| 1 | group: Hardware - AMD Build |
| 2 | |
| 3 | # ROCm image flow: |
| 4 | # 1. Select the content-matching ROCm base, building only on a cache miss. |
| 5 | # 2. Build ci_base from either the stable base or the newly selected base. |
| 6 | # 3. Build the build-scoped ROCm CI image and smoke-test it before GPU jobs run. |
| 7 | steps: |
| 8 | - label: "AMD: :docker: refresh ROCm base" |
| 9 | key: refresh-rocm-base-amd |
| 10 | depends_on: [] |
| 11 | device: amd_cpu |
| 12 | no_plugin: true |
| 13 | commands: |
| 14 | - bash .buildkite/scripts/rocm/refresh-base-image.sh |
| 15 | concurrency: 1 |
| 16 | concurrency_group: "vllm/rocm/base-build/$BUILDKITE_COMMIT" |
| 17 | env: |
| 18 | DOCKER_BUILDKIT: "1" |
| 19 | BUILDKIT_PROGRESS: "tty" |
| 20 | TERM: "xterm-256color" |
| 21 | retry: |
| 22 | automatic: |
| 23 | - exit_status: 1 # Transient Docker/BuildKit failure |
| 24 | limit: 1 |
| 25 | - exit_status: -1 # Agent was lost |
| 26 | limit: 1 |
| 27 | - exit_status: -10 # Agent was lost |
| 28 | limit: 1 |
| 29 | |
| 30 | # Ensure ci_base is up-to-date before building the test image. |
| 31 | # Compares a content hash of ci_base-affecting files against the remote |
| 32 | # image label. If hashes match the build is skipped (< 30 s); if they |
| 33 | # differ ci_base is rebuilt and pushed automatically. |
| 34 | - label: "AMD: :docker: ensure ci_base" |
| 35 | key: ensure-ci-base-amd |
| 36 | soft_fail: false |
| 37 | depends_on: |
| 38 | - refresh-rocm-base-amd |
| 39 | device: amd_cpu |
| 40 | no_plugin: true |
| 41 | commands: |
| 42 | - bash .buildkite/scripts/rocm/build-ci-base.sh |
| 43 | concurrency: 1 |
| 44 | concurrency_group: "vllm/rocm/ci-base-build/$BUILDKITE_COMMIT" |
| 45 | env: |
| 46 | DOCKER_BUILDKIT: "1" |
| 47 | BUILDKIT_PROGRESS: "tty" |
| 48 | TERM: "xterm-256color" |
| 49 | VLLM_BAKE_FILE: "docker/docker-bake-rocm.hcl" |
| 50 | PYTORCH_ROCM_ARCH: "gfx90a;gfx942;gfx950" |
| 51 | REMOTE_VLLM: "0" |
| 52 | retry: |
| 53 | automatic: |
| 54 | - exit_status: 1 # Transient Docker/BuildKit failure |
| 55 | limit: 1 |
| 56 | - exit_status: 2 # Exact image found but registry handoff failed |
| 57 | limit: 1 |
| 58 | - exit_status: -1 # Agent was lost |
| 59 | limit: 1 |
| 60 | - exit_status: -10 # Agent was lost |
| 61 | limit: 1 |
| 62 | |
| 63 | - label: "AMD: :docker: build test image and artifacts" |
| 64 | key: image-build-amd |
| 65 | soft_fail: false |
| 66 | depends_on: |
| 67 | - ensure-ci-base-amd |
| 68 | device: amd_cpu |
| 69 | no_plugin: true |
| 70 | commands: |
| 71 | - bash .buildkite/scripts/rocm/build-test-image.sh |
| 72 | - bash .buildkite/scripts/rocm/smoke-test-image.sh |
| 73 | env: |
| 74 | DOCKER_BUILDKIT: "1" |
| 75 | BUILDKIT_PROGRESS: "tty" |
| 76 | TERM: "xterm-256color" |
| 77 | VLLM_BAKE_FILE: "docker/docker-bake-rocm.hcl" |
| 78 | PYTORCH_ROCM_ARCH: "gfx90a;gfx942;gfx950" |
| 79 | IMAGE_TAG: "rocm/vllm-ci:build-$BUILDKITE_BUILD_ID" |
| 80 | REMOTE_VLLM: "0" |
| 81 | retry: |
| 82 | automatic: |
| 83 | - exit_status: 1 # Transient Docker/BuildKit failure |
| 84 | limit: 1 |
| 85 | - exit_status: -1 # Agent was lost |
| 86 | limit: 1 |
| 87 | - exit_status: -10 # Agent was lost |
| 88 | limit: 1 |