Implementation:Vllm project Vllm Release Pipeline Config
| Knowledge Sources | |
|---|---|
| Domains | CI/CD, Release Engineering |
| Last Updated | 2026-02-08 00:00 GMT |
Overview
Defines the Buildkite pipeline for building and releasing vLLM Python wheels and Docker images across multiple platforms and hardware configurations.
Description
This YAML configuration file orchestrates the full vLLM release process through Buildkite CI. It defines pipeline steps for building Python wheels on x86_64 and aarch64 architectures for CUDA 12.9, CUDA 13.0, and CPU-only targets, then builds release Docker images, creates multi-arch manifests, and publishes versioned images to AWS ECR. The pipeline begins with a manual input step to collect the release version number.
Usage
This file is triggered manually through the Buildkite UI when preparing a new vLLM release. A release engineer initiates the pipeline, provides the target release version, and the pipeline automates wheel building, Docker image construction, multi-architecture manifest creation, and artifact publishing to S3 and ECR.
Code Reference
Source Location
- Repository: vllm
- File: .buildkite/release-pipeline.yaml
- Lines: 1-713
Signature
steps:
- input: "Provide Release version here"
id: input-release-version
fields:
- text: "What is the release version?"
key: release-version
- group: "Build Python wheels"
key: "build-wheels"
steps: [...]
- group: "Build release Docker images"
key: "build-release-images"
steps: [...]
- group: "Publish release images"
key: "publish-release-images"
steps: [...]
Import
# Triggered via Buildkite pipeline configuration.
# This pipeline is uploaded to Buildkite and triggered manually for releases.
# Reference: .buildkite/release-pipeline.yaml
I/O Contract
Inputs
| Name | Type | Required | Description |
|---|---|---|---|
| release-version | string | Yes | The version string for the release, collected via manual input step |
| BUILDKITE_COMMIT | string | Yes | The git commit SHA being released, provided by Buildkite environment |
| DOCKER_BUILDKIT | string | No | Enables Docker BuildKit (set to "1" in env blocks) |
Outputs
| Name | Type | Description |
|---|---|---|
| Python wheels | .whl files | Built wheel artifacts for x86_64 and aarch64 (CUDA 12.9, CUDA 13.0, CPU) |
| Docker images | Docker images | Release images pushed to public.ecr.aws/q9t5s3a7/vllm-release-repo |
| Multi-arch manifests | Docker manifest | Combined x86_64/aarch64 manifests for CUDA 12.9 and CUDA 13.0 |
Usage Examples
# The pipeline is triggered manually in Buildkite.
# Build targets include:
# - build-wheel-arm64-cuda-12-9
# - build-wheel-arm64-cuda-13-0
# - build-wheel-arm64-cpu
# - build-wheel-x86-cuda-12-9
# - build-wheel-x86-cuda-13-0
# - build-wheel-x86-cpu
#
# Docker images are built and pushed to ECR:
# public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT
# public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-cu130