minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[AMD][CI] Retire the ROCm 7.0 kernel wheel (#38767)

Browse Source 
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: quitenode <quitenode@users.noreply.github.com>
This commit is contained in:
Michael
2026-09-11 08:25:55 +08:00
committed by GitHub
co-authored by Cursor Agent quitenode
parent 00143e9c23
commit d006f40e24
4 changed files with 9 additions and 90 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/release-whl-kernel.yml
+2 -56
View File
@@ -16,7 +16,6 @@ on:
options:
- 'all'
- 'cu130'
- 'rocm700'
- 'rocm720'
- 'rocm1000'
- 'musa43'
@@ -169,7 +168,7 @@ jobs:
build-rocm-matrix:
if: |
github.repository == 'sgl-project/sglang' &&
(github.event_name == 'push' || github.event.inputs.target == 'all' || github.event.inputs.target == 'rocm700' || github.event.inputs.target == 'rocm720' || github.event.inputs.target == 'rocm1000')
(github.event_name == 'push' || github.event.inputs.target == 'all' || github.event.inputs.target == 'rocm720' || github.event.inputs.target == 'rocm1000')
runs-on: amd-docker-scale
strategy:
matrix:
@@ -182,7 +181,7 @@ jobs:
# leg is another image build on the shared AMD node, and its release job
# appends the same wheel to the index a second time. `inputs` is empty
# on push, which falls through to the full set.
rocm-version: ${{ fromJson(inputs.target == 'rocm700' && '["700"]' || inputs.target == 'rocm720' && '["720"]' || inputs.target == 'rocm1000' && '["1000"]' || '["700", "720", "1000"]') }}
rocm-version: ${{ fromJson(inputs.target == 'rocm720' && '["720"]' || inputs.target == 'rocm1000' && '["1000"]' || '["720", "1000"]') }}
steps:
# Self-hosted build nodes retain the workspace across jobs. Prior builds
# leave root-owned artifacts under python/sglang/kernels/aot/build/ that actions/checkout
@@ -216,59 +215,6 @@ jobs:
name: wheel-python${{ matrix.python-version }}-rocm${{ matrix.rocm-version }}
path: python/sglang/kernels/aot/dist/*
release-rocm700:
needs: build-rocm-matrix
# Only publish the flavors this run built; see build-rocm-matrix.
if: ${{ github.event_name == 'push' || inputs.target == 'all' || inputs.target == 'rocm700' }}
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
with:
ref: ${{ inputs.pr_number && format('refs/pull/{0}/head', inputs.pr_number) || '' }}
- name: Download artifacts
uses: actions/download-artifact@v4
with:
path: python/sglang/kernels/aot/dist/
merge-multiple: true
pattern: wheel-*-rocm700
- name: Set tag name
id: set_tag_name
run: |
if [ -z "${{ inputs.tag_name }}" ]; then
TAG_NAME="v$(cat python/sglang/kernels/aot/python/sgl_kernel/version.py | cut -d'"' -f2)"
echo "tag_name=$TAG_NAME" >> $GITHUB_OUTPUT
else
echo "tag_name=${{ inputs.tag_name }}" >> $GITHUB_OUTPUT
fi
- name: Release
uses: softprops/action-gh-release@v2
with:
tag_name: ${{ steps.set_tag_name.outputs.tag_name }}
repository: sgl-project/whl
token: ${{ secrets.GH_PAT_FOR_WHL_RELEASE }}
files: |
python/sglang/kernels/aot/dist/*
- name: Clone wheel index
run: git clone https://oauth2:${WHL_TOKEN}@github.com/sgl-project/whl.git sgl-whl
env:
WHL_TOKEN: ${{ secrets.GH_PAT_FOR_WHL_RELEASE }}
- name: Update wheel index
run: python3 scripts/update_kernel_whl_index.py --rocm 700
- name: Push wheel index
run: |
cd sgl-whl
git config --local user.name "sglang-bot"
git config --local user.email "sglangbot@gmail.com"
git add -A
git commit -m "update whl index"
git push
release-rocm720:
needs: build-rocm-matrix
if: ${{ github.event_name == 'push' || inputs.target == 'all' || inputs.target == 'rocm720' }}
3rdparty/amd/wheel/README.md
Vendored
+3 -13
View File
@@ -4,7 +4,7 @@ Building and releasing `sglang-kernel` as a wheel is a part of the release workf
# sglang
`3rdparty/amd/wheel/sglang/pyproject.toml` is the AMD-specific pyproject for building the `amd-sglang` wheel. It extends `python/pyproject_other.toml` with two ROCm-version extras (`rocm700`, `rocm720`) that pin the matching torch/triton/torchaudio/torchvision/`sglang-kernel` wheels, and renames the package to `amd-sglang`.
`3rdparty/amd/wheel/sglang/pyproject.toml` is the AMD-specific pyproject for building the `amd-sglang` wheel. It extends `python/pyproject_other.toml` with a `rocm720` extra that pins the matching torch/triton/torchaudio/torchvision/`sglang-kernel` wheels, and renames the package to `amd-sglang`.
## Operation to build sglang wheel
@@ -18,12 +18,6 @@ $ cd python && python -m build
### v0.5.9
ROCm 7.0.0:
```
pip uninstall sglang-kernel sglang amd-sglang
pip install "amd-sglang[all-hip,rocm700]" -i https://pypi.amd.com/rocm-7.0.0/simple --extra-index-url https://pypi.org/simple
```
ROCm 7.2.0:
```
pip uninstall sglang-kernel sglang amd-sglang
@@ -39,13 +33,9 @@ Note: You must resolve the two dependencies, AITER and triton, below. Others ar
[AITER](https://github.com/ROCm/aiter) is a fundamental dependency. Wheel-izing it is ongoing.
Until we can pin it reliably, install it manually (typically following the [ROCm docker recipe](https://github.com/sgl-project/sglang/blob/main/docker/rocm.Dockerfile#L106).
### Revolving triton
### Resolving triton
To avoid known issues in triton 3.5.1 installed by default, we recommend upgrading triton after installation. In ROCm 7.0.0 environment,
```
pip install triton==3.6.0
```
or ROCm 7.2.0,
To avoid known issues in triton 3.5.1 installed by default, we recommend upgrading triton after installation in a ROCm 7.2.0 environment:
```
pip install https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/triton-3.6.0%2Brocm7.2.0.gitba5c1517-cp310-cp310-linux_x86_64.whl
```
3rdparty/amd/wheel/sgl-kernel/build_rocm.sh
Vendored
+3 -7
View File
@@ -3,9 +3,7 @@ set -euo pipefail
ROCM_VERSION=${1:-}
if [[ "${ROCM_VERSION}" == "700" ]]; then
IMAGE="lmsysorg/sglang:v0.5.8.post1-rocm700-mi35x"
elif [[ "${ROCM_VERSION}" == "720" ]]; then
if [[ "${ROCM_VERSION}" == "720" ]]; then
IMAGE="rocm/pytorch:rocm7.2_ubuntu22.04_py3.10_pytorch_release_2.9.1"
elif [[ "${ROCM_VERSION}" == "1000" ]]; then
# Ubuntu 24.04 / Python 3.12 / torch 2.11 matches the stack the released
@@ -13,7 +11,7 @@ elif [[ "${ROCM_VERSION}" == "1000" ]]; then
# wheel can still cover both gfx942 and gfx950.
IMAGE="rocm/pytorch:rocm10.0_ubuntu24.04_py3.12_pytorch_release_2.11.0"
else
echo "ERROR: Unsupported ROCM_VERSION='${ROCM_VERSION}'. Only '700', '720' and '1000' are supported." >&2
echo "ERROR: Unsupported ROCM_VERSION='${ROCM_VERSION}'. Only '720' and '1000' are supported." >&2
exit 1
fi
@@ -69,9 +67,7 @@ docker run --rm \
${IMAGE} \
bash -c "
# Install torch, triton, and friends, depending on the ROCm version
if [[ "${ROCM_VERSION}" == "700" ]]; then
${PYTHON_ROOT_PATH}/pip install https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/torch-2.9.1.dev20251204%2Brocm7.0.2.lw.git351ff442-cp310-cp310-linux_x86_64.whl https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/triton-3.5.1%2Brocm7.0.2.gita272dfa8-cp310-cp310-linux_x86_64.whl https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/torchaudio-2.9.0%2Brocm7.0.2.gite3c6ee2b-cp310-cp310-linux_x86_64.whl https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/torchvision-0.24.0%2Brocm7.0.2.gitb919bd0c-cp310-cp310-linux_x86_64.whl
elif [[ "${ROCM_VERSION}" == "720" ]]; then
if [[ "${ROCM_VERSION}" == "720" ]]; then
${PYTHON_ROOT_PATH}/pip install https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torch-2.9.1%2Brocm7.2.0.lw.git7e1940d4-cp310-cp310-linux_x86_64.whl https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/triton-3.5.1%2Brocm7.2.0.gita272dfa8-cp310-cp310-linux_x86_64.whl https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torchaudio-2.9.0%2Brocm7.2.0.gite3c6ee2b-cp310-cp310-linux_x86_64.whl https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torchvision-0.24.0%2Brocm7.2.0.gitb919bd0c-cp310-cp310-linux_x86_64.whl
fi
${ROCM_SETUP}
3rdparty/amd/wheel/sglang/pyproject.toml
Vendored
+1 -14
View File
@@ -67,18 +67,6 @@ runtime_common = [
]
# ROCm specific packages (https://repo.radeon.com/rocm/manylinux/)
# Existing practice for daily rocm700 docker images relies on 700-rc
# versions of software that are not public available. Here we pin some
# from rocm702 as the closest set as daily rocm700 images.
rocm700 = [
"torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/torch-2.9.1.dev20251204%2Brocm7.0.2.lw.git351ff442-cp310-cp310-linux_x86_64.whl",
"triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/triton-3.5.1%2Brocm7.0.2.gita272dfa8-cp310-cp310-linux_x86_64.whl",
"torchaudio @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/torchaudio-2.9.0%2Brocm7.0.2.gite3c6ee2b-cp310-cp310-linux_x86_64.whl",
"torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/torchvision-0.24.0%2Brocm7.0.2.gitb919bd0c-cp310-cp310-linux_x86_64.whl",
"mooncake-transfer-engine-non-cuda==0.3.8.post1",
"sglang-kernel @ https://github.com/sgl-project/whl/releases/download/v0.4.0/sglang_kernel-0.4.0+rocm700-cp310-abi3-manylinux2014_x86_64.whl",
]
rocm720 = [
"torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torch-2.9.1%2Brocm7.2.0.lw.git7e1940d4-cp310-cp310-linux_x86_64.whl",
"triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/triton-3.5.1%2Brocm7.2.0.gita272dfa8-cp310-cp310-linux_x86_64.whl",
@@ -89,8 +77,7 @@ rocm720 = [
]
# HIP (Heterogeneous-computing Interface for Portability) for AMD
# Install with one of:
# pip install "amd-sglang[srt_hip,rocm700]"
# Install with:
# pip install "amd-sglang[srt_hip,rocm720]"
srt_hip = [
"amd-sglang[runtime_common]",