Skip to content

Limit to 1 stage at a time #3

Limit to 1 stage at a time

Limit to 1 stage at a time #3

Workflow file for this run

name: Build vLLM Docker Image Matrix
env:
PARALLELISM: 1
TRITON_REF: release/3.3.x
TRITON_BUILD_VERSION: 3.3.0
XFORMERS_REF: v0.0.29.post3
XFORMERS_BUILD_VERSION: 0.0.29.post3
FLASHINFER_REF: v0.2.2.post1
FLASHINFER_BUILD_VERSION: 0.2.2.post1
VLLM_REF: d47807ba
VLLM_BUILD_VERSION: 0.7.4
on:
push:
branches: [main]
pull_request:
jobs:
build:
strategy:
matrix:
arch: [amd64, arm64]
cuda_version: [12.8.0]
image_distro: [ubuntu24.04]
runs-on: [self-hosted, "${{ matrix.arch }}"]
steps:
- name: Generate image name
run: |
echo "GHCR_IMAGE=ghcr.io/${GITHUB_REPOSITORY@L}" >> ${GITHUB_ENV}
- name: Login to GHCR
uses: docker/login-action@v3
with:
registry: ghcr.io
username: ${{ github.actor }}
password: ${{ secrets.GITHUB_TOKEN }}
- name: Setup docker buildx
uses: docker/setup-buildx-action@v3
with:
buildkitd-config-inline: |
[worker.oci]
max-parallelism = ${{ env.PARALLELISM }}
- name: Build and push builder image
uses: docker/build-push-action@v6
with:
build-args: |
CUDA_VERSION=${{ matrix.cuda_version }}
IMAGE_DISTRO=${{ matrix.image_distro }}
TRITON_REF=${{ env.TRITON_REF }}
TRITON_BUILD_VERSION=${{ env.TRITON_BUILD_VERSION }}
XFORMERS_REF=${{ env.XFORMERS_REF }}
XFORMERS_BUILD_VERSION=${{ env.XFORMERS_BUILD_VERSION }}
FLASHINFER_REF=${{ env.FLASHINFER_REF }}
FLASHINFER_BUILD_VERSION=${{ env.FLASHINFER_BUILD_VERSION }}
VLLM_REF=${{ env.VLLM_REF }}
VLLM_BUILD_VERSION=${{ env.VLLM_BUILD_VERSION }}
cache-from: type=registry,ref=${{ env.GHCR_IMAGE }}:cache-${{ matrix.arch }}
cache-to: type=registry,ref=${{ env.GHCR_IMAGE }}:cache-${{ matrix.arch }},mode=max
context: .
file: Dockerfile
platforms: linux/${{ matrix.arch }}
push: true
tags: ${{ env.GHCR_IMAGE }}:${{ matrix.arch }}
ghcr:
needs: build
runs-on: self-hosted
steps:
- name: Login to GHCR
uses: docker/login-action@v3
with:
registry: ghcr.io
username: ${{ github.actor }}
password: ${{ secrets.GITHUB_TOKEN }}
- name: Append images
run: |
ARCHS=(amd64 arm64)
docker buildx imagetools create -t ${GHCR_IMAGE}:latest ${ARCHS[@]/#/${GHCR_IMAGE}:}