134 | # agent_pool: mi300_1 |
| 2135 | # parallelism: 8 |
| 2136 | # optional: true |
| 2137 | # working_dir: "/vllm-workspace/" |
| 2138 | # source_file_dependencies: |
| 2139 | # - vllm/model_executor/models/ |
| 2140 | # - vllm/model_executor/model_loader/ |
| 2141 | # - vllm/multimodal/ |
| 2142 | # - vllm/model_executor/layers/ |
| 2143 | # - vllm/v1/attention/backends/ |
| 2144 | # - vllm/v1/attention/selector.py |
| 2145 | # - vllm/_aiter_ops.py |
| 2146 | # - vllm/platforms/rocm.py |
| 2147 | # - tests/models/ |
| 2148 | # commands: |
| 2149 | # - pip install --upgrade git+https://github.com/huggingface/transformers |
| 2150 | # - pytest -v -s tests/models/multimodal/processing/ --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB |
| 2151 | |
| 2152 | # - label: ":amd: (MI300) Transformers Nightly Models (Single)" # TBD |
| 2153 | # timeout_in_minutes: 50 |
| 2154 | # mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2155 | # dind: false |
| 2156 | # agent_pool: mi300_1 |
| 2157 | # optional: true |
| 2158 | # working_dir: "/vllm-workspace/" |
| 2159 | # source_file_dependencies: |
| 2160 | # - vllm/model_executor/models/ |
| 2161 | # - vllm/model_executor/model_loader/ |
| 2162 | # - vllm/multimodal/ |
| 2163 | # - vllm/model_executor/layers/ |
| 2164 | # - vllm/v1/attention/backends/ |
| 2165 | # - vllm/v1/attention/selector.py |
| 2166 | # - vllm/_aiter_ops.py |
| 2167 | # - vllm/platforms/rocm.py |
| 2168 | # - tests/models/ |
| 2169 | # - examples/ |
| 2170 | # commands: |
| 2171 | # - pip install --upgrade git+https://github.com/huggingface/transformers |
| 2172 | # - pytest -v -s tests/models/transformers/test_backend.py |
| 2173 | # - pytest -v -s tests/models/multimodal/test_mapping.py |
| 2174 | # - python3 examples/basic/offline_inference/chat.py |
| 2175 | # - python3 examples/generate/multimodal/vision_language_offline.py --model-type qwen2_5_vl |
| 2176 | # - VLLM_WORKER_MULTIPROC_METHOD=spawn python3 examples/generate/multimodal/audio_language_offline.py --model-type whisper |
| 2177 | |
| 2178 | #---------------------------------------------------------- mi300 · plugins ----------------------------------------------------------# |
| 2179 | |
| 2180 | - label: ":amd: (MI300) Plugin Integration" # TBD |
| 2181 | timeout_in_minutes: 50 |
| 2182 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2183 | dind: false |
| 2184 | agent_pool: mi300_2 |
| 2185 | num_gpus: 2 |
| 2186 | optional: true |
| 2187 | working_dir: "/vllm-workspace/tests" |
| 2188 | source_file_dependencies: |
| 2189 | - vllm/plugins/ |
| 2190 | - tests/plugins/ |
| 2191 | - vllm/platforms/rocm.py |
| 2192 | commands: |
| 2193 | # BEGIN: platform plugin and general plugin tests, all the code in-between runs on dummy platform |
| 2194 | # END: platform plugin tests |
| 2195 | # BEGIN: `io_processor` plugins test, all the code in between uses the `prithvi_io_processor` plugin |
| 2196 | # END: `io_processor` plugins test |
| 2197 | # BEGIN: `bge_m3_sparse io_processor` test |
| 2198 | # END: `bge_m3_sparse io_processor` test |
| 2199 | # BEGIN: `colbert_query io_processor` test |
| 2200 | # END: `colbert_query io_processor` test |
| 2201 | # BEGIN: `stat_logger` plugins test |
| 2202 | # END: `stat_logger` plugins test |
| 2203 | # BEGIN: `endpoint` plugins test |
| 2204 | # END: `endpoint` plugins test |
| 2205 | # BEGIN: other tests |
| 2206 | - pip install -e ./plugins/vllm_add_dummy_platform |
| 2207 | - pytest -v -s plugins_tests/test_platform_plugins.py |
| 2208 | - pip uninstall vllm_add_dummy_platform -y |
| 2209 | - pytest -v -s ./plugins_tests/test_io_processor_plugins.py |
| 2210 | - pip install -e ./plugins/prithvi_io_processor_plugin |
| 2211 | - pytest -v -s plugins_tests/test_terratorch_io_processor_plugins.py |
| 2212 | - pip uninstall prithvi_io_processor_plugin -y |
| 2213 | - pip install -e ./plugins/bge_m3_sparse_plugin |
| 2214 | - pytest -v -s plugins_tests/test_bge_m3_sparse_io_processor_plugins.py |
| 2215 | - pip uninstall bge_m3_sparse_plugin -y |
| 2216 | - pip install -e ./plugins/colbert_query_plugin |
| 2217 | - pytest -v -s plugins_tests/test_colbert_query_io_processor_plugins.py |
| 2218 | - pip uninstall colbert_query_plugin -y |
| 2219 | - pip install -e ./plugins/vllm_add_dummy_stat_logger |
| 2220 | - pytest -v -s plugins_tests/test_stats_logger_plugins.py |
| 2221 | - pip uninstall dummy_stat_logger -y |
| 2222 | - pip install -e ./plugins/vllm_add_dummy_endpoint_plugin |
| 2223 | - pytest -v -s plugins_tests/test_endpoint_plugins.py |
| 2224 | - pip uninstall vllm_add_dummy_endpoint_plugin -y |
| 2225 | - pytest -v -s plugins_tests/test_scheduler_plugins.py |
| 2226 | - pip install -e ./plugins/vllm_add_dummy_model |
| 2227 | - pytest -v -s distributed/test_distributed_oot.py |
| 2228 | - pytest -v -s plugins_tests/test_oot_registration_online.py |
| 2229 | - pytest -v -s plugins_tests/test_oot_registration_offline.py |
| 2230 | - pytest -v -s plugins_tests/lora_resolvers |
| 2231 | |
| 2232 | - label: ":amd: (MI300) GGUF Plugin" # TBD |
| 2233 | timeout_in_minutes: 45 |
| 2234 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2235 | dind: false |
| 2236 | agent_pool: mi300_1 |
| 2237 | optional: true |
| 2238 | working_dir: "/vllm-workspace/tests" |
| 2239 | source_file_dependencies: |
| 2240 | - vllm/model_executor/layers/quantization |
| 2241 | - tests/plugins_tests/test_gguf_plugin.py |
| 2242 | - tests/plugins_tests/gguf |
| 2243 | - vllm/platforms/rocm.py |
| 2244 | commands: |
| 2245 | - pip install "vllm-gguf-plugin >= 0.0.2" |
| 2246 | - pytest -v -s plugins_tests/gguf |
| 2247 | |
| 2248 | #------------------------------------------------------- mi300 · rust_frontend -------------------------------------------------------# |
| 2249 | |
| 2250 | - label: ":amd: (MI300) Rust Frontend OpenAI Coverage" # TBD |
| 2251 | timeout_in_minutes: 35 |
| 2252 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2253 | dind: false |
| 2254 | agent_pool: mi300_1 |
| 2255 | optional: true |
| 2256 | working_dir: "/vllm-workspace/tests" |
| 2257 | source_file_dependencies: |
| 2258 | - rust/ |
| 2259 | - vllm/benchmarks/ |
| 2260 | - vllm/entrypoints/openai/ |
| 2261 | - vllm/entrypoints/serve/ |
| 2262 | - vllm/v1/sample/ |
| 2263 | - tests/utils.py |
| 2264 | - tests/benchmarks/test_serve_cli.py |
| 2265 | - tests/entrypoints/openai/chat_completion/test_chat_completion.py |
| 2266 | - tests/entrypoints/openai/chat_completion/test_chat_logit_bias_validation.py |
| 2267 | - tests/entrypoints/launchers/test_shutdown.py |
| 2268 | - tests/entrypoints/openai/test_return_token_ids.py |
| 2269 | - tests/entrypoints/serve/instrumentator/test_uds.py |
| 2270 | - tests/v1/sample/test_logprobs_e2e.py |
| 2271 | - vllm/platforms/rocm.py |
| 2272 | - tests/entrypoints/openai/chat_completion/test_include_reasoning.py |
| 2273 | - tests/entrypoints/openai/chat_completion/test_serving_chat.py |
| 2274 | - tests/entrypoints/openai/completion/test_shutdown.py |
| 2275 | - tests/entrypoints/openai/test_uds.py |
| 2276 | commands: |
| 2277 | - export VLLM_USE_RUST_FRONTEND=1 |
| 2278 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 2279 | - pytest -v -s benchmarks/test_serve_cli.py -k "not insecure and not (test_bench_serve and not test_bench_serve_chat)" |
| 2280 | - pytest -v -s entrypoints/openai/chat_completion/test_chat_completion.py -k "not test_invalid_json_schema and not test_invalid_regex and not test_kv_transfer_prompt_token_ids_round_trip and not test_kv_transfer_prompt_token_ids_streaming" |
| 2281 | - pytest -v -s entrypoints/openai/chat_completion/test_include_reasoning.py |
| 2282 | - pytest -v -s entrypoints/openai/chat_completion/test_chat_logit_bias_validation.py -k "not multiple" |
| 2283 | - pytest -v -s entrypoints/openai/chat_completion/test_serving_chat.py -k "test_chat_per_request_metrics_follow_server_flag or test_streaming_reasoning_usage_counts_across_deltas or test_completion_tokens_details" |
| 2284 | - pytest -v -s entrypoints/launchers/test_shutdown.py -k "not engine_failure and not test_abort_timeout_exits_quickly" |
| 2285 | - pytest -v -s entrypoints/openai/test_return_token_ids.py -k "not test_comparison" |
| 2286 | - pytest -v -s entrypoints/serve/instrumentator/test_uds.py |
| 2287 | - pytest -v -s v1/sample/test_logprobs_e2e.py -k "test_prompt_logprobs_e2e_server" |
| 2288 | |
| 2289 | - label: ":amd: (MI300) Rust Frontend Serve-Admin Coverage" |
| 2290 | timeout_in_minutes: 30 |
| 2291 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2292 | dind: false |
| 2293 | agent_pool: mi300_1 |
| 2294 | optional: true |
| 2295 | working_dir: "/vllm-workspace/tests" |
| 2296 | source_file_dependencies: |
| 2297 | - rust/ |
| 2298 | - vllm/entrypoints/openai/ |
| 2299 | - vllm/entrypoints/serve/ |
| 2300 | - vllm/v1/engine/ |
| 2301 | - tests/utils.py |
| 2302 | - tests/entrypoints/serve/dev/rpc/test_collective_rpc.py |
| 2303 | - tests/entrypoints/scale_out/token_in_token_out/test_serving_tokens.py |
| 2304 | - tests/entrypoints/serve/instrumentator/test_basic.py |
| 2305 | - tests/entrypoints/serve/instrumentator/test_metrics.py |
| 2306 | - tests/entrypoints/serve/tokenize/test_tokenization.py |
| 2307 | - vllm/platforms/rocm.py |
| 2308 | commands: |
| 2309 | - export VLLM_USE_RUST_FRONTEND=1 |
| 2310 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 2311 | - PYTHONPATH=/vllm-workspace pytest -v -s entrypoints/serve/dev/rpc/test_collective_rpc.py |
| 2312 | - pytest -v -s entrypoints/serve/instrumentator/test_basic.py -k "not server_load" |
| 2313 | - pytest -v -s entrypoints/scale_out/token_in_token_out/test_serving_tokens.py -k "not stream and not lora and not test_generate_logprobs and not stop_string_workflow" |
| 2314 | - pytest -v -s entrypoints/serve/instrumentator/test_metrics.py -k "text and not show and not run_batch and not test_metrics_counts and not test_metrics_exist" |
| 2315 | - pytest -v -s entrypoints/serve/tokenize/test_tokenization.py -k "not tokenizer_info" |
| 2316 | |
| 2317 | - label: ":amd: (MI300) Rust Frontend Core Correctness" # TBD |
| 2318 | timeout_in_minutes: 25 |
| 2319 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2320 | dind: false |
| 2321 | agent_pool: mi300_1 |
| 2322 | optional: true |
| 2323 | working_dir: "/vllm-workspace/tests" |
| 2324 | source_file_dependencies: |
| 2325 | - rust/ |
| 2326 | - vllm/entrypoints/openai/ |
| 2327 | - tests/utils.py |
| 2328 | - tests/entrypoints/openai/correctness/test_lmeval.py |
| 2329 | - vllm/platforms/rocm.py |
| 2330 | commands: |
| 2331 | - export VLLM_USE_RUST_FRONTEND=1 |
| 2332 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 2333 | - pytest -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine |
| 2334 | |
| 2335 | - label: ":amd: (MI300) Rust Frontend Tool Use" # TBD |
| 2336 | timeout_in_minutes: 30 |
| 2337 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2338 | dind: false |
| 2339 | agent_pool: mi300_1 |
| 2340 | optional: true |
| 2341 | working_dir: "/vllm-workspace/tests" |
| 2342 | source_file_dependencies: |
| 2343 | - rust/ |
| 2344 | - vllm/entrypoints/openai/ |
| 2345 | - vllm/tool_parsers/ |
| 2346 | - tests/utils.py |
| 2347 | - tests/tool_use/ |
| 2348 | - vllm/platforms/rocm.py |
| 2349 | commands: |
| 2350 | - export VLLM_USE_RUST_FRONTEND=1 |
| 2351 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 2352 | - pytest -v -s tool_use --ignore=tool_use/mistral --models llama3.2 -k "not test_response_format_with_tool_choice_required and not test_parallel_tool_calls_false and not test_tool_call_and_choice" |
| 2353 | |
| 2354 | - label: ":amd: (MI300) Rust Frontend Distributed" # TBD |
| 2355 | timeout_in_minutes: 45 |
| 2356 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2357 | dind: false |
| 2358 | agent_pool: mi300_4 |
| 2359 | num_gpus: 4 |
| 2360 | optional: true |
| 2361 | working_dir: "/vllm-workspace/tests" |
| 2362 | source_file_dependencies: |
| 2363 | - rust/ |
| 2364 | - vllm/distributed/ |
| 2365 | - vllm/engine/ |
| 2366 | - vllm/executor/ |
| 2367 | - vllm/v1/engine/ |
| 2368 | - vllm/v1/worker/ |
| 2369 | - tests/utils.py |
| 2370 | - tests/v1/distributed/test_dense_dp_world_size.py |
| 2371 | - tests/v1/distributed/test_external_lb_dp.py |
| 2372 | - tests/v1/distributed/test_hybrid_lb_dp.py |
| 2373 | - tests/v1/distributed/test_internal_lb_dp.py |
| 2374 | - vllm/platforms/rocm.py |
| 2375 | commands: |
| 2376 | - export VLLM_USE_RUST_FRONTEND=1 |
| 2377 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 2378 | - TP_SIZE=1 DP_SIZE=4 pytest -v -s v1/distributed/test_dense_dp_world_size.py |
| 2379 | - VLLM_ENGINE_READY_TIMEOUT_S=1800 TP_SIZE=1 DP_SIZE=4 pytest -v -s v1/distributed/test_internal_lb_dp.py -k "not 4 and not server_info" |
| 2380 | - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_external_lb_dp.py -k "not 4 and not server_info" |
| 2381 | - TP_SIZE=1 DP_SIZE=4 pytest -v -s v1/distributed/test_hybrid_lb_dp.py -k "not 4 and not server_info" |
| 2382 | |
| 2383 | #------------------------------------------------------- mi300 · quantization --------------------------------------------------------# |
| 2384 | |
| 2385 | - label: ":amd: (MI300) Quantization Shard %N" |
| 2386 | timeout_in_minutes: 115 |
| 2387 | parallelism: 4 |
| 2388 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2389 | dind: false |
| 2390 | agent_pool: mi300_1 |
| 2391 | working_dir: "/vllm-workspace/tests" |
| 2392 | source_file_dependencies: |
| 2393 | - csrc/ |
| 2394 | - vllm/model_executor/layers/quantization |
| 2395 | - vllm/_aiter_ops.py |
| 2396 | - vllm/platforms/rocm.py |
| 2397 | - vllm/model_executor/layers/fused_moe/oracle/unquantized.py |
| 2398 | - vllm/model_executor/layers/fused_moe/unquantized_fused_moe_method.py |
| 2399 | - tests/quantization |
| 2400 | - tests/rocm/test_moe_weight_replay.py |
| 2401 | optional: true |
| 2402 | commands: |
| 2403 | # temporary install here since we need nightly, will move to requirements/test.in |
| 2404 | # after torchao 0.12 release, and pin a working version of torchao nightly here |
| 2405 | # since torchao nightly is only compatible with torch nightly currently |
| 2406 | # https://github.com/pytorch/ao/issues/2919, we'll have to skip new torchao tests for now |
| 2407 | # we can only upgrade after this is resolved |
| 2408 | # TODO(jerryzh168): resolve the above comment |
| 2409 | - unset VLLM_USE_V2_MODEL_RUNNER |
| 2410 | - uv pip install --system torchao==0.18.0 |
| 2411 | - uv pip install --system conch-triton-kernels |
| 2412 | - VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization/ --ignore quantization/test_blackwell_moe.py --ignore quantization/test_rocm_moe.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 2413 | - if [ "$$BUILDKITE_PARALLEL_JOB" = "0" ]; then pytest -v -s rocm/test_moe_weight_replay.py; fi |
| 2414 | |
| 2415 | - label: ":amd: (MI300) Quantized Fusions" # TBD |
| 2416 | timeout_in_minutes: 45 |
| 2417 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2418 | dind: false |
| 2419 | agent_pool: mi300_1 |
| 2420 | optional: true |
| 2421 | working_dir: "/vllm-workspace/tests" |
| 2422 | source_file_dependencies: |
| 2423 | - tests/fusion |
| 2424 | - vllm/model_executor/layers/fusion |
| 2425 | - vllm/model_executor/kernels/linear |
| 2426 | - vllm/model_executor/layers/quantization/compressed_tensors |
| 2427 | - vllm/model_executor/layers/quantization/modelopt.py |
| 2428 | - vllm/platforms/rocm.py |
| 2429 | commands: |
| 2430 | - pytest -v -s fusion/ |
| 2431 | |
| 2432 | #--------------------------------------------------------- mi300 · samplers ----------------------------------------------------------# |
| 2433 | |
| 2434 | - label: ":amd: (MI300) Samplers" # TBD |
| 2435 | timeout_in_minutes: 40 |
| 2436 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2437 | dind: false |
| 2438 | agent_pool: mi300_1 |
| 2439 | optional: true |
| 2440 | working_dir: "/vllm-workspace/tests" |
| 2441 | source_file_dependencies: |
| 2442 | - vllm/model_executor/layers |
| 2443 | - vllm/sampling_metadata.py |
| 2444 | - vllm/v1/sample/ |
| 2445 | - vllm/entrypoints/generate/beam_search/ |
| 2446 | - tests/samplers |
| 2447 | - tests/conftest.py |
| 2448 | - vllm/_aiter_ops.py |
| 2449 | - vllm/platforms/rocm.py |
| 2450 | - vllm/entrypoints/generate/beam_search |
| 2451 | commands: |
| 2452 | - pytest -v -s samplers -k "not test_beam_search_passes_multimodal_data" |
| 2453 | |
| 2454 | #------------------------------------------------------------ mi300 · misc ------------------------------------------------------------# |
| 2455 | |
| 2456 | - label: ":amd: (MI300) Regression" # TBD |
| 2457 | timeout_in_minutes: 25 |
| 2458 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2459 | dind: false |
| 2460 | agent_pool: mi300_1 |
| 2461 | working_dir: "/vllm-workspace/tests" |
| 2462 | source_file_dependencies: |
| 2463 | - vllm/ |
| 2464 | - tests/test_regression |
| 2465 | - vllm/config/ |
| 2466 | - vllm/distributed/ |
| 2467 | - vllm/engine/ |
| 2468 | - vllm/inputs/ |
| 2469 | - vllm/model_executor/ |
| 2470 | - vllm/multimodal/ |
| 2471 | - vllm/platforms/ |
| 2472 | - vllm/sampling_params.py |
| 2473 | - vllm/transformers_utils/ |
| 2474 | - vllm/utils/ |
| 2475 | - vllm/v1/ |
| 2476 | optional: true |
| 2477 | commands: |
| 2478 | - pip install 'modelscope<1.38' |
| 2479 | - pytest -v -s test_regression.py |
| 2480 | |
| 2481 | #--------------------------------------------------------- mi300 · ray_compat ---------------------------------------------------------# |
| 2482 | |
| 2483 | - label: ":amd: (MI300) Ray Dependency Compatibility Check" # TBD |
| 2484 | timeout_in_minutes: 25 |
| 2485 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2486 | dind: false |
| 2487 | agent_pool: mi300_1 |
| 2488 | working_dir: "/" |
| 2489 | source_file_dependencies: |
| 2490 | - requirements/ |
| 2491 | - setup.py |
| 2492 | - vllm/platforms/rocm.py |
| 2493 | - .buildkite/scripts/check-ray-compatibility.sh |
| 2494 | optional: true |
| 2495 | commands: |
| 2496 | - bash /vllm-workspace/.buildkite/scripts/check-ray-compatibility.sh |
| 2497 | |
| 2498 | #------------------------------------------------------------ mi300 · v1 -------------------------------------------------------------# |
| 2499 | |
| 2500 | - label: ":amd: (MI300) Spec Decode Speculators + MTP" # TBD |
| 2501 | timeout_in_minutes: 75 |
| 2502 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2503 | dind: false |
| 2504 | agent_pool: mi300_1 |
| 2505 | optional: true |
| 2506 | working_dir: "/vllm-workspace/tests" |
| 2507 | source_file_dependencies: |
| 2508 | - vllm/v1/spec_decode/ |
| 2509 | - vllm/v1/worker/gpu/spec_decode/ |
| 2510 | - vllm/model_executor/model_loader/ |
| 2511 | - vllm/v1/sample/ |
| 2512 | - vllm/model_executor/layers/ |
| 2513 | - vllm/transformers_utils/configs/speculators/ |
| 2514 | - tests/v1/e2e/spec_decode/ |
| 2515 | - vllm/platforms/rocm.py |
| 2516 | - vllm/v1/attention/backends/ |
| 2517 | commands: |
| 2518 | - pytest -v -s v1/e2e/spec_decode/speculators/ |
| 2519 | - pytest -v -s v1/e2e/spec_decode/mtp/ |
| 2520 | |
| 2521 | - label: ":amd: (MI300) V1 Spec Decode" # TBD |
| 2522 | timeout_in_minutes: 50 |
| 2523 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2524 | dind: false |
| 2525 | agent_pool: mi300_1 |
| 2526 | working_dir: "/vllm-workspace/tests" |
| 2527 | source_file_dependencies: |
| 2528 | - vllm/ |
| 2529 | - tests/v1/spec_decode |
| 2530 | - vllm/config/ |
| 2531 | - vllm/distributed/ |
| 2532 | - vllm/inputs/ |
| 2533 | - vllm/model_executor/ |
| 2534 | - vllm/models/qwen4_exp/ |
| 2535 | - vllm/platforms/ |
| 2536 | - vllm/sampling_params.py |
| 2537 | - vllm/transformers_utils/ |
| 2538 | - vllm/utils/ |
| 2539 | - vllm/v1/ |
| 2540 | optional: true |
| 2541 | commands: |
| 2542 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 2543 | - pytest -v -s -m 'not slow_test' v1/spec_decode |
| 2544 | |
| 2545 | - label: ":amd: (MI300) Acceptance Length (Large Models)" # TBD |
| 2546 | timeout_in_minutes: 45 |
| 2547 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2548 | dind: false |
| 2549 | agent_pool: mi300_1 |
| 2550 | optional: true |
| 2551 | working_dir: "/vllm-workspace/tests" |
| 2552 | source_file_dependencies: |
| 2553 | - vllm/v1/spec_decode/ |
| 2554 | - vllm/model_executor/models/mlp_speculator.py |
| 2555 | - tests/v1/spec_decode/test_acceptance_length.py |
| 2556 | - vllm/platforms/rocm.py |
| 2557 | commands: |
| 2558 | - export VLLM_ALLOW_INSECURE_SERIALIZATION=1 |
| 2559 | - pytest -v -s v1/spec_decode/test_acceptance_length.py -m slow_test |
| 2560 | |
| 2561 | - label: ":amd: (MI300) E2E Scheduling" # TBD |
| 2562 | timeout_in_minutes: 55 |
| 2563 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2564 | dind: false |
| 2565 | agent_pool: mi300_1 |
| 2566 | optional: true |
| 2567 | working_dir: "/vllm-workspace/tests" |
| 2568 | source_file_dependencies: |
| 2569 | - vllm/v1/ |
| 2570 | - tests/v1/e2e/general/ |
| 2571 | - vllm/platforms/rocm.py |
| 2572 | commands: |
| 2573 | - pytest -v -s v1/e2e/general/test_async_scheduling.py |
| 2574 | |
| 2575 | - label: ":amd: (MI300) E2E Scheduling Accuracy" # TBD |
| 2576 | timeout_in_minutes: 55 |
| 2577 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2578 | dind: false |
| 2579 | agent_pool: mi300_1 |
| 2580 | optional: true |
| 2581 | working_dir: "/vllm-workspace/tests" |
| 2582 | source_file_dependencies: |
| 2583 | - vllm/v1/ |
| 2584 | - tests/v1/e2e/general/ |
| 2585 | - vllm/platforms/rocm.py |
| 2586 | commands: |
| 2587 | - pytest -v -s v1/e2e/general/test_async_scheduling_accuracy.py |
| 2588 | |
| 2589 | - label: ":amd: (MI300) V1 Engine" # TBD |
| 2590 | timeout_in_minutes: 45 |
| 2591 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2592 | dind: false |
| 2593 | agent_pool: mi300_1 |
| 2594 | optional: true |
| 2595 | working_dir: "/vllm-workspace/tests" |
| 2596 | source_file_dependencies: |
| 2597 | - vllm/v1/engine/ |
| 2598 | - tests/v1/engine/ |
| 2599 | - tests/v1/test_tensor_ipc_queue.py |
| 2600 | - vllm/platforms/rocm.py |
| 2601 | commands: |
| 2602 | - pytest -v -s v1/engine/test_preprocess_error_handling.py |
| 2603 | - pytest -v -s v1/engine --ignore v1/engine/test_preprocess_error_handling.py |
| 2604 | - pytest -v -s v1/test_tensor_ipc_queue.py |
| 2605 | |
| 2606 | - label: ":amd: (MI300) Speculators Correctness Nightly" |
| 2607 | timeout_in_minutes: 40 |
| 2608 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2609 | dind: false |
| 2610 | agent_pool: mi300_1 |
| 2611 | optional: true |
| 2612 | working_dir: "/vllm-workspace/tests" |
| 2613 | source_file_dependencies: |
| 2614 | - vllm/config/speculative.py |
| 2615 | - vllm/engine/arg_utils.py |
| 2616 | - vllm/transformers_utils/config.py |
| 2617 | - vllm/transformers_utils/configs/speculators/ |
| 2618 | - vllm/v1/spec_decode/ |
| 2619 | - vllm/v1/worker/gpu/spec_decode/ |
| 2620 | - vllm/v1/worker/gpu_model_runner.py |
| 2621 | - vllm/v1/sample/ |
| 2622 | - vllm/v1/attention/backends/ |
| 2623 | - vllm/v1/attention/selector.py |
| 2624 | - vllm/model_executor/model_loader/ |
| 2625 | - vllm/model_executor/layers/ |
| 2626 | - vllm/model_executor/models/llama_eagle3.py |
| 2627 | - vllm/model_executor/models/qwen3.py |
| 2628 | - vllm/model_executor/models/qwen3_dflash.py |
| 2629 | - vllm/model_executor/models/registry.py |
| 2630 | - vllm/_aiter_ops.py |
| 2631 | - tests/evals/gsm8k/ |
| 2632 | - tests/v1/spec_decode/test_speculators_correctness.py |
| 2633 | - vllm/platforms/rocm.py |
| 2634 | - vllm/v1/worker/gpu/spec_decode/dflash/ |
| 2635 | commands: |
| 2636 | - export VLLM_ALLOW_INSECURE_SERIALIZATION=1 |
| 2637 | - pytest -v -s v1/spec_decode/test_speculators_correctness.py -m slow_test |
| 2638 | |
| 2639 | - label: ":amd: (MI300) Extract Hidden States Integration" # TBD |
| 2640 | timeout_in_minutes: 30 |
| 2641 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2642 | dind: false |
| 2643 | agent_pool: mi300_2 |
| 2644 | num_gpus: 2 |
| 2645 | working_dir: /vllm-workspace |
| 2646 | source_file_dependencies: |
| 2647 | - vllm/config/speculative.py |
| 2648 | - vllm/distributed/kv_transfer/kv_connector/ |
| 2649 | - vllm/model_executor/layers/attention/ |
| 2650 | - vllm/model_executor/layers/mamba/ |
| 2651 | - vllm/model_executor/model_loader/ |
| 2652 | - vllm/model_executor/models/extract_hidden_states.py |
| 2653 | - vllm/model_executor/models/llama.py |
| 2654 | - vllm/model_executor/models/qwen3_5.py |
| 2655 | - vllm/model_executor/models/qwen3_next.py |
| 2656 | - vllm/model_executor/models/registry.py |
| 2657 | - vllm/transformers_utils/configs/extract_hidden_states.py |
| 2658 | - vllm/transformers_utils/configs/qwen3_5.py |
| 2659 | - vllm/v1/attention/backends/ |
| 2660 | - vllm/v1/attention/selector.py |
| 2661 | - vllm/v1/kv_cache_interface.py |
| 2662 | - vllm/v1/spec_decode/extract_hidden_states.py |
| 2663 | - vllm/v1/worker/gpu_model_runner.py |
| 2664 | - vllm/_aiter_ops.py |
| 2665 | - tests/v1/kv_connector/extract_hidden_states_integration/ |
| 2666 | - vllm/platforms/rocm.py |
| 2667 | - vllm/distributed/kv_transfer/kv_connector/v1/example_hidden_states_connector.py |
| 2668 | - tests/v1/kv_connector/extract_hidden_states_integration |
| 2669 | optional: true |
| 2670 | commands: |
| 2671 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 2672 | - pytest -v -s -m 'distributed' tests/v1/kv_connector/extract_hidden_states_integration |
| 2673 | |
| 2674 | - label: ":amd: (MI300) V1 Attention Shard %N" # TBD |
| 2675 | timeout_in_minutes: 125 |
| 2676 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2677 | dind: false |
| 2678 | agent_pool: mi300_1 |
| 2679 | parallelism: 2 |
| 2680 | optional: true |
| 2681 | working_dir: "/vllm-workspace/tests" |
| 2682 | source_file_dependencies: |
| 2683 | - vllm/config/attention.py |
| 2684 | - vllm/model_executor/layers/attention |
| 2685 | - vllm/v1/attention |
| 2686 | - tests/v1/attention |
| 2687 | - vllm/_aiter_ops.py |
| 2688 | - vllm/envs.py |
| 2689 | - vllm/platforms/rocm.py |
| 2690 | commands: |
| 2691 | - pytest -v -s v1/attention --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 2692 | |
| 2693 | - label: ":amd: (MI300) Metrics, Tracing" # TBD |
| 2694 | timeout_in_minutes: 30 |
| 2695 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2696 | dind: false |
| 2697 | agent_pool: mi300_2 |
| 2698 | optional: true |
| 2699 | num_gpus: 2 |
| 2700 | working_dir: "/vllm-workspace/tests" |
| 2701 | source_file_dependencies: |
| 2702 | - vllm/ |
| 2703 | - tests/v1/tracing |
| 2704 | - tests/tracing/ |
| 2705 | - vllm/config/ |
| 2706 | - vllm/distributed/ |
| 2707 | - vllm/engine/ |
| 2708 | - vllm/inputs/ |
| 2709 | - vllm/model_executor/ |
| 2710 | - vllm/multimodal/ |
| 2711 | - vllm/platforms/ |
| 2712 | - vllm/sampling_params.py |
| 2713 | - vllm/tracing/ |
| 2714 | - vllm/transformers_utils/ |
| 2715 | - vllm/utils/ |
| 2716 | - vllm/v1/ |
| 2717 | commands: |
| 2718 | - "pip install \ |
| 2719 | 'opentelemetry-sdk>=1.26.0' \ |
| 2720 | 'opentelemetry-api>=1.26.0' \ |
| 2721 | 'opentelemetry-exporter-otlp>=1.26.0' \ |
| 2722 | 'opentelemetry-semantic-conventions-ai>=0.4.1'" |
| 2723 | - pytest -v -s v1/tracing |
| 2724 | - pytest -v -s tracing |
| 2725 | |
| 2726 | - label: ":amd: (MI300) V1 E2E" # TBD |
| 2727 | timeout_in_minutes: 30 |
| 2728 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2729 | dind: false |
| 2730 | agent_pool: mi300_2 |
| 2731 | num_gpus: 2 |
| 2732 | optional: true |
| 2733 | working_dir: "/vllm-workspace/tests" |
| 2734 | source_file_dependencies: |
| 2735 | - vllm/ |
| 2736 | - tests/v1/e2e |
| 2737 | - vllm/compilation/ |
| 2738 | - vllm/config/ |
| 2739 | - vllm/distributed/ |
| 2740 | - vllm/engine/ |
| 2741 | - vllm/envs.py |
| 2742 | - vllm/forward_context.py |
| 2743 | - vllm/inputs/ |
| 2744 | - vllm/logger.py |
| 2745 | - vllm/logging_utils/ |
| 2746 | - vllm/model_executor/ |
| 2747 | - vllm/multimodal/ |
| 2748 | - vllm/platforms/ |
| 2749 | - vllm/sampling_params.py |
| 2750 | - vllm/transformers_utils/ |
| 2751 | - vllm/triton_utils/ |
| 2752 | - vllm/utils/ |
| 2753 | - vllm/v1/ |
| 2754 | - tests/v1/e2e/spec_decode/ |
| 2755 | commands: |
| 2756 | - >- |
| 2757 | pytest -v -s |
| 2758 | v1/e2e/spec_decode/draft_model/test_draft_model.py::test_draft_model_tensor_parallelism |
| 2759 | v1/e2e/spec_decode/draft_model/test_draft_model.py::test_draft_model_engine_args_tensor_parallelism |
| 2760 | |
| 2761 | - label: ":amd: (MI300) Mooncake EC TCP E2E" |
| 2762 | timeout_in_minutes: 60 |
| 2763 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2764 | dind: false |
| 2765 | agent_pool: mi300_2 |
| 2766 | num_gpus: 2 |
| 2767 | optional: true |
| 2768 | working_dir: "/vllm-workspace" |
| 2769 | source_file_dependencies: |
| 2770 | - vllm/distributed/ec_transfer/ |
| 2771 | - vllm/config/ec_transfer.py |
| 2772 | - vllm/config/multimodal.py |
| 2773 | - vllm/config/vllm.py |
| 2774 | - vllm/multimodal/ |
| 2775 | - vllm/v1/core/encoder_cache_manager.py |
| 2776 | - vllm/v1/core/sched/ |
| 2777 | - vllm/v1/engine/ |
| 2778 | - vllm/v1/worker/ec_connector_model_runner_mixin.py |
| 2779 | - vllm/v1/worker/gpu_model_runner.py |
| 2780 | - vllm/v1/worker/gpu_worker.py |
| 2781 | - vllm/v1/worker/gpu/ec_connector.py |
| 2782 | - vllm/v1/worker/gpu/model_runner.py |
| 2783 | - vllm/v1/worker/mm_encoder_model_runner.py |
| 2784 | - vllm/v1/worker/encoder_cudagraph.py |
| 2785 | - vllm/v1/worker/encoder_cudagraph_defs.py |
| 2786 | - vllm/v1/attention/ops/vit_attn_wrappers.py |
| 2787 | - vllm/v1/worker/gpu/model_states/ |
| 2788 | - vllm/v1/worker/gpu/mm/ |
| 2789 | - tests/v1/cudagraph/test_encoder_cudagraph.py |
| 2790 | - tests/models/multimodal/generation/test_vit_cudagraph.py |
| 2791 | - examples/disaggregated/disaggregated_encoder/ |
| 2792 | - tests/v1/ec_connector/ |
| 2793 | - requirements/kv_connectors.txt |
| 2794 | - vllm/platforms/rocm.py |
| 2795 | - requirements/rocm.txt |
| 2796 | commands: |
| 2797 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 2798 | - export PYTHON_BIN='python' |
| 2799 | - export PYTHONUNBUFFERED='1' |
| 2800 | - export VLLM_HOST_IP='127.0.0.1' |
| 2801 | - export VLLM_USE_V2_MODEL_RUNNER='1' |
| 2802 | - export E_CUDAGRAPH_MM_ENCODER='1' |
| 2803 | - export E_MM_ENCODER_ATTN_BACKEND='TRITON_ATTN' |
| 2804 | - export MOONCAKE_EC_PROTOCOL='tcp' |
| 2805 | - export USE_MM_PROMPTS='1' |
| 2806 | - export SKIP_BASELINE='0' |
| 2807 | - export CONCURRENCY='3' |
| 2808 | - export REPEAT='2' |
| 2809 | - export LOG_PATH='/tmp/mooncake-ec-e2e' |
| 2810 | - export BASELINE_FILE='/tmp/mooncake-ec-e2e/baseline.json' |
| 2811 | - bash tests/v1/ec_connector/integration/run_epd_mooncake_ec_full_pipeline.sh |
| 2812 | |
| 2813 | - label: ":amd: (MI300) NixlConnector PD edge case" # TBD |
| 2814 | timeout_in_minutes: 65 |
| 2815 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2816 | dind: false |
| 2817 | agent_pool: mi300_2 |
| 2818 | num_gpus: 2 |
| 2819 | optional: true |
| 2820 | working_dir: "/vllm-workspace/tests" |
| 2821 | source_file_dependencies: |
| 2822 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 2823 | - vllm/v1/core/sched/ |
| 2824 | - tests/v1/kv_connector/nixl_integration/ |
| 2825 | - vllm/platforms/rocm.py |
| 2826 | commands: |
| 2827 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 2828 | - PREFILL_GPU_ID=0 DECODE_GPU_ID=1 ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/run_edge_case_test.sh |
| 2829 | |
| 2830 | - label: ":amd: (MI300) CrossLayer KV layout Distributed NixlConnector PD accuracy" # TBD |
| 2831 | timeout_in_minutes: 60 |
| 2832 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2833 | dind: false |
| 2834 | agent_pool: mi300_4 |
| 2835 | num_gpus: 4 |
| 2836 | optional: true |
| 2837 | working_dir: "/vllm-workspace/tests" |
| 2838 | source_file_dependencies: |
| 2839 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 2840 | - tests/v1/kv_connector/nixl_integration/ |
| 2841 | - vllm/platforms/rocm.py |
| 2842 | commands: |
| 2843 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 2844 | - CROSS_LAYERS_BLOCKS=True ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh |
| 2845 | |
| 2846 | - label: ":amd: (MI300) Distributed DP Extended" # TBD |
| 2847 | timeout_in_minutes: 75 |
| 2848 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2849 | dind: false |
| 2850 | agent_pool: mi300_4 |
| 2851 | num_gpus: 4 |
| 2852 | working_dir: "/vllm-workspace/tests" |
| 2853 | source_file_dependencies: |
| 2854 | - vllm/distributed/ |
| 2855 | - tests/v1/distributed |
| 2856 | - tests/v1/engine/test_engine_core_client.py |
| 2857 | - tests/distributed/test_utils |
| 2858 | - vllm/platforms/rocm.py |
| 2859 | commands: |
| 2860 | - export NCCL_CUMEM_HOST_ENABLE=0 |
| 2861 | - TP_SIZE=2 DP_SIZE=2 pytest -v -s v1/distributed/test_async_llm_dp.py |
| 2862 | - TP_SIZE=2 DP_SIZE=2 pytest -v -s v1/distributed/test_eagle_dp.py |
| 2863 | - TP_SIZE=2 DP_SIZE=2 pytest -v -s v1/distributed/test_external_lb_dp.py |
| 2864 | - TP_SIZE=1 DP_SIZE=4 pytest -v -s v1/distributed/test_dense_dp_world_size.py |
| 2865 | - TP_SIZE=1 DP_SIZE=4 pytest -v -s v1/distributed/test_internal_lb_dp.py |
| 2866 | - TP_SIZE=1 DP_SIZE=4 pytest -v -s v1/distributed/test_hybrid_lb_dp.py |
| 2867 | - pytest -v -s v1/engine/test_engine_core_client.py::test_kv_cache_events_dp |
| 2868 | - pytest -v -s distributed/test_utils.py |
| 2869 | |
| 2870 | - label: ":amd: (MI300) Distributed NixlConnector PD accuracy" # TBD |
| 2871 | timeout_in_minutes: 45 |
| 2872 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2873 | dind: false |
| 2874 | agent_pool: mi300_4 |
| 2875 | num_gpus: 4 |
| 2876 | optional: true |
| 2877 | working_dir: "/vllm-workspace/tests" |
| 2878 | source_file_dependencies: |
| 2879 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 2880 | - tests/v1/kv_connector/nixl_integration/ |
| 2881 | - vllm/platforms/rocm.py |
| 2882 | commands: |
| 2883 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 2884 | - ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh |
| 2885 | |
| 2886 | - label: ":amd: (MI300) Push NixlConnector PP prefill PD accuracy" # TBD |
| 2887 | timeout_in_minutes: 55 |
| 2888 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2889 | dind: false |
| 2890 | agent_pool: mi300_4 |
| 2891 | num_gpus: 4 |
| 2892 | optional: true |
| 2893 | working_dir: "/vllm-workspace/tests" |
| 2894 | source_file_dependencies: |
| 2895 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 2896 | - tests/v1/kv_connector/nixl_integration/ |
| 2897 | - tests/v1/kv_connector/nixl_push_integration/ |
| 2898 | - vllm/platforms/rocm.py |
| 2899 | commands: |
| 2900 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 2901 | - ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_push_integration/config_sweep_accuracy_test.sh |
| 2902 | |
| 2903 | - label: ":amd: (MI300) DP EP Distributed NixlConnector PD accuracy" # TBD |
| 2904 | timeout_in_minutes: 30 |
| 2905 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2906 | dind: false |
| 2907 | agent_pool: mi300_4 |
| 2908 | num_gpus: 4 |
| 2909 | optional: true |
| 2910 | working_dir: "/vllm-workspace/tests" |
| 2911 | source_file_dependencies: |
| 2912 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 2913 | - tests/v1/kv_connector/nixl_integration/ |
| 2914 | - vllm/platforms/rocm.py |
| 2915 | commands: |
| 2916 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 2917 | - DP_EP=1 ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh |
| 2918 | |
| 2919 | - label: ":amd: (MI300) Hybrid SSM NixlConnector PD accuracy" # TBD |
| 2920 | timeout_in_minutes: 80 |
| 2921 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2922 | dind: false |
| 2923 | agent_pool: mi300_4 |
| 2924 | num_gpus: 4 |
| 2925 | optional: true |
| 2926 | working_dir: "/vllm-workspace/tests" |
| 2927 | source_file_dependencies: |
| 2928 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 2929 | - tests/v1/kv_connector/nixl_integration/ |
| 2930 | - vllm/platforms/rocm.py |
| 2931 | commands: |
| 2932 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 2933 | - HYBRID_SSM=1 ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh |
| 2934 | |
| 2935 | - label: ":amd: (MI300) Hybrid SSM NixlConnector PD prefix cache" # TBD |
| 2936 | timeout_in_minutes: 25 |
| 2937 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2938 | dind: false |
| 2939 | agent_pool: mi300_2 |
| 2940 | num_gpus: 2 |
| 2941 | optional: true |
| 2942 | working_dir: "/vllm-workspace/tests" |
| 2943 | source_file_dependencies: |
| 2944 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 2945 | - vllm/v1/core/sched/ |
| 2946 | - vllm/v1/core/kv_cache_coordinator.py |
| 2947 | - tests/v1/kv_connector/nixl_integration/ |
| 2948 | - vllm/platforms/rocm.py |
| 2949 | commands: |
| 2950 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 2951 | - ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/run_mamba_prefix_cache_test.sh |
| 2952 | |
| 2953 | - label: ":amd: (MI300) MultiConnector (Nixl+Offloading) PD accuracy" # TBD |
| 2954 | timeout_in_minutes: 30 |
| 2955 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2956 | dind: false |
| 2957 | agent_pool: mi300_2 |
| 2958 | num_gpus: 2 |
| 2959 | optional: true |
| 2960 | working_dir: "/vllm-workspace/tests" |
| 2961 | source_file_dependencies: |
| 2962 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 2963 | - vllm/distributed/kv_transfer/kv_connector/v1/multi_connector.py |
| 2964 | - vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py |
| 2965 | - vllm/distributed/kv_transfer/kv_connector/v1/offloading/ |
| 2966 | - tests/v1/kv_connector/nixl_integration/ |
| 2967 | - vllm/platforms/rocm.py |
| 2968 | commands: |
| 2969 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 2970 | - ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/run_multi_connector_accuracy_test.sh |
| 2971 | |
| 2972 | - label: ":amd: (MI300) MultiConnector (Nixl+Offloading) PD edge cases" # TBD |
| 2973 | timeout_in_minutes: 25 |
| 2974 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2975 | dind: false |
| 2976 | agent_pool: mi300_2 |
| 2977 | num_gpus: 2 |
| 2978 | optional: true |
| 2979 | working_dir: "/vllm-workspace/tests" |
| 2980 | source_file_dependencies: |
| 2981 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 2982 | - vllm/distributed/kv_transfer/kv_connector/v1/multi_connector.py |
| 2983 | - vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py |
| 2984 | - vllm/distributed/kv_transfer/kv_connector/v1/offloading/ |
| 2985 | - tests/v1/kv_connector/nixl_integration/ |
| 2986 | - vllm/platforms/rocm.py |
| 2987 | commands: |
| 2988 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 2989 | - ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/run_multi_connector_edge_case_test.sh |
| 2990 | |
| 2991 | - label: ":amd: (MI300) V1 E2E Hybrid Chunked Prefill" # TBD |
| 2992 | timeout_in_minutes: 35 |
| 2993 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 2994 | dind: false |
| 2995 | agent_pool: mi300_4 |
| 2996 | num_gpus: 4 |
| 2997 | optional: true |
| 2998 | source_file_dependencies: |
| 2999 | - vllm/v1/attention/backends/utils.py |
| 3000 | - vllm/v1/worker/gpu_model_runner.py |
| 3001 | - tests/v1/e2e/test_hybrid_chunked_prefill.py |
| 3002 | - vllm/v1/core/ |
| 3003 | working_dir: /vllm-workspace/tests |
| 3004 | commands: |
| 3005 | - pytest -v -s v1/e2e/test_hybrid_chunked_prefill.py |
| 3006 | |
| 3007 | ######################################################################################################################################### |
| 3008 | # # |
| 3009 | # MI355 (gfx950) tests # |
| 3010 | # # |
| 3011 | ######################################################################################################################################### |
| 3012 | |
| 3013 | #----------------------------------------------------- mi355 · basic_correctness -----------------------------------------------------# |
| 3014 | |
| 3015 | - label: ":amd: (MI355 DPX) Basic Correctness Models %N" # TBD |
| 3016 | timeout_in_minutes: 60 |
| 3017 | parallelism: 2 |
| 3018 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3019 | dind: false |
| 3020 | agent_pool: mi355_dpx |
| 3021 | fast_check: true |
| 3022 | working_dir: "/vllm-workspace/tests" |
| 3023 | source_file_dependencies: |
| 3024 | - vllm/ |
| 3025 | - tests/basic_correctness/ |
| 3026 | optional: true |
| 3027 | commands: |
| 3028 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3029 | - VLLM_TARGET_TEST_SUITE=MI355 pytest -v -s basic_correctness/models --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB |
| 3030 | |
| 3031 | - label: ":amd: (MI355 DPX) Basic Correctness CuMem" # TBD |
| 3032 | timeout_in_minutes: 60 |
| 3033 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3034 | dind: false |
| 3035 | agent_pool: mi355_dpx |
| 3036 | fast_check: true |
| 3037 | working_dir: "/vllm-workspace/tests" |
| 3038 | source_file_dependencies: |
| 3039 | - vllm/ |
| 3040 | - tests/basic_correctness/ |
| 3041 | optional: true |
| 3042 | commands: |
| 3043 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3044 | - pytest -v -s basic_correctness/memory/cumem |
| 3045 | |
| 3046 | - label: ":amd: (MI300) Basic Correctness Sleep Mode" # TBD |
| 3047 | timeout_in_minutes: 60 |
| 3048 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300] |
| 3049 | dind: false |
| 3050 | agent_pool: mi300_1 |
| 3051 | fast_check: true |
| 3052 | working_dir: "/vllm-workspace/tests" |
| 3053 | source_file_dependencies: |
| 3054 | - vllm/ |
| 3055 | - tests/basic_correctness/ |
| 3056 | optional: true |
| 3057 | commands: |
| 3058 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3059 | - pytest -v -s basic_correctness/memory/sleep_mode |
| 3060 | |
| 3061 | - label: ":amd: (MI355 DPX) Basic Correctness CPU Offload" # TBD |
| 3062 | timeout_in_minutes: 60 |
| 3063 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3064 | dind: false |
| 3065 | agent_pool: mi355_dpx |
| 3066 | fast_check: true |
| 3067 | working_dir: "/vllm-workspace/tests" |
| 3068 | source_file_dependencies: |
| 3069 | - vllm/ |
| 3070 | - tests/basic_correctness/ |
| 3071 | optional: true |
| 3072 | commands: |
| 3073 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3074 | - pytest -v -s basic_correctness/cpu_offload |
| 3075 | |
| 3076 | - label: ":amd: (MI355 DPX) Basic Correctness Prefetch Offload" # TBD |
| 3077 | timeout_in_minutes: 60 |
| 3078 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3079 | dind: false |
| 3080 | agent_pool: mi355_dpx |
| 3081 | fast_check: true |
| 3082 | working_dir: "/vllm-workspace/tests" |
| 3083 | source_file_dependencies: |
| 3084 | - vllm/ |
| 3085 | - tests/basic_correctness/ |
| 3086 | optional: true |
| 3087 | commands: |
| 3088 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3089 | - pytest -v -s basic_correctness/prefetch_offload |
| 3090 | |
| 3091 | #-------------------------------------------------------- mi355 · benchmarks ---------------------------------------------------------# |
| 3092 | |
| 3093 | - label: ":amd: (MI355 DPX) Benchmarks CLI" # TBD |
| 3094 | timeout_in_minutes: 40 |
| 3095 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3096 | dind: false |
| 3097 | agent_pool: mi355_dpx |
| 3098 | optional: true |
| 3099 | working_dir: "/vllm-workspace/tests" |
| 3100 | source_file_dependencies: |
| 3101 | - vllm/ |
| 3102 | - tests/benchmarks/ |
| 3103 | - rust/src/bench/tests/python_serve_flags.txt |
| 3104 | commands: |
| 3105 | - pytest -v -s benchmarks/ |
| 3106 | |
| 3107 | - label: ":amd: (MI355) Attention Benchmark Smoke" # TBD |
| 3108 | timeout_in_minutes: 25 |
| 3109 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3110 | dind: false |
| 3111 | agent_pool: mi355_2 |
| 3112 | num_gpus: 2 |
| 3113 | optional: true |
| 3114 | working_dir: "/vllm-workspace/" |
| 3115 | source_file_dependencies: |
| 3116 | - benchmarks/attention_benchmarks/ |
| 3117 | - vllm/v1/attention/ |
| 3118 | - vllm/_aiter_ops.py |
| 3119 | - vllm/platforms/rocm.py |
| 3120 | commands: |
| 3121 | - python3 benchmarks/attention_benchmarks/benchmark.py --backends ROCM_ATTN ROCM_AITER_FA ROCM_AITER_UNIFIED_ATTN --batch-specs "8q1s1k" |
| 3122 | |
| 3123 | #---------------------------------------------------------- mi355 · compile ----------------------------------------------------------# |
| 3124 | |
| 3125 | - label: ":amd: (MI355 DPX) PyTorch Compilation H100 Cases" |
| 3126 | timeout_in_minutes: 45 |
| 3127 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3128 | dind: false |
| 3129 | agent_pool: mi355_dpx |
| 3130 | num_gpus: 1 |
| 3131 | optional: true |
| 3132 | working_dir: "/vllm-workspace/tests" |
| 3133 | source_file_dependencies: |
| 3134 | - vllm/__init__.py |
| 3135 | - vllm/_aiter_ops.py |
| 3136 | - vllm/_custom_ops.py |
| 3137 | - vllm/compilation/ |
| 3138 | - vllm/config/ |
| 3139 | - vllm/distributed/ |
| 3140 | - "!vllm/distributed/kv_transfer/" |
| 3141 | - vllm/engine/ |
| 3142 | - vllm/env_override.py |
| 3143 | - vllm/envs.py |
| 3144 | - vllm/forward_context.py |
| 3145 | - vllm/inputs/ |
| 3146 | - vllm/ir/ |
| 3147 | - vllm/kernels/ |
| 3148 | - vllm/logger.py |
| 3149 | - vllm/model_executor/ |
| 3150 | - vllm/multimodal/ |
| 3151 | - vllm/platforms/ |
| 3152 | - vllm/plugins/ |
| 3153 | - vllm/sampling_params.py |
| 3154 | - vllm/sequence.py |
| 3155 | - vllm/transformers_utils/ |
| 3156 | - vllm/triton_utils/ |
| 3157 | - vllm/utils/ |
| 3158 | - vllm/v1/ |
| 3159 | - tests/compile/h100/ |
| 3160 | - tests/utils.py |
| 3161 | - csrc/ |
| 3162 | commands: |
| 3163 | - pytest -v -s compile/h100/test_startup.py |
| 3164 | |
| 3165 | - label: ":amd: (MI355) Fusion E2E TP2 AR-RMS FP8 Config Sweep" |
| 3166 | timeout_in_minutes: 45 |
| 3167 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3168 | dind: false |
| 3169 | agent_pool: mi355_2 |
| 3170 | num_gpus: 2 |
| 3171 | optional: true |
| 3172 | working_dir: "/vllm-workspace/" |
| 3173 | source_file_dependencies: |
| 3174 | - vllm/compilation/ |
| 3175 | - vllm/model_executor/ |
| 3176 | - vllm/distributed/ |
| 3177 | - vllm/v1/attention/ |
| 3178 | - vllm/_aiter_ops.py |
| 3179 | - vllm/platforms/rocm.py |
| 3180 | - tests/compile/fusions_e2e/ |
| 3181 | - tests/utils.py |
| 3182 | - tests/conftest.py |
| 3183 | - csrc/quantization/ |
| 3184 | commands: |
| 3185 | # Existing MI300 jobs own BF16; TP2 Quick owns the excluded Llama cell. |
| 3186 | - amd-smi |
| 3187 | - VLLM_ROCM_USE_AITER=1 VLLM_ROCM_USE_AITER_CUSTOM_AR=1 pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py -k 'test_tp2_ar_rms_fp8_fusions and Llama-3 and ROCM_AITER_UNIFIED_ATTN and (dynamo_partition or +quant_fp8 or +rms_norm)' |
| 3188 | |
| 3189 | - label: ":amd: (MI355) Fusion E2E TP2 FP8 BF16 rewrites" |
| 3190 | timeout_in_minutes: 45 |
| 3191 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3192 | dind: false |
| 3193 | agent_pool: mi355_2 |
| 3194 | num_gpus: 2 |
| 3195 | optional: true |
| 3196 | working_dir: "/vllm-workspace/" |
| 3197 | source_file_dependencies: |
| 3198 | - vllm/compilation/ |
| 3199 | - vllm/model_executor/ |
| 3200 | - vllm/distributed/ |
| 3201 | - vllm/v1/attention/ |
| 3202 | - vllm/_aiter_ops.py |
| 3203 | - vllm/platforms/rocm.py |
| 3204 | - tests/compile/fusions_e2e/ |
| 3205 | - tests/utils.py |
| 3206 | - tests/conftest.py |
| 3207 | - csrc/quantization/ |
| 3208 | commands: |
| 3209 | # Dummy-weight rewrite coverage; trained FP8 AR/RMS equivalence is unresolved. |
| 3210 | # TODO: FP4 output-quant and trained MXFP4 AR/RMS fusion remain unsupported. |
| 3211 | - amd-smi |
| 3212 | - VLLM_ROCM_USE_AITER=1 VLLM_ROCM_USE_AITER_CUSTOM_AR=1 pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py -k 'test_tp2_ar_rms_fp8_fusions and qwen3 and ROCM_AITER_UNIFIED_ATTN and +quant_fp8 and (dynamo_partition or not +rms_norm)' |
| 3213 | - pytest -v -s tests/compile/fusions_e2e/test_tp2_rocm_moe.py |
| 3214 | |
| 3215 | - label: ":amd: (MI355) Fusion and Compile FP8" |
| 3216 | timeout_in_minutes: 45 |
| 3217 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3218 | dind: false |
| 3219 | agent_pool: mi355_2 |
| 3220 | num_gpus: 2 |
| 3221 | optional: true |
| 3222 | working_dir: "/vllm-workspace/" |
| 3223 | source_file_dependencies: |
| 3224 | - vllm/compilation/ |
| 3225 | - vllm/model_executor/layers/ |
| 3226 | - vllm/v1/attention/ |
| 3227 | - vllm/distributed/ |
| 3228 | - vllm/_aiter_ops.py |
| 3229 | - tests/compile/passes/ |
| 3230 | - tests/compile/correctness_e2e/test_attn_quant.py |
| 3231 | - vllm/model_executor/kernels/linear/ |
| 3232 | - vllm/model_executor/models/llama.py |
| 3233 | commands: |
| 3234 | - amd-smi |
| 3235 | - pytest -v -s tests/compile/passes/test_fusion_attn.py tests/compile/passes/test_silu_mul_quant_fusion.py tests/compile/passes/test_silu_mul_quant_manual_fusion.py |
| 3236 | - pytest -v -s tests/compile/correctness_e2e/test_attn_quant.py |
| 3237 | |
| 3238 | - label: ":amd: (MI355) Sequence Parallel Correctness" |
| 3239 | timeout_in_minutes: 180 |
| 3240 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3241 | dind: false |
| 3242 | agent_pool: mi355_2 |
| 3243 | num_gpus: 2 |
| 3244 | optional: true |
| 3245 | working_dir: "/vllm-workspace/" |
| 3246 | source_file_dependencies: |
| 3247 | - .buildkite/test_areas/compile.yaml |
| 3248 | - vllm/compilation/ |
| 3249 | - vllm/config/ |
| 3250 | - vllm/distributed/ |
| 3251 | - vllm/model_executor/kernels/linear/ |
| 3252 | - vllm/model_executor/layers/ |
| 3253 | - vllm/model_executor/models/llama.py |
| 3254 | - vllm/v1/attention/ |
| 3255 | - vllm/v1/worker/ |
| 3256 | - vllm/v1/cudagraph_dispatcher.py |
| 3257 | - vllm/platforms/rocm.py |
| 3258 | - tests/compile/correctness_e2e/test_sequence_parallel.py |
| 3259 | - tests/models/registry.py |
| 3260 | - tests/utils.py |
| 3261 | commands: |
| 3262 | - export VLLM_TEST_CLEAN_GPU_MEMORY=1 |
| 3263 | - pytest -v -s tests/compile/correctness_e2e/test_sequence_parallel.py |
| 3264 | |
| 3265 | - label: ":amd: (MI355) AsyncTP Correctness" |
| 3266 | timeout_in_minutes: 60 |
| 3267 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3268 | dind: false |
| 3269 | agent_pool: mi355_2 |
| 3270 | num_gpus: 2 |
| 3271 | optional: true |
| 3272 | working_dir: "/vllm-workspace/" |
| 3273 | source_file_dependencies: |
| 3274 | - .buildkite/test_areas/compile.yaml |
| 3275 | - vllm/compilation/ |
| 3276 | - vllm/config/ |
| 3277 | - vllm/distributed/ |
| 3278 | - vllm/model_executor/layers/ |
| 3279 | - vllm/model_executor/kernels/linear/ |
| 3280 | - vllm/model_executor/models/llama.py |
| 3281 | - vllm/model_executor/models/qwen3.py |
| 3282 | - vllm/v1/attention/ |
| 3283 | - vllm/_aiter_ops.py |
| 3284 | - vllm/platforms/rocm.py |
| 3285 | - vllm/transformers_utils/repo_utils.py |
| 3286 | - tests/compile/correctness_e2e/test_async_tp.py |
| 3287 | - tests/compile/fusions_e2e/common.py |
| 3288 | - tests/models/registry.py |
| 3289 | - tests/utils.py |
| 3290 | commands: |
| 3291 | - export VLLM_TEST_CLEAN_GPU_MEMORY=1 |
| 3292 | - pytest -v -s tests/compile/correctness_e2e/test_async_tp.py::test_rocm_async_tp_bf16_output_correctness |
| 3293 | - pytest -v -s tests/compile/correctness_e2e/test_async_tp.py::test_async_tp_pass_correctness -k RedHatAI |
| 3294 | - pytest -v -s tests/compile/correctness_e2e/test_async_tp.py::test_async_tp_pass_nvfp4_correctness |
| 3295 | |
| 3296 | - label: ":amd: (MI355 DPX) Fusion E2E FP8 Config Sweep" |
| 3297 | timeout_in_minutes: 45 |
| 3298 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3299 | dind: false |
| 3300 | agent_pool: mi355_dpx |
| 3301 | num_gpus: 1 |
| 3302 | optional: true |
| 3303 | working_dir: "/vllm-workspace/" |
| 3304 | source_file_dependencies: |
| 3305 | - .buildkite/test_areas/compile.yaml |
| 3306 | - csrc/quantization/ |
| 3307 | - vllm/compilation/ |
| 3308 | - vllm/model_executor/ |
| 3309 | - vllm/v1/attention/ |
| 3310 | - vllm/_aiter_ops.py |
| 3311 | - vllm/platforms/rocm.py |
| 3312 | - tests/compile/fusions_e2e/ |
| 3313 | commands: |
| 3314 | - rocm-smi |
| 3315 | - pytest -v -s tests/compile/fusions_e2e/test_tp1_quant.py -k "test_tp1_fp8_fusions and inductor_partition and not +rms_norm and not True and ((Llama-3 and TRITON_ATTN) or (Qwen3-30B and +quant_fp8 and TRITON_ATTN) or (DeepSeek-Coder and +quant_fp8 and TRITON_MLA))" |
| 3316 | |
| 3317 | #-------------------------------------------------------- mi355 · distributed --------------------------------------------------------# |
| 3318 | |
| 3319 | - label: ":amd: (MI355 DPX) EPLB Algorithm" # TBD |
| 3320 | timeout_in_minutes: 30 |
| 3321 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3322 | dind: false |
| 3323 | agent_pool: mi355_dpx |
| 3324 | optional: true |
| 3325 | working_dir: "/vllm-workspace/tests" |
| 3326 | source_file_dependencies: |
| 3327 | - vllm/distributed/eplb |
| 3328 | - tests/distributed/test_eplb_algo.py |
| 3329 | - tests/distributed/test_eplb_utils.py |
| 3330 | - vllm/platforms/rocm.py |
| 3331 | commands: |
| 3332 | - pytest -v -s distributed/test_eplb_algo.py |
| 3333 | - pytest -v -s distributed/test_eplb_utils.py |
| 3334 | |
| 3335 | - label: ":amd: (MI355 DPX) Sharded RDT Weight Transfer" |
| 3336 | timeout_in_minutes: 45 |
| 3337 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3338 | dind: false |
| 3339 | agent_pool: mi355_dpx |
| 3340 | num_gpus: 1 |
| 3341 | optional: true |
| 3342 | working_dir: "/vllm-workspace/tests" |
| 3343 | source_file_dependencies: |
| 3344 | - vllm/distributed/weight_transfer/ |
| 3345 | - tests/distributed/test_sharded_rdt_plan.py |
| 3346 | - tests/distributed/test_sharded_rdt_producer.py |
| 3347 | - tests/distributed/test_sharded_rdt_trainer.py |
| 3348 | - vllm/platforms/rocm.py |
| 3349 | commands: |
| 3350 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3351 | - uv pip install --system --no-deps 'ray==2.56.1' |
| 3352 | - pytest -v -s distributed/test_sharded_rdt_plan.py |
| 3353 | - pytest -v -s distributed/test_sharded_rdt_producer.py |
| 3354 | - pytest -v -s distributed/test_sharded_rdt_trainer.py |
| 3355 | |
| 3356 | - label: ":amd: (MI355) GEMM-RS-AR" |
| 3357 | timeout_in_minutes: 20 |
| 3358 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3359 | dind: false |
| 3360 | agent_pool: mi355_4 |
| 3361 | num_gpus: 4 |
| 3362 | optional: true |
| 3363 | working_dir: "/vllm-workspace/" |
| 3364 | source_file_dependencies: |
| 3365 | - .buildkite/test-amd.yaml |
| 3366 | - tests/kernels/test_gemm_rs_ar.py |
| 3367 | - tests/kernels/test_rocm_gemm_rs_ar.py |
| 3368 | - tests/conftest.py |
| 3369 | - vllm/models/common/ops/sequence_parallel.py |
| 3370 | - vllm/model_executor/layers/linear.py |
| 3371 | - vllm/model_executor/layers/utils.py |
| 3372 | - vllm/model_executor/layers/quantization/online/ |
| 3373 | - vllm/model_executor/layers/quantization/utils/mxfp8_utils.py |
| 3374 | - vllm/model_executor/kernels/linear/ |
| 3375 | - vllm/distributed/ |
| 3376 | - vllm/_aiter_ops.py |
| 3377 | - vllm/_custom_ops.py |
| 3378 | - csrc/rocm/skinny_gemms.cu |
| 3379 | - vllm/platforms/rocm.py |
| 3380 | commands: |
| 3381 | - pytest -v -s tests/kernels/test_rocm_gemm_rs_ar.py |
| 3382 | |
| 3383 | - label: ":amd: (MI355) Distributed DP Basic" # TBD |
| 3384 | timeout_in_minutes: 45 |
| 3385 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3386 | dind: false |
| 3387 | agent_pool: mi355_2 |
| 3388 | optional: true |
| 3389 | num_gpus: 2 |
| 3390 | working_dir: "/vllm-workspace/tests" |
| 3391 | source_file_dependencies: |
| 3392 | - vllm/distributed/ |
| 3393 | - vllm/engine/ |
| 3394 | - vllm/executor/ |
| 3395 | - vllm/worker/worker_base.py |
| 3396 | - vllm/v1/engine/ |
| 3397 | - vllm/v1/worker/ |
| 3398 | - tests/v1/distributed |
| 3399 | - tests/entrypoints/launchers/api_server/test_multi_api_servers.py |
| 3400 | - vllm/platforms/rocm.py |
| 3401 | - tests/v1/e2e/general/test_sharded_sampling.py |
| 3402 | - tests/v1/e2e/spec_decode/test_sharded_sampling.py |
| 3403 | - tests/v1/e2e/spec_decode/utils.py |
| 3404 | commands: |
| 3405 | - export NCCL_CUMEM_HOST_ENABLE=0 |
| 3406 | - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_async_llm_dp.py |
| 3407 | - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_eagle_dp.py |
| 3408 | - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_external_lb_dp.py |
| 3409 | - DP_SIZE=2 pytest -v -s entrypoints/launchers/api_server/test_multi_api_servers.py |
| 3410 | - pytest -v -s v1/e2e/general/test_sharded_sampling.py |
| 3411 | - pytest -v -s v1/e2e/spec_decode/test_sharded_sampling.py |
| 3412 | |
| 3413 | - label: ":amd: (MI355) MoRI EP Healthy Baseline" |
| 3414 | timeout_in_minutes: 10 |
| 3415 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3416 | dind: false |
| 3417 | agent_pool: mi355_2 |
| 3418 | num_gpus: 2 |
| 3419 | working_dir: "/vllm-workspace/tests" |
| 3420 | source_file_dependencies: |
| 3421 | - csrc/moe/ |
| 3422 | - csrc/rocm/ |
| 3423 | - tests/kernels/moe/ |
| 3424 | - vllm/model_executor/layers/fused_moe/ |
| 3425 | - vllm/model_executor/layers/quantization/ |
| 3426 | - vllm/distributed/ |
| 3427 | - vllm/config/ |
| 3428 | - vllm/forward_context.py |
| 3429 | - vllm/v1/worker/workspace.py |
| 3430 | - vllm/utils/import_utils.py |
| 3431 | - vllm/utils/math_utils.py |
| 3432 | - vllm/utils/torch_utils.py |
| 3433 | - vllm/platforms/ |
| 3434 | - vllm/_aiter_ops.py |
| 3435 | optional: true |
| 3436 | # TODO: Add fault injection, detection and recovery once MoRI supports them. |
| 3437 | # This job currently checks healthy EP execution only. |
| 3438 | commands: |
| 3439 | - VLLM_WORKER_MULTIPROC_METHOD=spawn pytest -v -s -rs kernels/moe/test_moe_layer.py::test_moe_layer_mori_graph |
| 3440 | - VLLM_TEST_ENABLE_MORI_MOE_LAYER=1 VLLM_WORKER_MULTIPROC_METHOD=spawn pytest -v -s -rs 'kernels/moe/test_moe_layer.py::test_moe_layer[False-mori_high_throughput-2-1-True]' --subtests='[32-512-512-8-2-bfloat16-None-False-False-False-False-mori_high_throughput-2-2-1]' |
| 3441 | |
| 3442 | - label: ":amd: (MI355) NixlConnector PD + Spec Decode acceptance" # TBD |
| 3443 | timeout_in_minutes: 45 |
| 3444 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3445 | dind: false |
| 3446 | agent_pool: mi355_2 |
| 3447 | num_gpus: 2 |
| 3448 | optional: true |
| 3449 | working_dir: "/vllm-workspace/tests" |
| 3450 | source_file_dependencies: |
| 3451 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 3452 | - vllm/v1/worker/kv_connector_model_runner_mixin.py |
| 3453 | - tests/v1/kv_connector/nixl_integration/ |
| 3454 | - vllm/platforms/rocm.py |
| 3455 | commands: |
| 3456 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 3457 | - KV_CACHE_MEMORY_BYTES=8G ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_spec_decode_test.sh |
| 3458 | |
| 3459 | - label: ":amd: (MI355) Distributed NixlConnector PD accuracy" # TBD |
| 3460 | timeout_in_minutes: 60 |
| 3461 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3462 | dind: false |
| 3463 | agent_pool: mi355_4 |
| 3464 | num_gpus: 4 |
| 3465 | optional: true |
| 3466 | working_dir: "/vllm-workspace/tests" |
| 3467 | source_file_dependencies: |
| 3468 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 3469 | - tests/v1/kv_connector/nixl_integration/ |
| 3470 | - vllm/platforms/rocm.py |
| 3471 | commands: |
| 3472 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 3473 | - ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh |
| 3474 | |
| 3475 | - label: ":amd: (MI355) Distributed AITER NixlConnector PD accuracy" |
| 3476 | timeout_in_minutes: 60 |
| 3477 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3478 | dind: false |
| 3479 | agent_pool: mi355_4 |
| 3480 | num_gpus: 4 |
| 3481 | optional: true |
| 3482 | working_dir: "/vllm-workspace/tests" |
| 3483 | source_file_dependencies: |
| 3484 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 3485 | - tests/v1/kv_connector/nixl_integration/ |
| 3486 | - tests/v1/kv_connector/rocm_pd_accuracy_utils.py |
| 3487 | - tests/utils.py |
| 3488 | - vllm/v1/attention/backends/rocm_aiter_unified_attn.py |
| 3489 | - vllm/platforms/rocm.py |
| 3490 | - requirements/common.txt |
| 3491 | - requirements/kv_connectors_rocm.txt |
| 3492 | commands: |
| 3493 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 3494 | - pytest -v -s v1/kv_connector/nixl_integration/test_accuracy_rocm.py |
| 3495 | |
| 3496 | - label: ":amd: (MI355) Distributed MooncakeConnector HIP PD accuracy" |
| 3497 | timeout_in_minutes: 30 |
| 3498 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3499 | dind: false |
| 3500 | agent_pool: mi355_4 |
| 3501 | num_gpus: 4 |
| 3502 | optional: true |
| 3503 | working_dir: "/vllm-workspace/tests" |
| 3504 | source_file_dependencies: |
| 3505 | - vllm/distributed/kv_transfer/kv_connector/v1/mooncake/ |
| 3506 | - tests/v1/kv_connector/mooncake_integration/ |
| 3507 | - tests/v1/kv_connector/rocm_pd_accuracy_utils.py |
| 3508 | - tests/utils.py |
| 3509 | - examples/disaggregated/mooncake_connector/ |
| 3510 | - vllm/v1/attention/backends/rocm_aiter_unified_attn.py |
| 3511 | - vllm/platforms/rocm.py |
| 3512 | - requirements/common.txt |
| 3513 | - requirements/kv_connectors_rocm.txt |
| 3514 | commands: |
| 3515 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 3516 | - pytest -v -s v1/kv_connector/mooncake_integration/test_accuracy_rocm.py |
| 3517 | |
| 3518 | - label: ":amd: (MI355) DP EP Distributed NixlConnector PD accuracy" # TBD |
| 3519 | timeout_in_minutes: 40 |
| 3520 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3521 | dind: false |
| 3522 | agent_pool: mi355_4 |
| 3523 | num_gpus: 4 |
| 3524 | optional: true |
| 3525 | working_dir: "/vllm-workspace/tests" |
| 3526 | source_file_dependencies: |
| 3527 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 3528 | - tests/v1/kv_connector/nixl_integration/ |
| 3529 | - vllm/platforms/rocm.py |
| 3530 | commands: |
| 3531 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 3532 | - DP_EP=1 ATTENTION_BACKEND=TRITON_ATTN bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh |
| 3533 | |
| 3534 | - label: ":amd: (MI355) Kimi-Linear-48B-A3B Disaggregated DP EP" |
| 3535 | timeout_in_minutes: 105 |
| 3536 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3537 | dind: false |
| 3538 | agent_pool: mi355_4 |
| 3539 | num_gpus: 4 |
| 3540 | optional: true |
| 3541 | working_dir: "/vllm-workspace/tests" |
| 3542 | source_file_dependencies: |
| 3543 | - vllm/platforms/rocm.py |
| 3544 | - vllm/models/kimi_k3/ |
| 3545 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 3546 | - tests/v1/kv_connector/nixl_integration/ |
| 3547 | - requirements/kv_connectors_rocm.txt |
| 3548 | commands: |
| 3549 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3550 | - export ENABLE_HMA_FLAG='1' |
| 3551 | - export DP_EP='1' |
| 3552 | - export GPU_MEMORY_UTILIZATION='0.9' |
| 3553 | - export VLLM_SSM_CONV_STATE_LAYOUT='DS' |
| 3554 | - export PREFILLER_TP_SIZE='2' |
| 3555 | - export DECODER_TP_SIZE='2' |
| 3556 | - export MODEL_NAMES='moonshotai/Kimi-Linear-48B-A3B-Instruct' |
| 3557 | - export PREFILL_BLOCK_SIZE='2048' |
| 3558 | - export DECODE_BLOCK_SIZE='2048' |
| 3559 | - export VLLM_SERVE_EXTRA_ARGS='--trust-remote-code,--mamba-cache-mode align,--max-model-len 32768' |
| 3560 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 3561 | - bash v1/kv_connector/nixl_integration/run_accuracy_test.sh |
| 3562 | |
| 3563 | - label: ":amd: (MI355) Distributed Features" # TBD |
| 3564 | timeout_in_minutes: 90 |
| 3565 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3566 | dind: false |
| 3567 | agent_pool: mi355_2 |
| 3568 | num_gpus: 2 |
| 3569 | optional: true |
| 3570 | working_dir: "/vllm-workspace/" |
| 3571 | source_file_dependencies: |
| 3572 | - csrc/custom_quickreduce.cu |
| 3573 | - csrc/ops.h |
| 3574 | - csrc/torch_bindings.cpp |
| 3575 | - vllm/distributed/ |
| 3576 | - vllm/model_executor/layers/ |
| 3577 | - vllm/entrypoints/llm.py |
| 3578 | - vllm/config/parallel.py |
| 3579 | - vllm/model_executor/layers/fused_moe/ |
| 3580 | - vllm/v1/engine/ |
| 3581 | - vllm/v1/executor/ |
| 3582 | - vllm/v1/worker/ |
| 3583 | - vllm/v1/distributed/ |
| 3584 | - vllm/v1/attention/backends/ |
| 3585 | - vllm/v1/attention/selector.py |
| 3586 | - vllm/_aiter_ops.py |
| 3587 | - vllm/_custom_ops.py |
| 3588 | - vllm/platforms/rocm.py |
| 3589 | - vllm/envs.py |
| 3590 | - examples/offline_inference/data_parallel.py |
| 3591 | - tests/distributed/test_context_parallel.py |
| 3592 | - tests/distributed/test_rocm_aiter_custom_ar.py |
| 3593 | - tests/distributed/test_rocm_quick_reduce.py |
| 3594 | - tests/distributed/test_quick_all_reduce.py |
| 3595 | - tests/v1/e2e/general/test_rocm_aiter_custom_ar.py |
| 3596 | - tests/v1/distributed/test_dbo.py |
| 3597 | - tests/utils.py |
| 3598 | - examples/features/data_parallel/data_parallel_offline.py |
| 3599 | - examples/rl/rlhf_async_new_apis.py |
| 3600 | - tests/distributed/test_weight_transfer.py |
| 3601 | - tests/distributed/test_packed_tensor.py |
| 3602 | commands: |
| 3603 | - pytest -v -s tests/distributed/test_context_parallel.py |
| 3604 | - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 examples/rl/rlhf_async_new_apis.py |
| 3605 | - VLLM_LOGGING_LEVEL=DEBUG python3 examples/features/data_parallel/data_parallel_offline.py --model=Qwen/Qwen1.5-MoE-A2.7B -tp=1 -dp=2 --max-model-len=2048 --all2all-backend=deepep_high_throughput |
| 3606 | - VLLM_LOGGING_LEVEL=DEBUG python3 examples/features/data_parallel/data_parallel_offline.py --model=Qwen/Qwen1.5-MoE-A2.7B -tp=1 -dp=2 --max-model-len=2048 --all2all-backend=allgather_reducescatter --disable-nccl-for-dp-synchronization |
| 3607 | - pytest -v -s tests/v1/distributed/test_dbo.py |
| 3608 | - VLLM_ALLOW_INSECURE_SERIALIZATION=1 pytest -v -s tests/distributed/test_weight_transfer.py |
| 3609 | - pytest -v -s tests/distributed/test_packed_tensor.py |
| 3610 | - pytest -v -s tests/distributed/test_rocm_aiter_custom_ar.py |
| 3611 | - pytest -v -s tests/v1/e2e/general/test_rocm_aiter_custom_ar.py |
| 3612 | - pytest -v -s tests/distributed/test_rocm_quick_reduce.py |
| 3613 | - pytest -v -s tests/distributed/test_quick_all_reduce.py |
| 3614 | |
| 3615 | - label: ":amd: (MI355) Distributed AgRs All2All" |
| 3616 | timeout_in_minutes: 30 |
| 3617 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3618 | dind: false |
| 3619 | agent_pool: mi355_2 |
| 3620 | num_gpus: 2 |
| 3621 | optional: true |
| 3622 | working_dir: "/vllm-workspace/" |
| 3623 | source_file_dependencies: |
| 3624 | - .buildkite/test_areas/distributed.yaml |
| 3625 | - vllm/distributed/ |
| 3626 | - vllm/config/parallel.py |
| 3627 | - vllm/config/vllm.py |
| 3628 | - vllm/forward_context.py |
| 3629 | - vllm/platforms/rocm.py |
| 3630 | - tests/distributed/test_mnnvl_alltoall.py |
| 3631 | - tests/utils.py |
| 3632 | commands: |
| 3633 | - pytest -v -s tests/distributed/test_mnnvl_alltoall.py::test_args_dispatch_combine |
| 3634 | |
| 3635 | - label: ":amd: (MI355) Fusion E2E TP2 Quick" # TBD |
| 3636 | timeout_in_minutes: 60 |
| 3637 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3638 | dind: false |
| 3639 | agent_pool: mi355_2 |
| 3640 | num_gpus: 2 |
| 3641 | optional: true |
| 3642 | working_dir: "/vllm-workspace/" |
| 3643 | source_file_dependencies: |
| 3644 | - csrc/quantization/ |
| 3645 | - vllm/compilation/ |
| 3646 | - vllm/distributed/ |
| 3647 | - vllm/model_executor/ |
| 3648 | - vllm/v1/attention/ |
| 3649 | - tests/compile/fusions_e2e/ |
| 3650 | - vllm/_aiter_ops.py |
| 3651 | - vllm/platforms/rocm.py |
| 3652 | commands: |
| 3653 | - export VLLM_TEST_CLEAN_GPU_MEMORY=1 |
| 3654 | - >- |
| 3655 | VLLM_ROCM_USE_AITER=1 |
| 3656 | VLLM_ROCM_USE_AITER_CUSTOM_AR=1 |
| 3657 | pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py |
| 3658 | -k 'test_tp2_ar_rms_fp8_fusions and inductor_partition and |
| 3659 | not +quant_fp8 and not +rms_norm and Llama-3 and |
| 3660 | ROCM_AITER_UNIFIED_ATTN' |
| 3661 | - >- |
| 3662 | VLLM_ROCM_USE_AITER=1 |
| 3663 | VLLM_ROCM_USE_AITER_CUSTOM_AR=1 |
| 3664 | pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py |
| 3665 | -k 'test_tp2_ar_rms_fp8_fusions and inductor_partition and |
| 3666 | +quant_fp8 and +rms_norm and qwen3 and |
| 3667 | ROCM_AITER_UNIFIED_ATTN' |
| 3668 | - >- |
| 3669 | VLLM_ROCM_USE_AITER=1 |
| 3670 | VLLM_ROCM_USE_AITER_CUSTOM_AR=1 |
| 3671 | pytest -v -s tests/compile/fusions_e2e/test_tp2_ar_rms.py |
| 3672 | -k 'test_tp2_ar_rms_fp8_fusions and inductor_partition and |
| 3673 | not +quant_fp8 and not +rms_norm and DeepSeek-Coder and |
| 3674 | ROCM_AITER_MLA' |
| 3675 | |
| 3676 | - label: ":amd: (MI355) DSv4-Flash Disaggregated DP EP" # TBD |
| 3677 | timeout_in_minutes: 85 |
| 3678 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3679 | dind: false |
| 3680 | agent_pool: mi355_8 |
| 3681 | num_gpus: 8 |
| 3682 | optional: true |
| 3683 | working_dir: "/vllm-workspace/tests" |
| 3684 | source_file_dependencies: |
| 3685 | - requirements/kv_connectors_rocm.txt |
| 3686 | - vllm/distributed/kv_transfer/kv_connector/v1/nixl/ |
| 3687 | - vllm/model_executor/layers/fused_moe/ |
| 3688 | - vllm/model_executor/layers/quantization/ |
| 3689 | - vllm/models/deepseek_v4/ |
| 3690 | - vllm/v1/attention/ |
| 3691 | - tests/v1/kv_connector/nixl_integration/ |
| 3692 | - vllm/_aiter_ops.py |
| 3693 | - vllm/platforms/rocm.py |
| 3694 | commands: |
| 3695 | - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt |
| 3696 | # A mirror does not inherit the parent env, so it is repeated here in |
| 3697 | # the parent's order. The layouts must stay empty: unset, the script |
| 3698 | # defaults to HND, which the V4 indexer rejects. |
| 3699 | - >- |
| 3700 | VLLM_ROCM_USE_AITER=1 |
| 3701 | ENABLE_HMA_FLAG=1 |
| 3702 | DP_EP=1 |
| 3703 | GPU_MEMORY_UTILIZATION=0.85 |
| 3704 | PREFILLER_TP_SIZE=4 |
| 3705 | DECODER_TP_SIZE=4 |
| 3706 | PREFILL_BLOCK_SIZE=256 |
| 3707 | DECODE_BLOCK_SIZE=256 |
| 3708 | MODEL_NAMES=deepseek-ai/DeepSeek-V4-Flash |
| 3709 | VLLM_ENGINE_READY_TIMEOUT_S=1800 |
| 3710 | VLLM_SERVE_EXTRA_ARGS=--trust-remote-code,--kv-cache-dtype,fp8 |
| 3711 | PREFILLER_KV_LAYOUT= |
| 3712 | DECODER_KV_LAYOUT= |
| 3713 | bash v1/kv_connector/nixl_integration/run_accuracy_test.sh |
| 3714 | |
| 3715 | #---------------------------------------------------------- mi355 · engine -----------------------------------------------------------# |
| 3716 | |
| 3717 | - label: ":amd: (MI355 DPX) Engine Core" # TBD |
| 3718 | timeout_in_minutes: 40 |
| 3719 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3720 | dind: false |
| 3721 | agent_pool: mi355_dpx |
| 3722 | optional: true |
| 3723 | working_dir: "/vllm-workspace/tests" |
| 3724 | source_file_dependencies: |
| 3725 | - vllm/ |
| 3726 | - tests/engine |
| 3727 | - tests/test_sequence |
| 3728 | - tests/test_config |
| 3729 | - tests/test_logger |
| 3730 | - tests/test_vllm_port |
| 3731 | - tests/jit_monitor/test_hooks.py |
| 3732 | - tests/jit_monitor/test_hooks_gpu.py |
| 3733 | - vllm/compilation/ |
| 3734 | - vllm/config/ |
| 3735 | - vllm/engine/ |
| 3736 | - vllm/entrypoints/logger.py |
| 3737 | - vllm/envs.py |
| 3738 | - vllm/logger.py |
| 3739 | - vllm/logging_utils/ |
| 3740 | - vllm/platforms/ |
| 3741 | - vllm/sequence.py |
| 3742 | - vllm/triton_utils/ |
| 3743 | - vllm/utils/ |
| 3744 | commands: |
| 3745 | - 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 |
| 3746 | |
| 3747 | #-------------------------------------------------------- mi355 · entrypoints --------------------------------------------------------# |
| 3748 | |
| 3749 | - label: ":amd: (MI355 DPX) Entrypoints Integration (LLM)" # TBD |
| 3750 | timeout_in_minutes: 90 |
| 3751 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3752 | dind: false |
| 3753 | agent_pool: mi355_dpx |
| 3754 | optional: true |
| 3755 | fast_check: true |
| 3756 | working_dir: "/vllm-workspace/tests" |
| 3757 | source_file_dependencies: |
| 3758 | - vllm/ |
| 3759 | - tests/entrypoints/llm |
| 3760 | commands: |
| 3761 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3762 | - pytest -v -s entrypoints/llm --ignore=entrypoints/llm/test_generate.py --ignore=entrypoints/llm/test_collective_rpc.py --ignore=entrypoints/llm/offline_mode |
| 3763 | - pytest -v -s entrypoints/llm/test_generate.py # it needs a clean process |
| 3764 | - pytest -v -s entrypoints/llm/offline_mode # Needs to avoid interference with other tests |
| 3765 | |
| 3766 | - label: ":amd: (MI355 DPX) Entrypoints Integration (API Server) %N" |
| 3767 | timeout_in_minutes: 65 |
| 3768 | parallelism: 4 |
| 3769 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3770 | dind: false |
| 3771 | agent_pool: mi355_dpx |
| 3772 | optional: true |
| 3773 | fast_check: true |
| 3774 | working_dir: "/vllm-workspace/tests" |
| 3775 | source_file_dependencies: |
| 3776 | - vllm/ |
| 3777 | - tests/entrypoints/serve |
| 3778 | - tests/entrypoints/scale_out |
| 3779 | commands: |
| 3780 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3781 | - pytest -v -s entrypoints/serve --ignore=entrypoints/serve/dev/rpc --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB |
| 3782 | - if [ "$$BUILDKITE_PARALLEL_JOB" = "1" ]; then PYTHONPATH=/vllm-workspace pytest -v -s entrypoints/serve/dev/rpc; fi |
| 3783 | - pytest -v -s entrypoints/scale_out --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB |
| 3784 | |
| 3785 | - label: ":amd: (MI355 DPX) Entrypoints Integration (OpenAI API completion)" # TBD |
| 3786 | timeout_in_minutes: 65 |
| 3787 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3788 | dind: false |
| 3789 | agent_pool: mi355_dpx |
| 3790 | fast_check: true |
| 3791 | optional: true |
| 3792 | working_dir: "/vllm-workspace/tests" |
| 3793 | source_file_dependencies: |
| 3794 | - vllm/ |
| 3795 | - tests/entrypoints/openai |
| 3796 | commands: |
| 3797 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3798 | - pytest -v -s entrypoints/openai/completion --ignore=entrypoints/openai/completion/test_tensorizer_entrypoint.py |
| 3799 | - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/completion --ignore=entrypoints/openai/chat_completion --ignore=entrypoints/openai/responses --ignore=entrypoints/openai/correctness |
| 3800 | |
| 3801 | - label: ":amd: (MI355 DPX) Entrypoints Integration (OpenAI API chat_completion)" # TBD |
| 3802 | timeout_in_minutes: 70 |
| 3803 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3804 | dind: false |
| 3805 | agent_pool: mi355_dpx |
| 3806 | fast_check: true |
| 3807 | optional: true |
| 3808 | working_dir: "/vllm-workspace/tests" |
| 3809 | source_file_dependencies: |
| 3810 | - vllm/ |
| 3811 | - tests/entrypoints/openai |
| 3812 | commands: |
| 3813 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3814 | - pytest -v -s entrypoints/openai/chat_completion |
| 3815 | |
| 3816 | - label: ":amd: (MI355 DPX) Entrypoints Integration (API Server Generate)" # TBD |
| 3817 | timeout_in_minutes: 65 |
| 3818 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3819 | dind: false |
| 3820 | agent_pool: mi355_dpx |
| 3821 | optional: true |
| 3822 | fast_check: true |
| 3823 | working_dir: "/vllm-workspace/tests" |
| 3824 | source_file_dependencies: |
| 3825 | - vllm/ |
| 3826 | - tests/tool_use |
| 3827 | - tests/entrypoints/tool_parsers |
| 3828 | - tests/entrypoints/anthropic |
| 3829 | - tests/entrypoints/cohere |
| 3830 | - tests/entrypoints/generate |
| 3831 | commands: |
| 3832 | - pytest -v -s tool_use |
| 3833 | - pytest -v -s entrypoints/tool_parsers |
| 3834 | - pytest -v -s entrypoints/generate |
| 3835 | - pytest -v -s entrypoints/anthropic |
| 3836 | - pytest -v -s entrypoints/cohere |
| 3837 | |
| 3838 | - label: ":amd: (MI355 DPX) OpenAI API Correctness" # TBD |
| 3839 | timeout_in_minutes: 30 |
| 3840 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3841 | dind: false |
| 3842 | agent_pool: mi355_dpx |
| 3843 | optional: true |
| 3844 | working_dir: "/vllm-workspace/tests" |
| 3845 | source_file_dependencies: |
| 3846 | - csrc/ |
| 3847 | - vllm/entrypoints/openai/ |
| 3848 | - vllm/model_executor/layers/ |
| 3849 | - vllm/v1/attention/backends/ |
| 3850 | - vllm/v1/attention/selector.py |
| 3851 | - vllm/_aiter_ops.py |
| 3852 | - vllm/platforms/rocm.py |
| 3853 | - vllm/model_executor/model_loader/ |
| 3854 | commands: |
| 3855 | - bash ../tools/install_torchcodec_rocm.sh || exit 1 |
| 3856 | - pytest -s entrypoints/openai/correctness/ |
| 3857 | |
| 3858 | - label: ":amd: (MI355 DPX) Entrypoints Integration (Speech to Text)" # TBD |
| 3859 | timeout_in_minutes: 60 |
| 3860 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3861 | dind: false |
| 3862 | agent_pool: mi355_dpx |
| 3863 | fast_check: true |
| 3864 | working_dir: "/vllm-workspace/tests" |
| 3865 | source_file_dependencies: |
| 3866 | - vllm/ |
| 3867 | - tests/entrypoints/speech_to_text |
| 3868 | optional: true |
| 3869 | commands: |
| 3870 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3871 | - pytest -v -s entrypoints/speech_to_text |
| 3872 | |
| 3873 | - label: ":amd: (MI355 DPX) Entrypoints Integration (Multimodal)" |
| 3874 | timeout_in_minutes: 55 |
| 3875 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3876 | dind: false |
| 3877 | agent_pool: mi355_dpx |
| 3878 | fast_check: true |
| 3879 | working_dir: "/vllm-workspace/tests" |
| 3880 | source_file_dependencies: |
| 3881 | - vllm/ |
| 3882 | - tests/entrypoints/multimodal |
| 3883 | optional: true |
| 3884 | commands: |
| 3885 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3886 | - pytest -v -s entrypoints/multimodal |
| 3887 | |
| 3888 | - label: ":amd: (MI355 DPX) Entrypoints Integration (Pooling) Shard %N" # TBD |
| 3889 | timeout_in_minutes: 65 |
| 3890 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3891 | dind: false |
| 3892 | agent_pool: mi355_dpx |
| 3893 | fast_check: true |
| 3894 | parallelism: 4 |
| 3895 | working_dir: "/vllm-workspace/tests" |
| 3896 | source_file_dependencies: |
| 3897 | - vllm/ |
| 3898 | - tests/entrypoints/pooling |
| 3899 | optional: true |
| 3900 | commands: |
| 3901 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3902 | - pytest -v -s entrypoints/pooling --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB |
| 3903 | |
| 3904 | #----------------------------------------------------------- mi355 · evals -----------------------------------------------------------# |
| 3905 | |
| 3906 | - label: ":amd: (MI355 DPX) Multimodal Accuracy Eval (Small Models)" # TBD |
| 3907 | timeout_in_minutes: 35 |
| 3908 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3909 | dind: false |
| 3910 | agent_pool: mi355_dpx |
| 3911 | optional: true |
| 3912 | working_dir: "/vllm-workspace/.buildkite/lm-eval-harness" |
| 3913 | source_file_dependencies: |
| 3914 | - vllm/multimodal/ |
| 3915 | - vllm/inputs/ |
| 3916 | - vllm/v1/core/ |
| 3917 | - vllm/platforms/rocm.py |
| 3918 | - vllm/model_executor/model_loader/ |
| 3919 | commands: |
| 3920 | - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-mm-small.txt --tp-size=1 |
| 3921 | |
| 3922 | - label: ":amd: (MI355) GPQA Eval (GPT-OSS)" # TBD |
| 3923 | timeout_in_minutes: 40 |
| 3924 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3925 | dind: false |
| 3926 | agent_pool: mi355_2 |
| 3927 | num_gpus: 2 |
| 3928 | optional: true |
| 3929 | working_dir: "/vllm-workspace/tests" |
| 3930 | source_file_dependencies: |
| 3931 | - csrc/ |
| 3932 | - vllm/model_executor/layers/quantization |
| 3933 | - vllm/model_executor/models/ |
| 3934 | - vllm/model_executor/model_loader/ |
| 3935 | - vllm/v1/attention/backends/ |
| 3936 | - vllm/v1/attention/selector.py |
| 3937 | - vllm/model_executor/layers/fused_moe/ |
| 3938 | - tests/evals/gpt_oss/ |
| 3939 | - vllm/_aiter_ops.py |
| 3940 | - vllm/platforms/rocm.py |
| 3941 | commands: |
| 3942 | - pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-gfx950.txt |
| 3943 | |
| 3944 | - label: ":amd: (MI355 DPX) LM Eval Small Models" |
| 3945 | timeout_in_minutes: 45 |
| 3946 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3947 | dind: false |
| 3948 | agent_pool: mi355_dpx |
| 3949 | num_gpus: 1 |
| 3950 | optional: true |
| 3951 | working_dir: "/vllm-workspace/tests" |
| 3952 | source_file_dependencies: |
| 3953 | - .buildkite/test_areas/lm_eval.yaml |
| 3954 | - csrc/ |
| 3955 | - tests/evals/gsm8k/ |
| 3956 | - vllm/model_executor/layers/quantization |
| 3957 | - vllm/model_executor/kernels/linear/ |
| 3958 | - vllm/model_executor/layers/fused_moe/ |
| 3959 | - vllm/model_executor/models/ |
| 3960 | - vllm/model_executor/model_loader/ |
| 3961 | - vllm/v1/attention/backends/ |
| 3962 | - vllm/v1/attention/selector.py |
| 3963 | - vllm/_aiter_ops.py |
| 3964 | - vllm/platforms/rocm.py |
| 3965 | commands: |
| 3966 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-gfx950-small.txt |
| 3967 | |
| 3968 | - label: ":amd: (MI355 DPX) LM Eval Watermarking" |
| 3969 | timeout_in_minutes: 45 |
| 3970 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3971 | dind: false |
| 3972 | agent_pool: mi355_dpx |
| 3973 | num_gpus: 1 |
| 3974 | optional: true |
| 3975 | working_dir: "/vllm-workspace/tests" |
| 3976 | source_file_dependencies: |
| 3977 | - .buildkite/test_areas/lm_eval.yaml |
| 3978 | - tests/evals/gsm8k/ |
| 3979 | - vllm/config/watermarking.py |
| 3980 | - vllm/sampling_params.py |
| 3981 | - vllm/v1/watermarking/ |
| 3982 | - vllm/v1/worker/gpu/ |
| 3983 | - vllm/platforms/rocm.py |
| 3984 | commands: |
| 3985 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 3986 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-watermark.txt |
| 3987 | |
| 3988 | - label: ":amd: (MI355) LM Eval Watermark Feature Combination" |
| 3989 | timeout_in_minutes: 60 |
| 3990 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 3991 | dind: false |
| 3992 | agent_pool: mi355_2 |
| 3993 | num_gpus: 2 |
| 3994 | optional: true |
| 3995 | working_dir: "/vllm-workspace/tests" |
| 3996 | source_file_dependencies: |
| 3997 | - .buildkite/test_areas/lm_eval.yaml |
| 3998 | - tests/evals/gsm8k/ |
| 3999 | - vllm/config/speculative.py |
| 4000 | - vllm/config/vllm.py |
| 4001 | - vllm/config/watermarking.py |
| 4002 | - vllm/sampling_params.py |
| 4003 | - vllm/model_executor/models/qwen3_dspark.py |
| 4004 | - vllm/model_executor/models/qwen3_dflash.py |
| 4005 | - vllm/model_executor/warmup/ |
| 4006 | - vllm/v1/watermarking/ |
| 4007 | - vllm/v1/worker/gpu/ |
| 4008 | - vllm/platforms/rocm.py |
| 4009 | commands: |
| 4010 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 4011 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-watermark-dspark-tp2.txt |
| 4012 | |
| 4013 | - label: ":amd: (MI355) LM Eval Qwen3-5 Models" # TBD |
| 4014 | timeout_in_minutes: 75 |
| 4015 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4016 | dind: false |
| 4017 | agent_pool: mi355_2 |
| 4018 | num_gpus: 2 |
| 4019 | optional: true |
| 4020 | working_dir: "/vllm-workspace/tests" |
| 4021 | source_file_dependencies: |
| 4022 | - vllm/model_executor/models/qwen3_5.py |
| 4023 | - vllm/model_executor/models/qwen3_5_mtp.py |
| 4024 | - vllm/transformers_utils/configs/qwen3_5.py |
| 4025 | - vllm/transformers_utils/configs/qwen3_5_moe.py |
| 4026 | - vllm/model_executor/models/qwen2.py |
| 4027 | - vllm/model_executor/models/qwen3.py |
| 4028 | - vllm/model_executor/models/qwen3_next.py |
| 4029 | - vllm/model_executor/models/qwen3_next_mtp.py |
| 4030 | - vllm/distributed/ |
| 4031 | - vllm/model_executor/layers/fused_moe/ |
| 4032 | - vllm/third_party/flash_linear_attention/ops/ |
| 4033 | - vllm/model_executor/layers/quantization/modelopt.py |
| 4034 | - vllm/model_executor/kernels/linear/nvfp4/ |
| 4035 | - vllm/v1/attention/ |
| 4036 | - vllm/v1/worker/ |
| 4037 | - tests/evals/gsm8k/ |
| 4038 | - tests/utils.py |
| 4039 | - vllm/_aiter_ops.py |
| 4040 | - vllm/platforms/rocm.py |
| 4041 | commands: |
| 4042 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-qwen35-mi355.txt |
| 4043 | |
| 4044 | - label: ":amd: (MI355) LM Eval Large Models EP" |
| 4045 | timeout_in_minutes: 120 |
| 4046 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4047 | dind: false |
| 4048 | agent_pool: mi355_2 |
| 4049 | num_gpus: 2 |
| 4050 | optional: true |
| 4051 | working_dir: "/vllm-workspace/tests" |
| 4052 | source_file_dependencies: |
| 4053 | - .buildkite/test_areas/lm_eval.yaml |
| 4054 | - tests/evals/gsm8k/ |
| 4055 | - tests/utils.py |
| 4056 | - tests/conftest.py |
| 4057 | - csrc/ |
| 4058 | - vllm/config/ |
| 4059 | - vllm/distributed/ |
| 4060 | - vllm/model_executor/models/config.py |
| 4061 | - vllm/model_executor/models/registry.py |
| 4062 | - vllm/model_executor/models/utils.py |
| 4063 | - vllm/model_executor/models/qwen3_next.py |
| 4064 | - vllm/model_executor/models/qwen3_next_mtp.py |
| 4065 | - vllm/model_executor/models/nemotron_h.py |
| 4066 | - vllm/model_executor/models/nemotron_h_mtp.py |
| 4067 | - vllm/transformers_utils/configs/qwen3_next.py |
| 4068 | - vllm/transformers_utils/configs/nemotron_h.py |
| 4069 | - vllm/model_executor/model_loader/ |
| 4070 | - vllm/model_executor/layers/quantization/ |
| 4071 | - vllm/model_executor/layers/fused_moe/ |
| 4072 | - vllm/model_executor/layers/mamba/ |
| 4073 | - vllm/model_executor/layers/layernorm.py |
| 4074 | - vllm/model_executor/layers/linear.py |
| 4075 | - vllm/model_executor/kernels/linear/ |
| 4076 | - vllm/third_party/flash_linear_attention/ops/ |
| 4077 | - vllm/v1/attention/ |
| 4078 | - vllm/v1/spec_decode/ |
| 4079 | - vllm/v1/worker/ |
| 4080 | - vllm/_aiter_ops.py |
| 4081 | - vllm/platforms/rocm.py |
| 4082 | commands: |
| 4083 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 4084 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-blackwell-ep-mi355.txt |
| 4085 | |
| 4086 | - label: ":amd: (MI355) Qwen3-8-Flash-Next-FP8 Accuracy Eval" |
| 4087 | timeout_in_minutes: 105 |
| 4088 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4089 | dind: false |
| 4090 | agent_pool: mi355_4 |
| 4091 | num_gpus: 4 |
| 4092 | optional: true |
| 4093 | working_dir: "/vllm-workspace/tests" |
| 4094 | source_file_dependencies: |
| 4095 | - csrc/libtorch_stable/gdn/ |
| 4096 | - tests/evals/qwen4_exp/ |
| 4097 | - vllm/model_executor/layers/mamba/ |
| 4098 | - vllm/models/qwen4_exp/ |
| 4099 | - vllm/transformers_utils/configs/qwen4_exp.py |
| 4100 | - vllm/v1/attention/backends/short_conv_attn.py |
| 4101 | - vllm/platforms/rocm.py |
| 4102 | commands: |
| 4103 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 4104 | - uv pip install --system 'evalscope==1.10.0' |
| 4105 | - pytest -s -v evals/qwen4_exp/test_accuracy.py --config-list-file=configs/models-rocm.txt |
| 4106 | |
| 4107 | - label: ":amd: (MI355 DPX) LM Eval TurboQuant k8v4" |
| 4108 | timeout_in_minutes: 30 |
| 4109 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4110 | dind: false |
| 4111 | agent_pool: mi355_dpx |
| 4112 | num_gpus: 1 |
| 4113 | optional: true |
| 4114 | working_dir: "/vllm-workspace/tests" |
| 4115 | source_file_dependencies: |
| 4116 | - tests/evals/gsm8k/ |
| 4117 | - tests/utils.py |
| 4118 | - tests/conftest.py |
| 4119 | - vllm/model_executor/layers/quantization/turboquant/ |
| 4120 | - vllm/v1/attention/backends/turboquant_attn.py |
| 4121 | - vllm/v1/attention/backends/triton_attn.py |
| 4122 | - vllm/v1/attention/ops/triton_unified_attention.py |
| 4123 | - vllm/v1/attention/selector.py |
| 4124 | - vllm/v1/attention/ops/triton_turboquant_decode.py |
| 4125 | - vllm/v1/attention/ops/triton_turboquant_store.py |
| 4126 | - vllm/v1/attention/ops/flydsl_turboquant_decode.py |
| 4127 | - vllm/v1/attention/ops/flydsl_kernels/ |
| 4128 | - vllm/v1/attention/ops/turboquant_soa/ |
| 4129 | - vllm/platforms/rocm.py |
| 4130 | commands: |
| 4131 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant-k8v4-rocm.txt |
| 4132 | |
| 4133 | - label: ":amd: (MI355 DPX) LM Eval TurboQuant t4nc" |
| 4134 | timeout_in_minutes: 30 |
| 4135 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4136 | dind: false |
| 4137 | agent_pool: mi355_dpx |
| 4138 | num_gpus: 1 |
| 4139 | optional: true |
| 4140 | working_dir: "/vllm-workspace/tests" |
| 4141 | source_file_dependencies: |
| 4142 | - tests/evals/gsm8k/ |
| 4143 | - tests/utils.py |
| 4144 | - tests/conftest.py |
| 4145 | - vllm/model_executor/layers/quantization/turboquant/ |
| 4146 | - vllm/v1/attention/backends/turboquant_attn.py |
| 4147 | - vllm/v1/attention/backends/triton_attn.py |
| 4148 | - vllm/v1/attention/ops/triton_unified_attention.py |
| 4149 | - vllm/v1/attention/selector.py |
| 4150 | - vllm/v1/attention/ops/triton_turboquant_decode.py |
| 4151 | - vllm/v1/attention/ops/triton_turboquant_store.py |
| 4152 | - vllm/v1/attention/ops/flydsl_turboquant_decode.py |
| 4153 | - vllm/v1/attention/ops/flydsl_kernels/ |
| 4154 | - vllm/v1/attention/ops/turboquant_soa/ |
| 4155 | - vllm/platforms/rocm.py |
| 4156 | commands: |
| 4157 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant-t4nc-rocm.txt |
| 4158 | |
| 4159 | - label: ":amd: (MI355 DPX) LM Eval TurboQuant k3v4nc" |
| 4160 | timeout_in_minutes: 30 |
| 4161 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4162 | dind: false |
| 4163 | agent_pool: mi355_dpx |
| 4164 | num_gpus: 1 |
| 4165 | optional: true |
| 4166 | working_dir: "/vllm-workspace/tests" |
| 4167 | source_file_dependencies: |
| 4168 | - tests/evals/gsm8k/ |
| 4169 | - tests/utils.py |
| 4170 | - tests/conftest.py |
| 4171 | - vllm/model_executor/layers/quantization/turboquant/ |
| 4172 | - vllm/v1/attention/backends/turboquant_attn.py |
| 4173 | - vllm/v1/attention/backends/triton_attn.py |
| 4174 | - vllm/v1/attention/ops/triton_unified_attention.py |
| 4175 | - vllm/v1/attention/selector.py |
| 4176 | - vllm/v1/attention/ops/triton_turboquant_decode.py |
| 4177 | - vllm/v1/attention/ops/triton_turboquant_store.py |
| 4178 | - vllm/v1/attention/ops/flydsl_turboquant_decode.py |
| 4179 | - vllm/v1/attention/ops/flydsl_kernels/ |
| 4180 | - vllm/v1/attention/ops/turboquant_soa/ |
| 4181 | - vllm/platforms/rocm.py |
| 4182 | commands: |
| 4183 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant-k3v4nc-rocm.txt |
| 4184 | |
| 4185 | - label: ":amd: (MI355 DPX) LM Eval TurboQuant t3nc" |
| 4186 | timeout_in_minutes: 30 |
| 4187 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4188 | dind: false |
| 4189 | agent_pool: mi355_dpx |
| 4190 | num_gpus: 1 |
| 4191 | optional: true |
| 4192 | working_dir: "/vllm-workspace/tests" |
| 4193 | source_file_dependencies: |
| 4194 | - .buildkite/test-amd.yaml |
| 4195 | - tests/evals/gsm8k/ |
| 4196 | - tests/utils.py |
| 4197 | - tests/conftest.py |
| 4198 | - vllm/model_executor/layers/quantization/turboquant/ |
| 4199 | - vllm/v1/attention/backends/turboquant_attn.py |
| 4200 | - vllm/v1/attention/backends/triton_attn.py |
| 4201 | - vllm/v1/attention/ops/triton_unified_attention.py |
| 4202 | - vllm/v1/attention/selector.py |
| 4203 | - vllm/v1/attention/ops/triton_turboquant_decode.py |
| 4204 | - vllm/v1/attention/ops/triton_turboquant_store.py |
| 4205 | - vllm/v1/attention/ops/flydsl_turboquant_decode.py |
| 4206 | - vllm/v1/attention/ops/flydsl_kernels/ |
| 4207 | - vllm/v1/attention/ops/turboquant_soa/ |
| 4208 | - vllm/platforms/rocm.py |
| 4209 | commands: |
| 4210 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/models-turboquant-t3nc-rocm.txt |
| 4211 | |
| 4212 | - label: ":amd: (MI355) LM Eval Spec Decode (Fixed-length FP8)" # TBD |
| 4213 | timeout_in_minutes: 145 |
| 4214 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4215 | dind: false |
| 4216 | agent_pool: mi355_4 |
| 4217 | num_gpus: 4 |
| 4218 | optional: true |
| 4219 | working_dir: "/vllm-workspace/tests" |
| 4220 | source_file_dependencies: |
| 4221 | - tests/evals/gsm8k/ |
| 4222 | - tests/utils.py |
| 4223 | - vllm/distributed/ |
| 4224 | - vllm/models/deepseek_v4/ |
| 4225 | - vllm/model_executor/models/qwen3_dspark.py |
| 4226 | - vllm/model_executor/layers/fused_moe/ |
| 4227 | - vllm/v1/spec_decode/metrics.py |
| 4228 | - vllm/v1/worker/gpu/spec_decode/dspark/ |
| 4229 | - vllm/v1/worker/gpu/spec_decode/rejection_sampler.py |
| 4230 | - vllm/v1/worker/gpu/spec_decode/speculator.py |
| 4231 | - vllm/model_executor/layers/quantization/ |
| 4232 | - vllm/v1/attention/ |
| 4233 | - vllm/_aiter_ops.py |
| 4234 | - vllm/platforms/rocm.py |
| 4235 | commands: |
| 4236 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-spec-decode-rocm.txt |
| 4237 | |
| 4238 | - label: ':amd: (MI355) LM Eval Small Models Distributed' |
| 4239 | timeout_in_minutes: 145 |
| 4240 | mirror_hardwares: |
| 4241 | - amdexperimental |
| 4242 | - amdproduction |
| 4243 | - amdgfx950nightly |
| 4244 | - amdmi355 |
| 4245 | dind: false |
| 4246 | agent_pool: mi355_2 |
| 4247 | num_gpus: 2 |
| 4248 | optional: true |
| 4249 | working_dir: /vllm-workspace/tests |
| 4250 | source_file_dependencies: |
| 4251 | - csrc/ |
| 4252 | - vllm/model_executor/layers/quantization |
| 4253 | - vllm/model_executor/models/ |
| 4254 | - vllm/model_executor/model_loader/ |
| 4255 | - vllm/v1/attention/backends/ |
| 4256 | - vllm/v1/attention/selector.py |
| 4257 | - vllm/v1/worker/ |
| 4258 | - vllm/v1/core/ |
| 4259 | - vllm/config/ |
| 4260 | - tests/evals/gsm8k/configs/models-mi3xx-fp8-and-mixed.txt |
| 4261 | - vllm/_aiter_ops.py |
| 4262 | - vllm/platforms/rocm.py |
| 4263 | - tests/evals/gsm8k/ |
| 4264 | - tests/utils.py |
| 4265 | - vllm/distributed/ |
| 4266 | - vllm/transformers_utils/configs/diffusion_gemma.py |
| 4267 | - vllm/v1/attention/ops/ |
| 4268 | - tests/evals/gsm8k/configs/models-small-tp-mi355.txt |
| 4269 | commands: |
| 4270 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-mi3xx-fp8-and-mixed.txt |
| 4271 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-small-tp.txt |
| 4272 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-small-tp-mi355.txt |
| 4273 | |
| 4274 | - label: ":amd: (MI355) MoE Refactor Integration Shard %N" |
| 4275 | timeout_in_minutes: 105 |
| 4276 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4277 | dind: false |
| 4278 | agent_pool: mi355_2 |
| 4279 | num_gpus: 2 |
| 4280 | parallelism: 4 |
| 4281 | optional: true |
| 4282 | working_dir: "/vllm-workspace/tests" |
| 4283 | source_file_dependencies: |
| 4284 | - vllm/platforms/rocm.py |
| 4285 | - tests/evals/gsm8k/ |
| 4286 | - vllm/model_executor/layers/fused_moe/ |
| 4287 | - vllm/model_executor/layers/quantization/ |
| 4288 | - vllm/_aiter_ops.py |
| 4289 | commands: |
| 4290 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 4291 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=evals/gsm8k/configs/moe-refactor/config-rocm-shard-$$BUILDKITE_PARALLEL_JOB.txt |
| 4292 | |
| 4293 | |
| 4294 | - label: ":amd: (MI355) Qwen3-30B-A3B-FP8-block Sync EPLB Accuracy" # TBD |
| 4295 | timeout_in_minutes: 25 |
| 4296 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4297 | dind: false |
| 4298 | agent_pool: mi355_2 |
| 4299 | num_gpus: 2 |
| 4300 | optional: true |
| 4301 | working_dir: "/vllm-workspace" |
| 4302 | source_file_dependencies: |
| 4303 | - vllm/model_executor/models/ |
| 4304 | - vllm/model_executor/model_loader/ |
| 4305 | - vllm/model_executor/layers/quantization/ |
| 4306 | - vllm/model_executor/layers/fused_moe/ |
| 4307 | - vllm/distributed/eplb |
| 4308 | - vllm/v1/attention/backends/ |
| 4309 | - vllm/v1/attention/selector.py |
| 4310 | - .buildkite/scripts/scheduled_integration_test/ |
| 4311 | - vllm/_aiter_ops.py |
| 4312 | - vllm/platforms/rocm.py |
| 4313 | commands: |
| 4314 | - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_block_ep_eplb.sh 0.8 200 8020 2 1 |
| 4315 | |
| 4316 | - label: ":amd: (MI355) LM Eval Large Models FP8" # TBD |
| 4317 | timeout_in_minutes: 45 |
| 4318 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4319 | dind: false |
| 4320 | agent_pool: mi355_4 |
| 4321 | num_gpus: 4 |
| 4322 | optional: true |
| 4323 | working_dir: "/vllm-workspace/.buildkite/lm-eval-harness" |
| 4324 | source_file_dependencies: |
| 4325 | - csrc/ |
| 4326 | - vllm/model_executor/layers/quantization |
| 4327 | - vllm/model_executor/models/ |
| 4328 | - vllm/model_executor/model_loader/ |
| 4329 | - vllm/v1/attention/backends/ |
| 4330 | - vllm/v1/attention/selector.py |
| 4331 | - vllm/_aiter_ops.py |
| 4332 | - vllm/platforms/rocm.py |
| 4333 | commands: |
| 4334 | - export VLLM_USE_DEEP_GEMM=0 |
| 4335 | - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large-rocm-tp4.txt --tp-size=4 |
| 4336 | |
| 4337 | - label: ":amd: (MI355) LM Eval Large Models" # TBD |
| 4338 | timeout_in_minutes: 65 |
| 4339 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4340 | dind: false |
| 4341 | agent_pool: mi355_8 |
| 4342 | optional: true |
| 4343 | num_gpus: 8 |
| 4344 | working_dir: "/vllm-workspace/tests" |
| 4345 | source_file_dependencies: |
| 4346 | - vllm/model_executor/models/ |
| 4347 | - vllm/model_executor/model_loader/ |
| 4348 | - vllm/model_executor/layers/quantization/ |
| 4349 | - vllm/v1/attention/backends/ |
| 4350 | - vllm/v1/attention/selector.py |
| 4351 | - vllm/model_executor/layers/layernorm.py |
| 4352 | - csrc/ |
| 4353 | - vllm/_aiter_ops.py |
| 4354 | - vllm/platforms/rocm.py |
| 4355 | - tests/evals/ |
| 4356 | commands: |
| 4357 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 4358 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-gfx950-large.txt |
| 4359 | |
| 4360 | - label: ":amd: (MI355) LM Eval DeepSeek R1 TP" |
| 4361 | timeout_in_minutes: 65 |
| 4362 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4363 | dind: false |
| 4364 | agent_pool: mi355_4 |
| 4365 | optional: true |
| 4366 | num_gpus: 4 |
| 4367 | working_dir: "/vllm-workspace/tests" |
| 4368 | source_file_dependencies: |
| 4369 | - .buildkite/test_areas/lm_eval.yaml |
| 4370 | - tests/evals/gsm8k/ |
| 4371 | - tests/utils.py |
| 4372 | - tests/conftest.py |
| 4373 | - vllm/config/ |
| 4374 | - vllm/distributed/ |
| 4375 | - vllm/model_executor/models/ |
| 4376 | - vllm/models/ |
| 4377 | - vllm/model_executor/model_loader/ |
| 4378 | - vllm/model_executor/layers/quantization/ |
| 4379 | - vllm/model_executor/layers/fused_moe/ |
| 4380 | - vllm/v1/attention/ |
| 4381 | - vllm/v1/spec_decode/ |
| 4382 | - vllm/v1/worker/gpu/spec_decode/ |
| 4383 | - vllm/v1/core/kv_cache_utils.py |
| 4384 | - vllm/model_executor/layers/layernorm.py |
| 4385 | - csrc/ |
| 4386 | - vllm/_aiter_ops.py |
| 4387 | - vllm/platforms/rocm.py |
| 4388 | - vllm/model_executor/layers/sparse_attn_indexer.py |
| 4389 | commands: |
| 4390 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 4391 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-mi355-deepseek-r1-tp.txt |
| 4392 | |
| 4393 | - label: ":amd: (MI355) LM Eval DeepSeek R1 DP" |
| 4394 | timeout_in_minutes: 65 |
| 4395 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4396 | dind: false |
| 4397 | agent_pool: mi355_4 |
| 4398 | optional: true |
| 4399 | num_gpus: 4 |
| 4400 | working_dir: "/vllm-workspace/tests" |
| 4401 | source_file_dependencies: |
| 4402 | - .buildkite/test_areas/lm_eval.yaml |
| 4403 | - tests/evals/gsm8k/ |
| 4404 | - tests/utils.py |
| 4405 | - tests/conftest.py |
| 4406 | - vllm/config/ |
| 4407 | - vllm/distributed/ |
| 4408 | - vllm/model_executor/models/ |
| 4409 | - vllm/models/ |
| 4410 | - vllm/model_executor/model_loader/ |
| 4411 | - vllm/model_executor/layers/quantization/ |
| 4412 | - vllm/model_executor/layers/fused_moe/ |
| 4413 | - vllm/v1/attention/ |
| 4414 | - vllm/v1/spec_decode/ |
| 4415 | - vllm/v1/worker/gpu/spec_decode/ |
| 4416 | - vllm/v1/core/kv_cache_utils.py |
| 4417 | - vllm/model_executor/layers/layernorm.py |
| 4418 | - csrc/ |
| 4419 | - vllm/_aiter_ops.py |
| 4420 | - vllm/platforms/rocm.py |
| 4421 | - vllm/model_executor/layers/sparse_attn_indexer.py |
| 4422 | commands: |
| 4423 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 4424 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-mi355-deepseek-r1-dp.txt |
| 4425 | |
| 4426 | - label: ":amd: (MI355) LM Eval DeepSeek V3-2 TP" |
| 4427 | timeout_in_minutes: 65 |
| 4428 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4429 | dind: false |
| 4430 | agent_pool: mi355_4 |
| 4431 | optional: true |
| 4432 | num_gpus: 4 |
| 4433 | working_dir: "/vllm-workspace/tests" |
| 4434 | source_file_dependencies: |
| 4435 | - .buildkite/test_areas/lm_eval.yaml |
| 4436 | - tests/evals/gsm8k/ |
| 4437 | - tests/utils.py |
| 4438 | - tests/conftest.py |
| 4439 | - vllm/config/ |
| 4440 | - vllm/distributed/ |
| 4441 | - vllm/model_executor/models/ |
| 4442 | - vllm/models/ |
| 4443 | - vllm/model_executor/model_loader/ |
| 4444 | - vllm/model_executor/layers/quantization/ |
| 4445 | - vllm/model_executor/layers/fused_moe/ |
| 4446 | - vllm/v1/attention/ |
| 4447 | - vllm/v1/spec_decode/ |
| 4448 | - vllm/v1/worker/gpu/spec_decode/ |
| 4449 | - vllm/v1/core/kv_cache_utils.py |
| 4450 | - vllm/model_executor/layers/layernorm.py |
| 4451 | - csrc/ |
| 4452 | - vllm/_aiter_ops.py |
| 4453 | - vllm/platforms/rocm.py |
| 4454 | - vllm/model_executor/layers/sparse_attn_indexer.py |
| 4455 | commands: |
| 4456 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 4457 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-mi355-deepseek-v32-tp.txt |
| 4458 | |
| 4459 | - label: ":amd: (MI355) LM Eval DeepSeek V3-2 DP" |
| 4460 | timeout_in_minutes: 65 |
| 4461 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4462 | dind: false |
| 4463 | agent_pool: mi355_4 |
| 4464 | optional: true |
| 4465 | num_gpus: 4 |
| 4466 | working_dir: "/vllm-workspace/tests" |
| 4467 | source_file_dependencies: |
| 4468 | - .buildkite/test_areas/lm_eval.yaml |
| 4469 | - tests/evals/gsm8k/ |
| 4470 | - tests/utils.py |
| 4471 | - tests/conftest.py |
| 4472 | - vllm/config/ |
| 4473 | - vllm/distributed/ |
| 4474 | - vllm/model_executor/models/ |
| 4475 | - vllm/models/ |
| 4476 | - vllm/model_executor/model_loader/ |
| 4477 | - vllm/model_executor/layers/quantization/ |
| 4478 | - vllm/model_executor/layers/fused_moe/ |
| 4479 | - vllm/v1/attention/ |
| 4480 | - vllm/v1/spec_decode/ |
| 4481 | - vllm/v1/worker/gpu/spec_decode/ |
| 4482 | - vllm/v1/core/kv_cache_utils.py |
| 4483 | - vllm/model_executor/layers/layernorm.py |
| 4484 | - csrc/ |
| 4485 | - vllm/_aiter_ops.py |
| 4486 | - vllm/platforms/rocm.py |
| 4487 | - vllm/model_executor/layers/sparse_attn_indexer.py |
| 4488 | commands: |
| 4489 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 4490 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-mi355-deepseek-v32-dp.txt |
| 4491 | |
| 4492 | - label: ":amd: (MI355) LM Eval Nemotron 3 Super" |
| 4493 | timeout_in_minutes: 65 |
| 4494 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4495 | dind: false |
| 4496 | agent_pool: mi355_4 |
| 4497 | optional: true |
| 4498 | num_gpus: 4 |
| 4499 | working_dir: "/vllm-workspace/tests" |
| 4500 | source_file_dependencies: |
| 4501 | - .buildkite/test_areas/lm_eval.yaml |
| 4502 | - tests/evals/gsm8k/ |
| 4503 | - tests/utils.py |
| 4504 | - tests/conftest.py |
| 4505 | - vllm/config/ |
| 4506 | - vllm/distributed/ |
| 4507 | - vllm/model_executor/models/ |
| 4508 | - vllm/models/ |
| 4509 | - vllm/model_executor/model_loader/ |
| 4510 | - vllm/model_executor/layers/quantization/ |
| 4511 | - vllm/model_executor/layers/fused_moe/ |
| 4512 | - vllm/v1/attention/ |
| 4513 | - vllm/v1/spec_decode/ |
| 4514 | - vllm/v1/worker/gpu/spec_decode/ |
| 4515 | - vllm/v1/core/kv_cache_utils.py |
| 4516 | - vllm/model_executor/layers/layernorm.py |
| 4517 | - csrc/ |
| 4518 | - vllm/_aiter_ops.py |
| 4519 | - vllm/platforms/rocm.py |
| 4520 | - vllm/model_executor/layers/mamba/ |
| 4521 | commands: |
| 4522 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 4523 | - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-mi355-nemotron-3-super.txt |
| 4524 | |
| 4525 | #--------------------------------------------------------- mi355 · examples ----------------------------------------------------------# |
| 4526 | |
| 4527 | - label: ":amd: (MI355 DPX) Examples" # TBD |
| 4528 | timeout_in_minutes: 75 |
| 4529 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4530 | dind: false |
| 4531 | agent_pool: mi355_dpx |
| 4532 | optional: true |
| 4533 | working_dir: "/vllm-workspace/examples" |
| 4534 | source_file_dependencies: |
| 4535 | - vllm/entrypoints |
| 4536 | - vllm/multimodal |
| 4537 | - examples/ |
| 4538 | - vllm/platforms/rocm.py |
| 4539 | commands: |
| 4540 | # Basic |
| 4541 | # Multi-modal models |
| 4542 | # Pooling models |
| 4543 | # Features demo |
| 4544 | - pip install --no-deps tensorizer |
| 4545 | - python3 basic/offline_inference/chat.py |
| 4546 | - python3 basic/offline_inference/generate.py --model facebook/opt-125m |
| 4547 | - python3 basic/offline_inference/generate.py --model meta-llama/Llama-2-13b-chat-hf --cpu-offload-gb 10 |
| 4548 | - python3 basic/offline_inference/classify.py |
| 4549 | - python3 basic/offline_inference/embed.py |
| 4550 | - python3 basic/offline_inference/score.py |
| 4551 | - python3 generate/multimodal/audio_language_offline.py --seed 0 |
| 4552 | - python3 generate/multimodal/vision_language_offline.py --seed 0 |
| 4553 | - python3 generate/multimodal/vision_language_multi_image_offline.py --seed 0 |
| 4554 | - python3 generate/multimodal/encoder_decoder_multimodal_offline.py --model-type whisper --seed 0 |
| 4555 | - python3 pooling/embed/vision_embedding_offline.py --seed 0 |
| 4556 | - python3 features/automatic_prefix_caching/prefix_caching_offline.py |
| 4557 | - python3 deployment/llm_engine_example.py |
| 4558 | - python3 features/tensorize_vllm_model.py --model facebook/opt-125m serialize --serialized-directory /tmp/ --suffix v1 && python3 features/tensorize_vllm_model.py --model facebook/opt-125m deserialize --path-to-tensors /tmp/vllm/facebook/opt-125m/v1/model.tensors |
| 4559 | - python3 features/speculative_decoding/spec_decode_offline.py --test --method eagle --num_spec_tokens 3 --dataset-name hf --dataset-path philschmid/mt-bench --num-prompts 80 --temp 0 --top-p 1.0 --top-k -1 --tp 1 --enable-chunked-prefill --max-model-len 2048 |
| 4560 | - python3 features/speculative_decoding/spec_decode_offline.py --test --method eagle3 --num_spec_tokens 3 --dataset-name hf --dataset-path philschmid/mt-bench --num-prompts 80 --temp 0 --top-p 1.0 --top-k -1 --tp 1 --enable-chunked-prefill --max-model-len 1536 |
| 4561 | |
| 4562 | #---------------------------------------------------------- mi355 · kernels ----------------------------------------------------------# |
| 4563 | |
| 4564 | - label: ":amd: (MI355) mHC Kernels" |
| 4565 | timeout_in_minutes: 50 |
| 4566 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4567 | dind: false |
| 4568 | agent_pool: mi355_1 |
| 4569 | optional: true |
| 4570 | working_dir: "/vllm-workspace/tests" |
| 4571 | source_file_dependencies: |
| 4572 | - csrc/ |
| 4573 | - vllm/ |
| 4574 | - tests/kernels/ |
| 4575 | - vllm/platforms/rocm.py |
| 4576 | commands: |
| 4577 | - pytest -v -s kernels/mhc |
| 4578 | |
| 4579 | - label: ":amd: (MI355) FLA Ops Kernels" |
| 4580 | timeout_in_minutes: 50 |
| 4581 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4582 | dind: false |
| 4583 | agent_pool: mi355_1 |
| 4584 | optional: true |
| 4585 | working_dir: "/vllm-workspace/tests" |
| 4586 | source_file_dependencies: |
| 4587 | - csrc/ |
| 4588 | - vllm/ |
| 4589 | - tests/kernels/ |
| 4590 | - vllm/platforms/rocm.py |
| 4591 | commands: |
| 4592 | - pytest -v -s kernels/fla |
| 4593 | |
| 4594 | - label: ":amd: (MI355) Miscellaneous Kernels" |
| 4595 | timeout_in_minutes: 55 |
| 4596 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4597 | dind: false |
| 4598 | agent_pool: mi355_1 |
| 4599 | optional: true |
| 4600 | working_dir: "/vllm-workspace/tests" |
| 4601 | source_file_dependencies: |
| 4602 | - csrc/ |
| 4603 | - vllm/ |
| 4604 | - tests/kernels/ |
| 4605 | - vllm/platforms/rocm.py |
| 4606 | commands: |
| 4607 | - pytest -v -s |
| 4608 | kernels/test_apply_repetition_penalties.py |
| 4609 | kernels/test_awq_int4_to_int8.py |
| 4610 | kernels/test_bf16_skinny_gemm.py |
| 4611 | kernels/test_cache_kernels.py |
| 4612 | kernels/test_compressor_kv_cache.py |
| 4613 | kernels/test_cp_gather_fp8.py |
| 4614 | kernels/test_deepseek_v4_cpu_kernels.py |
| 4615 | kernels/test_dsv41_mega_attn_layouts.py |
| 4616 | kernels/test_engram.py |
| 4617 | kernels/test_flashinfer_bf16_gemm.py |
| 4618 | kernels/test_flex_attention.py |
| 4619 | kernels/test_fp32_router_gemm.py |
| 4620 | kernels/test_fused_deepseek_v32_norm_rope.py |
| 4621 | kernels/test_fused_indexer_q_rope_quant.py |
| 4622 | kernels/test_fused_inv_rope_fp8_quant.py |
| 4623 | kernels/test_fused_minimax_m3_qknorm_rope_kv_insert.py |
| 4624 | kernels/test_fused_quant_activation.py |
| 4625 | kernels/test_gate_linear_rocm_dispatch.py |
| 4626 | kernels/test_gemm_rs_ar.py |
| 4627 | kernels/test_kpool_decode_update_batched.py |
| 4628 | kernels/test_minimax_m3_amd_ops.py |
| 4629 | kernels/test_minimax_m3_sparse_attn_fp8_scale.py |
| 4630 | kernels/test_onednn.py |
| 4631 | kernels/test_relu2_fp8_quant.py |
| 4632 | kernels/test_rocm_fp32_router_gemm.py |
| 4633 | --deselect=tests/kernels/test_bf16_skinny_gemm.py::test_every_dsv3_routed_shape_is_instantiated |
| 4634 | |
| 4635 | - label: ":amd: (MI355 DPX) Attention DiffKV Kernels" |
| 4636 | timeout_in_minutes: 35 |
| 4637 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4638 | dind: false |
| 4639 | agent_pool: mi355_dpx |
| 4640 | num_gpus: 1 |
| 4641 | optional: true |
| 4642 | working_dir: "/vllm-workspace/tests" |
| 4643 | source_file_dependencies: |
| 4644 | - vllm/v1/attention/ops/triton_unified_attention_diffkv.py |
| 4645 | - vllm/v1/attention/backends/triton_attn_diffkv.py |
| 4646 | - tests/kernels/attention/test_triton_unified_attention.py |
| 4647 | - tests/kernels/attention/test_triton_unified_attention_diffkv.py |
| 4648 | - vllm/platforms/rocm.py |
| 4649 | commands: |
| 4650 | - pytest -v -s kernels/attention/test_triton_unified_attention_diffkv.py |
| 4651 | |
| 4652 | - label: ":amd: (MI355) MiniMax Reduce RMS Kernels" # TBD |
| 4653 | timeout_in_minutes: 45 |
| 4654 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4655 | dind: false |
| 4656 | agent_pool: mi355_2 |
| 4657 | num_gpus: 2 |
| 4658 | working_dir: "/vllm-workspace/tests" |
| 4659 | source_file_dependencies: |
| 4660 | - vllm/model_executor/layers/minimax_rms_norm/ |
| 4661 | - vllm/distributed/ |
| 4662 | - vllm/_aiter_ops.py |
| 4663 | - vllm/envs.py |
| 4664 | - vllm/platforms/rocm.py |
| 4665 | - tests/kernels/core/test_minimax_reduce_rms.py |
| 4666 | - tests/utils.py |
| 4667 | - csrc/libtorch_stable/minimax_reduce_rms_kernel.cu |
| 4668 | - csrc/libtorch_stable/minimax_reduce_rms_kernel.h |
| 4669 | - csrc/libtorch_stable/ops.h |
| 4670 | - csrc/libtorch_stable/torch_bindings.cpp |
| 4671 | - vllm/model_executor/layers/mamba/linear/minimax_linear_attn.py |
| 4672 | optional: true |
| 4673 | commands: |
| 4674 | - pytest -v -s kernels/core/test_minimax_reduce_rms.py |
| 4675 | |
| 4676 | - label: ":amd: (MI355 DPX) Kernels" # TBD |
| 4677 | timeout_in_minutes: 25 |
| 4678 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4679 | dind: false |
| 4680 | agent_pool: mi355_dpx |
| 4681 | working_dir: "/vllm-workspace/" |
| 4682 | source_file_dependencies: |
| 4683 | - csrc/quantization/fp4/ |
| 4684 | - csrc/attention/mla/ |
| 4685 | - csrc/quantization/cutlass_w8a8/moe/ |
| 4686 | - vllm/model_executor/layers/fused_moe/cutlass_moe.py |
| 4687 | - vllm/v1/attention/backends/triton_attn.py |
| 4688 | - vllm/v1/attention/backends/rocm_attn.py |
| 4689 | - vllm/v1/attention/backends/rocm_aiter_fa.py |
| 4690 | - vllm/v1/attention/backends/rocm_aiter_unified_attn.py |
| 4691 | - vllm/v1/attention/backends/mla/aiter_triton_mla.py |
| 4692 | - vllm/v1/attention/backends/mla/rocm_aiter_mla.py |
| 4693 | - vllm/v1/attention/selector.py |
| 4694 | - vllm/platforms/rocm.py |
| 4695 | - vllm/_aiter_ops.py |
| 4696 | - examples/basic/offline_inference/chat.py |
| 4697 | - tests/kernels/attention/test_attention_selector.py |
| 4698 | optional: true |
| 4699 | commands: |
| 4700 | - amd-smi |
| 4701 | - python3 examples/basic/offline_inference/chat.py --attention-backend TRITON_ATTN |
| 4702 | - pytest -v -s tests/kernels/attention/test_attention_selector.py |
| 4703 | |
| 4704 | - label: ":amd: (MI355 DPX) MLA Kernels" |
| 4705 | timeout_in_minutes: 40 |
| 4706 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4707 | dind: false |
| 4708 | agent_pool: mi355_dpx |
| 4709 | num_gpus: 1 |
| 4710 | optional: true |
| 4711 | working_dir: "/vllm-workspace/" |
| 4712 | source_file_dependencies: |
| 4713 | - vllm/v1/attention/backends/mla/rocm_aiter_mla.py |
| 4714 | - vllm/v1/attention/selector.py |
| 4715 | - vllm/platforms/rocm.py |
| 4716 | - vllm/_aiter_ops.py |
| 4717 | - tests/kernels/attention/test_rocm_aiter_mla_decode.py |
| 4718 | - tests/kernels/attention/test_rocm_aiter_mla_decode_metadata.py |
| 4719 | - tests/kernels/attention/test_rocm_aiter_mla_fp8_support.py |
| 4720 | - tests/kernels/attention/test_rocm_aiter_mla_op_registration.py |
| 4721 | commands: |
| 4722 | - amd-smi |
| 4723 | - pytest -v -s tests/kernels/attention/test_rocm_aiter_mla_decode_metadata.py |
| 4724 | - pytest -v -s tests/kernels/attention/test_rocm_aiter_mla_decode.py |
| 4725 | - pytest -v -s tests/kernels/attention/test_rocm_aiter_mla_fp8_support.py |
| 4726 | - pytest -v -s tests/kernels/attention/test_rocm_aiter_mla_op_registration.py |
| 4727 | |
| 4728 | - label: ":amd: (MI355 DPX) DeepSeek V4 Fused Kernels" |
| 4729 | timeout_in_minutes: 60 |
| 4730 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4731 | dind: false |
| 4732 | agent_pool: mi355_dpx |
| 4733 | num_gpus: 1 |
| 4734 | optional: true |
| 4735 | working_dir: "/vllm-workspace/tests" |
| 4736 | source_file_dependencies: |
| 4737 | - csrc/libtorch_stable/fused_deepseek_v4_qnorm_rope_kv_insert_kernel.cu |
| 4738 | - vllm/models/deepseek_v4/common/ops/ |
| 4739 | - tests/kernels/test_fused_deepseek_v4_qnorm_rope_kv_insert.py |
| 4740 | - vllm/platforms/rocm.py |
| 4741 | commands: |
| 4742 | - pytest -v -s kernels/test_fused_deepseek_v4_qnorm_rope_kv_insert.py |
| 4743 | |
| 4744 | - label: ":amd: (MI355 DPX) Attention Kernels Shard %N" # TBD |
| 4745 | timeout_in_minutes: 90 |
| 4746 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4747 | dind: false |
| 4748 | agent_pool: mi355_dpx |
| 4749 | parallelism: 2 |
| 4750 | working_dir: "/vllm-workspace/tests" |
| 4751 | source_file_dependencies: |
| 4752 | - csrc/attention/ |
| 4753 | - vllm/v1/attention |
| 4754 | - vllm/model_executor/layers/attention |
| 4755 | - tests/kernels/attention |
| 4756 | - vllm/_aiter_ops.py |
| 4757 | - vllm/envs.py |
| 4758 | - vllm/platforms/rocm.py |
| 4759 | - vllm/utils/flashinfer.py |
| 4760 | optional: true |
| 4761 | commands: |
| 4762 | - pytest -v -s kernels/attention --ignore=kernels/attention/test_triton_unified_attention_diffkv.py --ignore=kernels/attention/test_rocm_aiter_mla_decode.py --ignore=kernels/attention/test_rocm_aiter_mla_decode_metadata.py --ignore=kernels/attention/test_rocm_aiter_mla_fp8_support.py --ignore=kernels/attention/test_rocm_aiter_mla_op_registration.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 4763 | |
| 4764 | - label: ":amd: (MI355 DPX) MoE Kernels Shard %N" # TBD |
| 4765 | timeout_in_minutes: 55 |
| 4766 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4767 | dind: false |
| 4768 | agent_pool: mi355_dpx |
| 4769 | parallelism: 5 |
| 4770 | working_dir: "/vllm-workspace/tests" |
| 4771 | source_file_dependencies: |
| 4772 | - csrc/quantization/cutlass_w8a8/moe/ |
| 4773 | - csrc/moe/ |
| 4774 | - csrc/rocm/ |
| 4775 | - tests/kernels/moe |
| 4776 | - tests/kernels/utils.py |
| 4777 | - vllm/model_executor/layers/fused_moe/ |
| 4778 | - vllm/model_executor/layers/quantization/mxfp4.py |
| 4779 | - vllm/model_executor/layers/quantization/utils/quant_utils.py |
| 4780 | - vllm/distributed/device_communicators/ |
| 4781 | - vllm/envs.py |
| 4782 | - vllm/config |
| 4783 | - vllm/_aiter_ops.py |
| 4784 | - vllm/platforms/rocm.py |
| 4785 | optional: true |
| 4786 | commands: |
| 4787 | - pytest -v -s kernels/moe |
| 4788 | --ignore=kernels/moe/test_modular_oai_triton_moe.py |
| 4789 | --ignore=kernels/moe/test_gpt_oss_triton_kernels.py |
| 4790 | --ignore=kernels/moe/test_moe.py |
| 4791 | --ignore=kernels/moe/test_block_int8.py |
| 4792 | --ignore=kernels/moe/test_triton_moe_no_act_mul.py |
| 4793 | --ignore=kernels/moe/test_triton_moe_ptpc_fp8.py |
| 4794 | --ignore=kernels/moe/test_deepep_moe.py |
| 4795 | --ignore=kernels/moe/test_moe_layer.py |
| 4796 | --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 4797 | - pytest -v -s kernels/moe/test_modular_oai_triton_moe.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 4798 | |
| 4799 | - label: ":amd: (MI355) FusedMoE Layer Kernels" # TBD |
| 4800 | timeout_in_minutes: 35 |
| 4801 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4802 | dind: false |
| 4803 | agent_pool: mi355_2 |
| 4804 | num_gpus: 2 |
| 4805 | optional: true |
| 4806 | working_dir: "/vllm-workspace/tests" |
| 4807 | source_file_dependencies: |
| 4808 | - csrc/moe/ |
| 4809 | - csrc/rocm/ |
| 4810 | - tests/kernels/moe |
| 4811 | - vllm/model_executor/layers/fused_moe/ |
| 4812 | - vllm/model_executor/layers/quantization/ |
| 4813 | - vllm/distributed/ |
| 4814 | - vllm/config/ |
| 4815 | - vllm/forward_context.py |
| 4816 | - vllm/v1/worker/workspace.py |
| 4817 | - vllm/utils/import_utils.py |
| 4818 | - vllm/utils/math_utils.py |
| 4819 | - vllm/utils/torch_utils.py |
| 4820 | - vllm/platforms/ |
| 4821 | - vllm/_aiter_ops.py |
| 4822 | - csrc/quantization/cutlass_w8a8/moe/ |
| 4823 | - vllm/distributed/device_communicators/ |
| 4824 | - vllm/config |
| 4825 | commands: |
| 4826 | # MoRI graph cases run in the Fault Tolerance group's healthy EP baseline. |
| 4827 | - pytest -v -s kernels/moe/test_moe_layer.py -k "not test_moe_layer_mori_graph" |
| 4828 | |
| 4829 | - label: ":amd: (MI355) MoRI EP Numerics" |
| 4830 | timeout_in_minutes: 20 |
| 4831 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4832 | dind: false |
| 4833 | agent_pool: mi355_2 |
| 4834 | num_gpus: 2 |
| 4835 | working_dir: "/vllm-workspace/tests" |
| 4836 | source_file_dependencies: |
| 4837 | - tests/kernels/moe/test_mori_moe.py |
| 4838 | - tests/kernels/moe/utils.py |
| 4839 | - tests/kernels/moe/modular_kernel_tools/parallel_utils.py |
| 4840 | - tests/kernels/utils.py |
| 4841 | - tests/utils.py |
| 4842 | - tests/conftest.py |
| 4843 | - vllm/utils/torch_utils.py |
| 4844 | - vllm/model_executor/layers/fused_moe/ |
| 4845 | - vllm/model_executor/layers/quantization/ |
| 4846 | - vllm/distributed/ |
| 4847 | - vllm/config/ |
| 4848 | - vllm/forward_context.py |
| 4849 | - vllm/v1/worker/workspace.py |
| 4850 | - vllm/platforms/ |
| 4851 | - vllm/_aiter_ops.py |
| 4852 | - requirements/rocm-test.txt |
| 4853 | optional: true |
| 4854 | commands: |
| 4855 | - amd-smi |
| 4856 | - VLLM_ROCM_USE_AITER=1 VLLM_ROCM_USE_AITER_MOE=1 VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS=0 pytest -v -s kernels/moe/test_mori_moe.py |
| 4857 | |
| 4858 | - label: ":amd: (MI355) Quantization Kernels Shard %N" |
| 4859 | timeout_in_minutes: 120 |
| 4860 | parallelism: 6 |
| 4861 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4862 | dind: false |
| 4863 | agent_pool: mi355_1 |
| 4864 | working_dir: "/vllm-workspace/tests" |
| 4865 | source_file_dependencies: |
| 4866 | - csrc/quantization/ |
| 4867 | - csrc/rocm/ |
| 4868 | - vllm/model_executor/layers/quantization |
| 4869 | - tests/kernels/quantization |
| 4870 | - tests/kernels/quant_utils.py |
| 4871 | - tests/kernels/utils.py |
| 4872 | - vllm/_aiter_ops.py |
| 4873 | - vllm/kernels/aiter_ops.py |
| 4874 | - vllm/_custom_ops.py |
| 4875 | - vllm/envs.py |
| 4876 | - vllm/platforms/rocm.py |
| 4877 | - vllm/model_executor/kernels/ |
| 4878 | - vllm/v1/attention/backends/rocm_aiter_fa.py |
| 4879 | - vllm/config/ |
| 4880 | - tests/kernels/quantization/test_rocm_skinny_gemms.py |
| 4881 | optional: true |
| 4882 | commands: |
| 4883 | - pytest -v -s kernels/quantization -k "not test_nvfp4_dense_emulation" --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 4884 | |
| 4885 | - label: ":amd: (MI355) NVFP4 Dense Emulation Shard %N" |
| 4886 | timeout_in_minutes: 10 |
| 4887 | parallelism: 3 |
| 4888 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4889 | dind: false |
| 4890 | agent_pool: mi355_1 |
| 4891 | working_dir: "/vllm-workspace/tests" |
| 4892 | source_file_dependencies: |
| 4893 | - tests/kernels/quantization/test_nvfp4_emulation.py |
| 4894 | - tests/kernels/quantization/nvfp4_utils.py |
| 4895 | - tests/conftest.py |
| 4896 | - vllm/model_executor/kernels/linear/ |
| 4897 | - vllm/model_executor/layers/quantization/utils/nvfp4_emulation_utils.py |
| 4898 | - vllm/config/kernel.py |
| 4899 | - vllm/platforms/ |
| 4900 | - vllm/envs.py |
| 4901 | - vllm/utils/torch_utils.py |
| 4902 | - requirements/rocm-test.txt |
| 4903 | optional: true |
| 4904 | commands: |
| 4905 | - amd-smi |
| 4906 | - pytest -v -s kernels/quantization/test_nvfp4_emulation.py -k test_nvfp4_dense_emulation --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT |
| 4907 | |
| 4908 | - label: ":amd: (MI355 DPX) FP8 MoE Kernels" # TBD |
| 4909 | timeout_in_minutes: 40 |
| 4910 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4911 | dind: false |
| 4912 | agent_pool: mi355_dpx |
| 4913 | working_dir: "/vllm-workspace/tests" |
| 4914 | source_file_dependencies: |
| 4915 | - csrc/moe/ |
| 4916 | - vllm/model_executor/layers/fused_moe/ |
| 4917 | - tests/kernels/moe/test_deepep_moe.py |
| 4918 | - vllm/_aiter_ops.py |
| 4919 | - vllm/platforms/rocm.py |
| 4920 | - vllm/envs.py |
| 4921 | - tests/kernels/moe/test_gpt_oss_triton_kernels.py |
| 4922 | - tests/kernels/moe/test_moe.py |
| 4923 | - tests/kernels/moe/test_block_int8.py |
| 4924 | - tests/kernels/moe/test_triton_moe_no_act_mul.py |
| 4925 | - tests/kernels/moe/test_triton_moe_ptpc_fp8.py |
| 4926 | optional: true |
| 4927 | commands: |
| 4928 | - pytest -v -s kernels/moe/test_gpt_oss_triton_kernels.py |
| 4929 | - pytest -v -s kernels/moe/test_moe.py |
| 4930 | - pytest -v -s kernels/moe/test_block_int8.py |
| 4931 | - pytest -v -s kernels/moe/test_triton_moe_no_act_mul.py |
| 4932 | - pytest -v -s kernels/moe/test_triton_moe_ptpc_fp8.py |
| 4933 | |
| 4934 | - label: ":amd: (MI355) DeepEP FP8 MoE Kernels" # TBD |
| 4935 | timeout_in_minutes: 55 |
| 4936 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4937 | dind: false |
| 4938 | agent_pool: mi355_2 |
| 4939 | num_gpus: 2 |
| 4940 | optional: true |
| 4941 | working_dir: "/vllm-workspace/tests" |
| 4942 | source_file_dependencies: |
| 4943 | - csrc/moe/ |
| 4944 | - csrc/quantization/w8a8/cutlass/moe/ |
| 4945 | - vllm/model_executor/layers/fused_moe/ |
| 4946 | - tests/kernels/moe/test_deepep_moe.py |
| 4947 | - vllm/_aiter_ops.py |
| 4948 | - vllm/platforms/rocm.py |
| 4949 | - vllm/envs.py |
| 4950 | commands: |
| 4951 | - pytest -v -s kernels/moe/test_deepep_moe.py |
| 4952 | |
| 4953 | #----------------------------------------------------------- mi355 · lora ------------------------------------------------------------# |
| 4954 | |
| 4955 | - label: ":amd: (MI355 DPX) LoRA Shard %N" # TBD |
| 4956 | timeout_in_minutes: 85 |
| 4957 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4958 | dind: false |
| 4959 | agent_pool: mi355_dpx |
| 4960 | parallelism: 4 |
| 4961 | optional: true |
| 4962 | working_dir: "/vllm-workspace/tests" |
| 4963 | source_file_dependencies: |
| 4964 | - vllm/lora |
| 4965 | - tests/lora |
| 4966 | - vllm/platforms/rocm.py |
| 4967 | commands: |
| 4968 | - pytest -v -s lora --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --ignore=lora/test_chatglm3_tp.py --ignore=lora/test_llama_tp.py --ignore=lora/test_qwen3_with_multi_loras.py --ignore=lora/test_olmoe_tp.py --ignore=lora/test_deepseekv2_tp.py --ignore=lora/test_gptoss_tp.py --ignore=lora/test_qwen3moe_tp.py --ignore=lora/test_qwen35_densemodel_lora.py |
| 4969 | |
| 4970 | #------------------------------------------------------ mi355 · models / basic -------------------------------------------------------# |
| 4971 | |
| 4972 | - label: ":amd: (MI355) Basic Models (Other)" # TBD |
| 4973 | timeout_in_minutes: 65 |
| 4974 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4975 | dind: false |
| 4976 | agent_pool: mi355_1 |
| 4977 | optional: true |
| 4978 | working_dir: "/vllm-workspace/tests" |
| 4979 | source_file_dependencies: |
| 4980 | - vllm/ |
| 4981 | - tests/models/test_terratorch.py |
| 4982 | - tests/models/transformers/test_backend.py |
| 4983 | - tests/models/test_registry.py |
| 4984 | - tests/models/test_hyv4_rocm.py |
| 4985 | - tests/models/test_vision.py |
| 4986 | - tests/models/test_deepseek_v4_vl_rocm.py |
| 4987 | commands: |
| 4988 | - pytest -v -s models/test_terratorch.py models/transformers/test_backend.py models/test_registry.py models/test_deepseek_v4_vl_rocm.py |
| 4989 | - VLLM_ROCM_USE_AITER=1 pytest -v -s models/test_hyv4_rocm.py -m 'not distributed' |
| 4990 | - pytest -v -s models/test_vision.py::test_simple_mrope_vision_model_spatial_merge |
| 4991 | |
| 4992 | - label: ":amd: (MI355 DPX) GLM5Next" |
| 4993 | timeout_in_minutes: 35 |
| 4994 | mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] |
| 4995 | dind: false |
| 4996 | agent_pool: mi355_dpx |
| 4997 | num_gpus: 1 |
| 4998 | optional: true |
| 4999 | working_dir: "/vllm-workspace/tests" |
| 5000 | source_file_dependencies: |
Showing first 5,000 of 5,997 lines.
View raw