| 1 | group: Spec Decode |
| 2 | depends_on: |
| 3 | - image-build |
| 4 | steps: |
| 5 | - label: ":nvidia: (H200 MIG 35GB) V1 Spec Decode" |
| 6 | device: h200_35gb |
| 7 | key: v1-spec-decode |
| 8 | timeout_in_minutes: 40 |
| 9 | source_file_dependencies: |
| 10 | - vllm/config/ |
| 11 | - vllm/distributed/ |
| 12 | - "!vllm/distributed/kv_transfer/" |
| 13 | - vllm/inputs/ |
| 14 | - vllm/model_executor/ |
| 15 | - vllm/models/qwen4_exp/ |
| 16 | - vllm/platforms/ |
| 17 | - vllm/sampling_params.py |
| 18 | - vllm/transformers_utils/ |
| 19 | - vllm/utils/ |
| 20 | - vllm/v1/ |
| 21 | - tests/v1/spec_decode |
| 22 | commands: |
| 23 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 24 | # TODO: create another `optional` test group for slow tests |
| 25 | - pytest -v -s -m 'not slow_test' v1/spec_decode |
| 26 | mirror: |
| 27 | amd: |
| 28 | label: ":amd: (MI300) V1 Spec Decode" |
| 29 | dind: false |
| 30 | device: mi300_1 |
| 31 | timeout_in_minutes: 50 |
| 32 | depends_on: |
| 33 | - image-build-amd |
| 34 | |
| 35 | - label: ":nvidia: (H200 MIG 35GB) Spec Decode Eagle 1: DeepSeek + Qwen" |
| 36 | key: spec-decode-eagle-1-deepseek-qwen |
| 37 | timeout_in_minutes: 25 |
| 38 | device: h200_35gb |
| 39 | source_file_dependencies: |
| 40 | - vllm/v1/spec_decode/ |
| 41 | - vllm/v1/worker/gpu/spec_decode/ |
| 42 | - tests/v1/e2e/spec_decode/ |
| 43 | commands: |
| 44 | - pytest -v -s v1/e2e/spec_decode/eagle/ -k "deepseek_eagle or qwen3_eagle3" |
| 45 | mirror: |
| 46 | amd: |
| 47 | label: ":amd: (MI355 DPX) Spec Decode Eagle 1: DeepSeek + Qwen" |
| 48 | dind: false |
| 49 | device: mi355_dpx |
| 50 | timeout_in_minutes: 60 |
| 51 | depends_on: |
| 52 | - image-build-amd |
| 53 | source_file_dependencies: |
| 54 | - vllm/v1/spec_decode/ |
| 55 | - vllm/v1/worker/gpu/spec_decode/ |
| 56 | - vllm/model_executor/model_loader/ |
| 57 | - vllm/v1/sample/ |
| 58 | - vllm/model_executor/layers/ |
| 59 | - tests/v1/e2e/spec_decode/ |
| 60 | - vllm/platforms/rocm.py |
| 61 | |
| 62 | - label: ":nvidia: (H200 MIG 35GB) Spec Decode Eagle 2: Llama 3 + Qwen VL + Other" |
| 63 | key: spec-decode-eagle-2-llama3-qwen-vl-other |
| 64 | timeout_in_minutes: 25 |
| 65 | device: h200_35gb |
| 66 | source_file_dependencies: |
| 67 | - vllm/v1/spec_decode/ |
| 68 | - vllm/v1/worker/gpu/spec_decode/ |
| 69 | - tests/v1/e2e/spec_decode/ |
| 70 | commands: |
| 71 | - pytest -v -s v1/e2e/spec_decode/eagle/ -k "not deepseek_eagle and not qwen3_eagle3" |
| 72 | mirror: |
| 73 | amd: |
| 74 | label: ":amd: (MI355 DPX) Spec Decode Eagle 2: Llama 3 + Qwen VL + Other" |
| 75 | dind: false |
| 76 | device: mi355_dpx |
| 77 | timeout_in_minutes: 60 |
| 78 | depends_on: |
| 79 | - image-build-amd |
| 80 | source_file_dependencies: |
| 81 | - vllm/v1/spec_decode/ |
| 82 | - vllm/v1/worker/gpu/spec_decode/ |
| 83 | - vllm/model_executor/model_loader/ |
| 84 | - vllm/v1/sample/ |
| 85 | - vllm/model_executor/layers/ |
| 86 | - tests/v1/e2e/spec_decode/ |
| 87 | - vllm/platforms/rocm.py |
| 88 | |
| 89 | - label: ":nvidia: (B200) Spec Decode Eagle Nightly" |
| 90 | key: spec-decode-eagle-nightly-b200 |
| 91 | timeout_in_minutes: 25 |
| 92 | device: b200-k8s |
| 93 | optional: true |
| 94 | source_file_dependencies: |
| 95 | - vllm/v1/spec_decode/ |
| 96 | - vllm/v1/worker/gpu/spec_decode/ |
| 97 | - tests/v1/e2e/spec_decode/ |
| 98 | commands: |
| 99 | - pytest -v -s v1/e2e/spec_decode/eagle/ |
| 100 | |
| 101 | # MTP correctness is one parametrized test split into per-model folders; |
| 102 | # each job below runs whole directories. |
| 103 | - label: ":nvidia: (H200 MIG 35GB) Spec Decode Speculators" |
| 104 | key: spec-decode-speculators |
| 105 | timeout_in_minutes: 30 |
| 106 | device: h200_35gb |
| 107 | source_file_dependencies: |
| 108 | - vllm/v1/spec_decode/ |
| 109 | - vllm/v1/worker/gpu/spec_decode/ |
| 110 | - vllm/v1/attention/backends/ |
| 111 | - vllm/transformers_utils/configs/speculators/ |
| 112 | - tests/v1/e2e/spec_decode/ |
| 113 | commands: |
| 114 | - pytest -v -s v1/e2e/spec_decode/speculators/ |
| 115 | mirror: |
| 116 | amd: |
| 117 | label: ":amd: (MI300) Spec Decode Speculators" |
| 118 | dind: false |
| 119 | device: mi300_1 |
| 120 | timeout_in_minutes: 40 |
| 121 | depends_on: |
| 122 | - image-build-amd |
| 123 | source_file_dependencies: |
| 124 | - vllm/v1/spec_decode/ |
| 125 | - vllm/v1/worker/gpu/spec_decode/ |
| 126 | - vllm/model_executor/model_loader/ |
| 127 | - vllm/v1/sample/ |
| 128 | - vllm/model_executor/layers/ |
| 129 | - vllm/transformers_utils/configs/speculators/ |
| 130 | - tests/v1/e2e/spec_decode/ |
| 131 | - vllm/platforms/rocm.py |
| 132 | |
| 133 | - label: ":nvidia: (H200 MIG 35GB) Spec Decode MTP Gemma4" |
| 134 | key: spec-decode-mtp-gemma4 |
| 135 | timeout_in_minutes: 30 |
| 136 | device: h200_35gb |
| 137 | source_file_dependencies: |
| 138 | - vllm/v1/spec_decode/ |
| 139 | - vllm/v1/worker/gpu/spec_decode/ |
| 140 | - vllm/v1/attention/backends/ |
| 141 | - vllm/transformers_utils/configs/speculators/ |
| 142 | - tests/v1/e2e/spec_decode/ |
| 143 | commands: |
| 144 | - pytest -v -s v1/e2e/spec_decode/mtp/gemma4/ |
| 145 | mirror: |
| 146 | amd: |
| 147 | label: ":amd: (MI300) Spec Decode MTP Gemma4" |
| 148 | dind: false |
| 149 | device: mi300_1 |
| 150 | timeout_in_minutes: 40 |
| 151 | depends_on: |
| 152 | - image-build-amd |
| 153 | source_file_dependencies: |
| 154 | - vllm/v1/spec_decode/ |
| 155 | - vllm/v1/worker/gpu/spec_decode/ |
| 156 | - vllm/model_executor/model_loader/ |
| 157 | - vllm/v1/sample/ |
| 158 | - vllm/model_executor/layers/ |
| 159 | - vllm/transformers_utils/configs/speculators/ |
| 160 | - tests/v1/e2e/spec_decode/ |
| 161 | - vllm/platforms/rocm.py |
| 162 | |
| 163 | - label: ":nvidia: (H200 MIG 35GB) Spec Decode MTP DeepSeek + MiMo" |
| 164 | key: spec-decode-mtp-deepseek-mimo |
| 165 | timeout_in_minutes: 30 |
| 166 | device: h200_35gb |
| 167 | source_file_dependencies: |
| 168 | - vllm/v1/spec_decode/ |
| 169 | - vllm/v1/worker/gpu/spec_decode/ |
| 170 | - vllm/v1/attention/backends/ |
| 171 | - vllm/transformers_utils/configs/speculators/ |
| 172 | - tests/v1/e2e/spec_decode/ |
| 173 | commands: |
| 174 | - pytest -v -s v1/e2e/spec_decode/mtp/deepseek/ |
| 175 | - pytest -v -s v1/e2e/spec_decode/mtp/mimo/ |
| 176 | mirror: |
| 177 | amd: |
| 178 | label: ":amd: (MI300) Spec Decode MTP DeepSeek + MiMo" |
| 179 | dind: false |
| 180 | device: mi300_1 |
| 181 | timeout_in_minutes: 40 |
| 182 | depends_on: |
| 183 | - image-build-amd |
| 184 | source_file_dependencies: |
| 185 | - vllm/v1/spec_decode/ |
| 186 | - vllm/v1/worker/gpu/spec_decode/ |
| 187 | - vllm/model_executor/model_loader/ |
| 188 | - vllm/v1/sample/ |
| 189 | - vllm/model_executor/layers/ |
| 190 | - vllm/transformers_utils/configs/speculators/ |
| 191 | - tests/v1/e2e/spec_decode/ |
| 192 | - vllm/platforms/rocm.py |
| 193 | |
| 194 | - label: ":nvidia: (H200 MIG 35GB) Spec Decode MTP Qwen3.5" |
| 195 | key: spec-decode-mtp-qwen3-5 |
| 196 | timeout_in_minutes: 30 |
| 197 | device: h200_35gb |
| 198 | source_file_dependencies: |
| 199 | - vllm/v1/spec_decode/ |
| 200 | - vllm/v1/worker/gpu/spec_decode/ |
| 201 | - vllm/v1/attention/backends/ |
| 202 | - vllm/transformers_utils/configs/speculators/ |
| 203 | - tests/v1/e2e/spec_decode/ |
| 204 | commands: |
| 205 | - pytest -v -s v1/e2e/spec_decode/mtp/qwen3_5/ |
| 206 | mirror: |
| 207 | amd: |
| 208 | label: ":amd: (MI300) Spec Decode MTP Qwen3.5" |
| 209 | dind: false |
| 210 | device: mi300_1 |
| 211 | timeout_in_minutes: 40 |
| 212 | depends_on: |
| 213 | - image-build-amd |
| 214 | source_file_dependencies: |
| 215 | - vllm/v1/spec_decode/ |
| 216 | - vllm/v1/worker/gpu/spec_decode/ |
| 217 | - vllm/model_executor/model_loader/ |
| 218 | - vllm/v1/sample/ |
| 219 | - vllm/model_executor/layers/ |
| 220 | - vllm/transformers_utils/configs/speculators/ |
| 221 | - tests/v1/e2e/spec_decode/ |
| 222 | - vllm/platforms/rocm.py |
| 223 | |
| 224 | - label: ":nvidia: (B200) Spec Decode Speculators + MTP Nightly" |
| 225 | key: spec-decode-speculators-mtp-nightly-b200 |
| 226 | timeout_in_minutes: 30 |
| 227 | device: b200-k8s |
| 228 | optional: true |
| 229 | source_file_dependencies: |
| 230 | - vllm/v1/spec_decode/ |
| 231 | - vllm/v1/worker/gpu/spec_decode/ |
| 232 | - vllm/transformers_utils/configs/speculators/ |
| 233 | - tests/v1/e2e/spec_decode/ |
| 234 | commands: |
| 235 | - pytest -v -s v1/e2e/spec_decode/speculators/ |
| 236 | - pytest -v -s v1/e2e/spec_decode/mtp/ |
| 237 | |
| 238 | - label: ":nvidia: (H200 MIG 35GB) Spec Decode N-Gram + Suffix" |
| 239 | key: spec-decode-ngram-suffix |
| 240 | timeout_in_minutes: 20 |
| 241 | device: h200_35gb |
| 242 | source_file_dependencies: |
| 243 | - vllm/v1/spec_decode/ |
| 244 | - vllm/v1/worker/gpu/spec_decode/ |
| 245 | - tests/v1/e2e/spec_decode/ |
| 246 | - tests/spec_decode/ |
| 247 | commands: |
| 248 | - pytest -v -s v1/e2e/spec_decode/ngram_suffix/ |
| 249 | - python3 spec_decode/test_custom_proposer.py |
| 250 | mirror: |
| 251 | amd: |
| 252 | label: ":amd: (MI355 DPX) Spec Decode N-Gram + Suffix" |
| 253 | dind: false |
| 254 | device: mi355_dpx |
| 255 | timeout_in_minutes: 35 |
| 256 | depends_on: |
| 257 | - image-build-amd |
| 258 | source_file_dependencies: |
| 259 | - vllm/v1/spec_decode/ |
| 260 | - vllm/v1/worker/gpu/spec_decode/ |
| 261 | - vllm/model_executor/model_loader/ |
| 262 | - vllm/v1/sample/ |
| 263 | - vllm/model_executor/layers/ |
| 264 | - tests/v1/e2e/spec_decode/ |
| 265 | - vllm/platforms/rocm.py |
| 266 | |
| 267 | - label: ":nvidia: (H200 MIG 18GB) Spec Decode Draft Model Shard %N" |
| 268 | key: spec-decode-draft-model |
| 269 | timeout_in_minutes: 60 |
| 270 | device: h200_18gb |
| 271 | parallelism: 4 |
| 272 | source_file_dependencies: |
| 273 | - vllm/v1/spec_decode/ |
| 274 | - vllm/v1/worker/gpu/spec_decode/ |
| 275 | - tests/v1/e2e/spec_decode/ |
| 276 | commands: |
| 277 | - pytest -v -s v1/e2e/spec_decode/draft_model/ --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB |
| 278 | mirror: |
| 279 | amd: |
| 280 | label: ":amd: (MI355 DPX) Spec Decode Draft Model Shard %N" |
| 281 | dind: false |
| 282 | device: mi355_dpx |
| 283 | timeout_in_minutes: 60 |
| 284 | parallelism: 2 |
| 285 | depends_on: |
| 286 | - image-build-amd |
| 287 | source_file_dependencies: |
| 288 | - vllm/v1/spec_decode/ |
| 289 | - vllm/v1/worker/gpu/spec_decode/ |
| 290 | - vllm/model_executor/model_loader/ |
| 291 | - vllm/v1/sample/ |
| 292 | - vllm/model_executor/layers/ |
| 293 | - tests/v1/e2e/spec_decode/ |
| 294 | - vllm/platforms/rocm.py |
| 295 | |
| 296 | - label: ":nvidia: (B200) Spec Decode Draft Model Nightly" |
| 297 | key: spec-decode-draft-model-nightly-b200 |
| 298 | timeout_in_minutes: 40 |
| 299 | device: b200-k8s |
| 300 | optional: true |
| 301 | source_file_dependencies: |
| 302 | - vllm/v1/spec_decode/ |
| 303 | - vllm/v1/worker/gpu/spec_decode/ |
| 304 | - tests/v1/e2e/spec_decode/ |
| 305 | commands: |
| 306 | - pytest -v -s v1/e2e/spec_decode/draft_model/ |
| 307 | |
| 308 | - label: ":nvidia: (H100) Speculators Correctness Nightly" |
| 309 | key: speculators-correctness |
| 310 | timeout_in_minutes: 30 |
| 311 | device: h100 |
| 312 | optional: true |
| 313 | num_devices: 1 |
| 314 | source_file_dependencies: |
| 315 | - vllm/v1/spec_decode/ |
| 316 | - vllm/v1/worker/gpu/spec_decode/dflash/ |
| 317 | - vllm/model_executor/models/qwen3_dflash.py |
| 318 | - tests/v1/spec_decode/test_speculators_correctness.py |
| 319 | commands: |
| 320 | - export VLLM_ALLOW_INSECURE_SERIALIZATION=1 |
| 321 | - pytest -v -s v1/spec_decode/test_speculators_correctness.py -m slow_test |
| 322 | mirror: |
| 323 | amd: |
| 324 | label: ":amd: (MI300) Speculators Correctness Nightly" |
| 325 | dind: false |
| 326 | device: mi300_1 |
| 327 | timeout_in_minutes: 40 |
| 328 | depends_on: |
| 329 | - image-build-amd |
| 330 | source_file_dependencies: |
| 331 | - vllm/_aiter_ops.py |
| 332 | - vllm/platforms/rocm.py |
| 333 | |
| 334 | - label: ":nvidia: (B200) Spec Decode DeepSeek MTP Parallel Load" |
| 335 | key: spec-decode-deepseek-mtp-parallel-load-2xb200-2xmi300 |
| 336 | timeout_in_minutes: 30 |
| 337 | device: b200-k8s |
| 338 | optional: true |
| 339 | num_devices: 2 |
| 340 | source_file_dependencies: |
| 341 | - vllm/v1/spec_decode/llm_base_proposer.py |
| 342 | - vllm/v1/spec_decode/eagle.py |
| 343 | - vllm/v1/worker/gpu/spec_decode/eagle/ |
| 344 | - vllm/model_executor/models/deepseek_mtp.py |
| 345 | - vllm/model_executor/models/deepseek_v2.py |
| 346 | - tests/v1/e2e/spec_decode/test_mtp_parallel_load.py |
| 347 | commands: |
| 348 | - pytest -v -s v1/e2e/spec_decode/test_mtp_parallel_load.py |
| 349 | mirror: |
| 350 | amd: |
| 351 | label: ":amd: (MI300) Spec Decode DeepSeek MTP Parallel Load" |
| 352 | dind: false |
| 353 | device: mi300_2 |
| 354 | timeout_in_minutes: 45 |
| 355 | depends_on: |
| 356 | - image-build-amd |
| 357 | source_file_dependencies: |
| 358 | - vllm/distributed/ |
| 359 | - vllm/model_executor/model_loader/ |
| 360 | - vllm/model_executor/layers/ |
| 361 | - vllm/model_executor/models/deepseek_mtp.py |
| 362 | - vllm/model_executor/models/deepseek_v2.py |
| 363 | - vllm/v1/attention/ |
| 364 | - vllm/v1/engine/ |
| 365 | - vllm/v1/spec_decode/ |
| 366 | - vllm/v1/worker/gpu/spec_decode/ |
| 367 | - vllm/v1/worker/gpu_model_runner.py |
| 368 | - tests/v1/e2e/spec_decode/test_mtp_parallel_load.py |
| 369 | - vllm/platforms/rocm.py |
| 370 | |
| 371 | - label: ":nvidia: (H200 MIG 35GB) Spec Decode AL DFlash Nightly" |
| 372 | key: spec-decode-dflash-nightly |
| 373 | timeout_in_minutes: 90 |
| 374 | device: h200_35gb |
| 375 | optional: true |
| 376 | source_file_dependencies: |
| 377 | - vllm/v1/spec_decode/ |
| 378 | - vllm/v1/worker/gpu/spec_decode/ |
| 379 | - vllm/model_executor/models/qwen3_dflash.py |
| 380 | - vllm/model_executor/models/qwen3_dflash2.py |
| 381 | - vllm/model_executor/models/qwen3_5.py |
| 382 | - vllm/model_executor/models/laguna_dflash.py |
| 383 | - tests/v1/e2e/spec_decode/ |
| 384 | commands: |
| 385 | # DFlash2 NVFP4 (27B) does not fit a 35GB MIG slice; runs on B200 below. |
| 386 | - pytest -v -s v1/e2e/spec_decode/acceptance_rates/dflash/ -k "not dflash2" |
| 387 | mirror: |
| 388 | amd: |
| 389 | label: ":amd: (MI300) Spec Decode AL DFlash Nightly" |
| 390 | dind: false |
| 391 | device: mi300_1 |
| 392 | timeout_in_minutes: 120 |
| 393 | commands: |
| 394 | - pytest -v -s v1/e2e/spec_decode/acceptance_rates/dflash/ -k "not dflash2" |
| 395 | depends_on: |
| 396 | - image-build-amd |
| 397 | source_file_dependencies: |
| 398 | - vllm/v1/spec_decode/ |
| 399 | - vllm/v1/worker/gpu/spec_decode/ |
| 400 | - vllm/v1/attention/ |
| 401 | - vllm/v1/sample/ |
| 402 | - vllm/model_executor/model_loader/ |
| 403 | - vllm/model_executor/layers/ |
| 404 | - vllm/model_executor/models/qwen2.py |
| 405 | - vllm/model_executor/models/qwen3.py |
| 406 | - vllm/model_executor/models/qwen3_dflash.py |
| 407 | - vllm/model_executor/models/laguna.py |
| 408 | - vllm/model_executor/models/laguna_dflash.py |
| 409 | - tests/v1/e2e/spec_decode/ |
| 410 | - vllm/platforms/rocm.py |
| 411 | |
| 412 | - label: ":nvidia: (B200) Spec Decode AL DFlash2 Nightly" |
| 413 | key: spec-decode-dflash2-nightly-b200 |
| 414 | timeout_in_minutes: 60 |
| 415 | device: b200-k8s |
| 416 | optional: true |
| 417 | source_file_dependencies: |
| 418 | - vllm/v1/spec_decode/ |
| 419 | - vllm/v1/worker/gpu/spec_decode/ |
| 420 | - vllm/model_executor/models/qwen3_dflash.py |
| 421 | - vllm/model_executor/models/qwen3_dflash2.py |
| 422 | - vllm/model_executor/models/qwen3_5.py |
| 423 | - tests/v1/e2e/spec_decode/ |
| 424 | commands: |
| 425 | - pytest -v -s v1/e2e/spec_decode/acceptance_rates/dflash/ -k "dflash2" |
| 426 | mirror: |
| 427 | amd: |
| 428 | label: ":amd: (MI355) Spec Decode AL DFlash2 Nightly" |
| 429 | dind: false |
| 430 | device: mi355_1 |
| 431 | num_devices: 1 |
| 432 | timeout_in_minutes: 60 |
| 433 | depends_on: |
| 434 | - image-build-amd |
| 435 | source_file_dependencies: |
| 436 | - .buildkite/test_areas/spec_decode.yaml |
| 437 | - tests/v1/e2e/spec_decode/ |
| 438 | - tests/evals/gsm8k/ |
| 439 | - tests/utils.py |
| 440 | - vllm/v1/spec_decode/ |
| 441 | - vllm/v1/worker/gpu/spec_decode/ |
| 442 | - vllm/v1/worker/gpu/sample/ |
| 443 | - vllm/v1/worker/gpu/model_runner.py |
| 444 | - vllm/v1/attention/ |
| 445 | - vllm/model_executor/models/qwen3_dflash.py |
| 446 | - vllm/model_executor/models/qwen3_dflash2.py |
| 447 | - vllm/model_executor/models/qwen3_5.py |
| 448 | - vllm/model_executor/model_loader/ |
| 449 | - vllm/model_executor/layers/quantization/ |
| 450 | - vllm/model_executor/kernels/linear/ |
| 451 | - vllm/platforms/rocm.py |
| 452 | |
| 453 | - label: ":nvidia: (H200 MIG 35GB) Spec Decode AL DSpark Nightly" |
| 454 | key: spec-decode-dspark-nightly |
| 455 | timeout_in_minutes: 60 |
| 456 | device: h200_35gb |
| 457 | optional: true |
| 458 | source_file_dependencies: |
| 459 | - vllm/v1/spec_decode/ |
| 460 | - vllm/v1/worker/gpu/spec_decode/ |
| 461 | - vllm/model_executor/models/qwen3_dspark.py |
| 462 | - vllm/model_executor/models/gemma4_dspark.py |
| 463 | - tests/v1/e2e/spec_decode/ |
| 464 | commands: |
| 465 | - pytest -v -s v1/e2e/spec_decode/acceptance_rates/dspark/ |
| 466 | mirror: |
| 467 | amd: |
| 468 | label: ":amd: (MI300) Spec Decode AL DSpark Nightly" |
| 469 | dind: false |
| 470 | device: mi300_1 |
| 471 | timeout_in_minutes: 90 |
| 472 | depends_on: |
| 473 | - image-build-amd |
| 474 | source_file_dependencies: |
| 475 | - vllm/v1/spec_decode/ |
| 476 | - vllm/v1/worker/gpu/spec_decode/ |
| 477 | - vllm/v1/attention/ |
| 478 | - vllm/v1/sample/ |
| 479 | - vllm/model_executor/model_loader/ |
| 480 | - vllm/model_executor/layers/ |
| 481 | - vllm/model_executor/models/qwen2.py |
| 482 | - vllm/model_executor/models/qwen3.py |
| 483 | - vllm/model_executor/models/qwen3_dflash.py |
| 484 | - vllm/model_executor/models/qwen3_dspark.py |
| 485 | - vllm/model_executor/models/gemma4.py |
| 486 | - vllm/model_executor/models/gemma4_mm.py |
| 487 | - vllm/model_executor/models/gemma4_unified.py |
| 488 | - vllm/model_executor/models/gemma4_mtp.py |
| 489 | - vllm/model_executor/models/gemma4_dspark.py |
| 490 | - tests/v1/e2e/spec_decode/ |
| 491 | - vllm/platforms/rocm.py |
| 492 | |
| 493 | - label: ":nvidia: (H200 MIG 35GB) Spec Decode AL MTP + Other Acceptance Nightly" |
| 494 | key: spec-decode-mtp-other-acceptance-nightly |
| 495 | timeout_in_minutes: 35 |
| 496 | device: h200_35gb |
| 497 | optional: true |
| 498 | parallelism: 3 |
| 499 | source_file_dependencies: |
| 500 | - vllm/v1/spec_decode/ |
| 501 | - vllm/v1/worker/gpu/spec_decode/ |
| 502 | - tests/v1/e2e/spec_decode/conftest.py |
| 503 | - tests/v1/e2e/spec_decode/ |
| 504 | commands: |
| 505 | # Each Gemma4 parametrization is an indivisible ~23m floor. Keep them in |
| 506 | # separate shards and put the remaining tests in shard 0, where new tests |
| 507 | # land by default. |
| 508 | - if [ "$$BUILDKITE_PARALLEL_JOB" = "0" ]; then pytest -v -s v1/e2e/spec_decode/acceptance_rates/mtp_other/ --deselect "tests/v1/e2e/spec_decode/acceptance_rates/mtp_other/test_mtp.py::test_gemma4_mtp_acceptance_lengths[False]" --deselect "tests/v1/e2e/spec_decode/acceptance_rates/mtp_other/test_mtp.py::test_gemma4_mtp_acceptance_lengths[True]"; fi |
| 509 | - if [ "$$BUILDKITE_PARALLEL_JOB" = "1" ]; then pytest -v -s "v1/e2e/spec_decode/acceptance_rates/mtp_other/test_mtp.py::test_gemma4_mtp_acceptance_lengths[False]"; fi |
| 510 | - if [ "$$BUILDKITE_PARALLEL_JOB" = "2" ]; then pytest -v -s "v1/e2e/spec_decode/acceptance_rates/mtp_other/test_mtp.py::test_gemma4_mtp_acceptance_lengths[True]"; fi |
| 511 | mirror: |
| 512 | amd: |
| 513 | label: ":amd: (MI355 DPX) Spec Decode AL MTP + Other Acceptance Nightly" |
| 514 | dind: false |
| 515 | device: mi355_dpx |
| 516 | timeout_in_minutes: 55 |
| 517 | depends_on: |
| 518 | - image-build-amd |
| 519 | working_dir: /vllm-workspace/tests |
| 520 | source_file_dependencies: |
| 521 | - vllm/v1/spec_decode/ |
| 522 | - vllm/v1/sample/ |
| 523 | - vllm/v1/worker/gpu/ |
| 524 | - vllm/v1/worker/gpu_model_runner.py |
| 525 | - vllm/v1/worker/gpu_worker.py |
| 526 | - vllm/model_executor/model_loader/ |
| 527 | - vllm/model_executor/models/gemma4.py |
| 528 | - vllm/model_executor/models/gemma4_mm.py |
| 529 | - vllm/model_executor/models/gemma4_mtp.py |
| 530 | - vllm/model_executor/models/llama.py |
| 531 | - vllm/model_executor/models/llama_eagle3.py |
| 532 | - vllm/model_executor/models/medusa.py |
| 533 | - vllm/model_executor/models/registry.py |
| 534 | - tests/evals/gsm8k/ |
| 535 | - tests/utils.py |
| 536 | - tests/v1/e2e/spec_decode/ |
| 537 | - vllm/platforms/rocm.py |