| 1 | group: Basic Correctness |
| 2 | depends_on: |
| 3 | - image-build |
| 4 | steps: |
| 5 | - label: ":nvidia: (H200 MIG 18GB) Basic Correctness Models %N" |
| 6 | key: basic-correctness |
| 7 | timeout_in_minutes: 68 |
| 8 | device: h200_18gb |
| 9 | parallelism: 2 |
| 10 | source_file_dependencies: &basic-correctness-deps |
| 11 | - vllm/ |
| 12 | - "!vllm/distributed/kv_transfer/" |
| 13 | - tests/basic_correctness/ |
| 14 | commands: |
| 15 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 16 | - pytest -v -s basic_correctness/models --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB |
| 17 | mirror: |
| 18 | amd: |
| 19 | label: ":amd: (MI355 DPX) Basic Correctness Models %N" |
| 20 | dind: false |
| 21 | device: mi355_dpx |
| 22 | timeout_in_minutes: 60 |
| 23 | depends_on: |
| 24 | - image-build-amd |
| 25 | |
| 26 | - label: ":nvidia: (H200 MIG 18GB) Basic Correctness CuMem" |
| 27 | key: basic-correctness-cumem |
| 28 | timeout_in_minutes: 68 |
| 29 | device: h200_18gb |
| 30 | source_file_dependencies: *basic-correctness-deps |
| 31 | commands: |
| 32 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 33 | - pytest -v -s basic_correctness/memory/cumem |
| 34 | mirror: |
| 35 | amd: |
| 36 | label: ":amd: (MI355 DPX) Basic Correctness CuMem" |
| 37 | dind: false |
| 38 | device: mi355_dpx |
| 39 | timeout_in_minutes: 60 |
| 40 | depends_on: |
| 41 | - image-build-amd |
| 42 | |
| 43 | - label: ":nvidia: (H200 MIG 18GB) Basic Correctness Sleep Mode" |
| 44 | key: basic-correctness-sleep-mode |
| 45 | timeout_in_minutes: 68 |
| 46 | device: h200_18gb |
| 47 | source_file_dependencies: *basic-correctness-deps |
| 48 | commands: |
| 49 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 50 | - pytest -v -s basic_correctness/memory/sleep_mode |
| 51 | mirror: |
| 52 | amd: |
| 53 | label: ":amd: (MI300) Basic Correctness Sleep Mode" |
| 54 | dind: false |
| 55 | device: mi300_1 |
| 56 | timeout_in_minutes: 60 |
| 57 | depends_on: |
| 58 | - image-build-amd |
| 59 | |
| 60 | - label: ":nvidia: (H200 MIG 18GB) Basic Correctness CPU Offload" |
| 61 | key: basic-correctness-cpu-offload |
| 62 | timeout_in_minutes: 68 |
| 63 | device: h200_18gb |
| 64 | source_file_dependencies: *basic-correctness-deps |
| 65 | commands: |
| 66 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 67 | - pytest -v -s basic_correctness/cpu_offload |
| 68 | mirror: |
| 69 | amd: |
| 70 | label: ":amd: (MI355 DPX) Basic Correctness CPU Offload" |
| 71 | dind: false |
| 72 | device: mi355_dpx |
| 73 | timeout_in_minutes: 60 |
| 74 | depends_on: |
| 75 | - image-build-amd |
| 76 | |
| 77 | - label: ":nvidia: (H200 MIG 18GB) Basic Correctness Prefetch Offload" |
| 78 | key: basic-correctness-prefetch-offload |
| 79 | timeout_in_minutes: 68 |
| 80 | device: h200_18gb |
| 81 | source_file_dependencies: *basic-correctness-deps |
| 82 | commands: |
| 83 | - export VLLM_WORKER_MULTIPROC_METHOD=spawn |
| 84 | - pytest -v -s basic_correctness/prefetch_offload |
| 85 | mirror: |
| 86 | amd: |
| 87 | label: ":amd: (MI355 DPX) Basic Correctness Prefetch Offload" |
| 88 | dind: false |
| 89 | device: mi355_dpx |
| 90 | timeout_in_minutes: 60 |
| 91 | depends_on: |
| 92 | - image-build-amd |