[AMD] build gfx1250 release image from main (#37225)

This commit is contained in:
YC Yen-Ching Tseng
2026-08-31 17:04:04 +08:00
committed by GitHub
parent 6580d5cd9a
commit 0674be736c
+18 -19
View File
@@ -118,21 +118,17 @@ jobs:
docker tag "rocm/sgl-dev:${IMAGE_TAG}" "lmsysorg/sglang-rocm:${IMAGE_TAG}"
docker push "lmsysorg/sglang-rocm:${IMAGE_TAG}"
# gfx1250 (MI45x) is still being brought up on the amd_helios branch: main
# has no gfx1250 stage in docker/rocm.Dockerfile, so this job builds from
# that branch rather than from the ref the workflow was triggered on. It runs
# on the same schedule as the pair above but stays a separate job, because
# there is no MI450 runner pool consuming the image and a bring-up build
# gfx1250 (MI45x) runs on the same schedule as the pair above but stays a
# separate job: no MI450 runner pool consumes the image, so a bring-up build
# failure must not gate the released gfx942/gfx950 images.
publish-gfx1250:
if: github.repository == 'sgl-project/sglang' && (github.event_name != 'workflow_dispatch' || inputs.job_select == 'all' || inputs.job_select == 'publish-gfx1250')
runs-on: amd-docker-scale
environment: 'prod'
steps:
- name: Checkout amd_helios
- name: Checkout repository
uses: actions/checkout@v4
with:
ref: amd_helios
fetch-depth: 0
- name: Set up Python
@@ -143,7 +139,7 @@ jobs:
- name: Set date
run: echo "DATE=$(date +%Y%m%d)" >> "$GITHUB_ENV"
- name: Get version and branch head
- name: Get version from latest tag
id: version
run: |
VERSION=$(python3 scripts/release/get_version_tag.py --tag-only | sed 's/^v//')
@@ -153,16 +149,11 @@ jobs:
exit 1
fi
# The checkout above is amd_helios, so this is the branch head that
# both the build context and the cloned sglang source come from.
BRANCH_SHA=$(git rev-parse HEAD)
COMMIT_HASH=$(git rev-parse --short HEAD)
PRETEND_VERSION="${VERSION}.dev${DATE}+g${COMMIT_HASH}"
echo "version=${VERSION}" >> "$GITHUB_OUTPUT"
echo "branch_sha=${BRANCH_SHA}" >> "$GITHUB_OUTPUT"
echo "pretend_version=${PRETEND_VERSION}" >> "$GITHUB_OUTPUT"
echo "Building gfx1250 from amd_helios at ${BRANCH_SHA}"
- name: Login to Docker Hub (AMD)
uses: docker/login-action@v3
@@ -172,18 +163,15 @@ jobs:
- name: Build and push to rocm/sgl-dev
env:
BRANCH_SHA: ${{ steps.version.outputs.branch_sha }}
PRETEND_VERSION: ${{ steps.version.outputs.pretend_version }}
VERSION: ${{ steps.version.outputs.version }}
run: |
# `dev` marks this as built from a bring-up branch rather than from
# the ref the release ran on, so it never reads as a released tag.
image_tag="v${VERSION}-rocm10-mi45x-dev-${DATE}"
echo "Publishing rocm/sgl-dev:${image_tag}"
image_tag="v${VERSION}-rocm10-mi45x-${DATE}"
echo "IMAGE_TAG=${image_tag}" >> "$GITHUB_ENV"
docker build . \
-f docker/rocm.Dockerfile \
--build-arg "SGL_BRANCH=${BRANCH_SHA}" \
--build-arg "SGL_BRANCH=${GITHUB_SHA}" \
--build-arg BUILD_TYPE=all \
--build-arg GPU_ARCH=gfx1250-rocm1000 \
--build-arg ENABLE_MORI=1 \
@@ -193,3 +181,14 @@ jobs:
-t "rocm/sgl-dev:${image_tag}" \
--no-cache
docker push "rocm/sgl-dev:${image_tag}"
- name: Login to Docker Hub (lmsys)
uses: docker/login-action@v3
with:
username: ${{ secrets.DOCKERHUB_USERNAME }}
password: ${{ secrets.DOCKERHUB_TOKEN }}
- name: Push to lmsysorg/sglang-rocm
run: |
docker tag "rocm/sgl-dev:${IMAGE_TAG}" "lmsysorg/sglang-rocm:${IMAGE_TAG}"
docker push "lmsysorg/sglang-rocm:${IMAGE_TAG}"