265 lines
9.6 KiB
YAML
265 lines
9.6 KiB
YAML
name: Release sgl-deep-gemm
|
|
|
|
on:
|
|
workflow_dispatch:
|
|
inputs:
|
|
version:
|
|
description: "Wheel version (e.g. 0.1.0, 0.1.1rc0)"
|
|
type: string
|
|
required: true
|
|
target:
|
|
type: choice
|
|
description: "Build target (default: all)"
|
|
required: false
|
|
default: 'all'
|
|
options:
|
|
- 'all'
|
|
- 'cu130'
|
|
branch:
|
|
description: "DeepGEMM branch to build from (default: dev)"
|
|
type: string
|
|
required: false
|
|
default: 'dev'
|
|
validation_only:
|
|
description: "Build and test without publishing wheels"
|
|
type: boolean
|
|
required: false
|
|
default: false
|
|
|
|
concurrency:
|
|
group: release-sgl-deepgemm-${{ github.ref }}
|
|
cancel-in-progress: true
|
|
|
|
env:
|
|
NCCL_NVLS_ENABLE: "0"
|
|
# Must match TORCH_VER in docker/sgl-deep-gemm.Dockerfile — the wheel's
|
|
# pre-compiled _C.so links against this torch ABI.
|
|
TORCH_VER: "2.13.0"
|
|
|
|
jobs:
|
|
build-cu130-matrix:
|
|
if: |
|
|
github.repository == 'sgl-project/sglang' &&
|
|
(github.event.inputs.target == 'all' || github.event.inputs.target == 'cu130')
|
|
strategy:
|
|
matrix:
|
|
python-version: ["3.12"]
|
|
cuda-version: ["13.0"]
|
|
arch: [x86_64, aarch64]
|
|
include:
|
|
- arch: x86_64
|
|
runner: x64-kernel-build-node
|
|
- arch: aarch64
|
|
runner: arm-kernel-build-node
|
|
runs-on: ${{ matrix.runner }}
|
|
steps:
|
|
- name: Clean workspace (remove root-owned files from prior runs)
|
|
run: |
|
|
docker run --rm -v "${{ github.workspace }}:/workspace" alpine:3 \
|
|
sh -c 'rm -rf /workspace/..?* /workspace/.[!.]* /workspace/*' || true
|
|
|
|
- uses: actions/checkout@v4
|
|
|
|
- name: Checkout DeepGEMM
|
|
uses: actions/checkout@v4
|
|
with:
|
|
repository: sgl-project/DeepGEMM
|
|
ref: ${{ inputs.branch || 'dev' }}
|
|
path: DeepGEMM
|
|
submodules: recursive
|
|
|
|
- name: Set up Python ${{ matrix.python-version }}
|
|
uses: actions/setup-python@v5
|
|
with:
|
|
python-version: ${{ matrix.python-version }}
|
|
|
|
- name: Set wheel version
|
|
run: |
|
|
echo -n "${{ inputs.version }}" > DeepGEMM/sgl_deep_gemm/VERSION
|
|
cat DeepGEMM/sgl_deep_gemm/VERSION
|
|
|
|
- name: Build wheel
|
|
run: |
|
|
chmod +x ./scripts/build_sgl_deep_gemm.sh ./scripts/rename_sgl_deep_gemm_whl.sh
|
|
./scripts/build_sgl_deep_gemm.sh "${{ matrix.python-version }}" "${{ matrix.cuda-version }}" "${{ github.workspace }}/DeepGEMM" "${{ matrix.arch }}"
|
|
|
|
# PyPI upload moved to release-cu130 so it is gated on test-cu130.
|
|
- name: Upload artifacts
|
|
uses: actions/upload-artifact@v4
|
|
with:
|
|
name: deepgemm-wheel-cuda${{ matrix.cuda-version }}-${{ matrix.arch }}
|
|
path: DeepGEMM/dist/*.whl
|
|
|
|
- name: Upload PyPI artifacts
|
|
uses: actions/upload-artifact@v4
|
|
with:
|
|
name: deepgemm-pypi-cuda${{ matrix.cuda-version }}-${{ matrix.arch }}
|
|
path: DeepGEMM/dist-pypi/*.whl
|
|
|
|
test-cu130:
|
|
needs: build-cu130-matrix
|
|
strategy:
|
|
fail-fast: false
|
|
matrix:
|
|
include:
|
|
- arch_label: sm90
|
|
runner: 8-gpu-h200
|
|
wheel_arch: x86_64
|
|
timeout: 60
|
|
- arch_label: sm100
|
|
runner: 8-gpu-b200
|
|
wheel_arch: x86_64
|
|
timeout: 60
|
|
# The sm120 tests require a runner with RTX 6000. It causes OOM on 5090 runners.
|
|
# - arch_label: sm120
|
|
# runner: 1-gpu-5090
|
|
# wheel_arch: x86_64
|
|
- arch_label: sm100-aarch64
|
|
runner: 4-gpu-gb300
|
|
wheel_arch: aarch64
|
|
timeout: 60
|
|
runs-on: ${{ matrix.runner }}
|
|
timeout-minutes: ${{ matrix.timeout }}
|
|
steps:
|
|
- uses: actions/checkout@v4
|
|
|
|
- name: Checkout DeepGEMM
|
|
uses: actions/checkout@v4
|
|
with:
|
|
repository: sgl-project/DeepGEMM
|
|
ref: ${{ inputs.branch || 'dev' }}
|
|
path: DeepGEMM
|
|
submodules: recursive
|
|
|
|
- name: Download wheel
|
|
uses: actions/download-artifact@v4
|
|
with:
|
|
path: dist/
|
|
merge-multiple: true
|
|
pattern: deepgemm-wheel-cuda13.0-${{ matrix.wheel_arch }}
|
|
|
|
- name: Install wheel and deps
|
|
env:
|
|
CU_TAG: cu130
|
|
run: |
|
|
python3 -m pip install --upgrade pip
|
|
# deep_gemm imports torch before loading _C.so, so torch must preload the
|
|
# libcudart the cu130 wheel links. Pin the CUDA-matched torch (see sgl-deep-gemm.Dockerfile).
|
|
python3 -m pip install "torch==${TORCH_VER}" --index-url "https://download.pytorch.org/whl/${CU_TAG}" --force-reinstall
|
|
python3 -m pip install numpy "tilelang==0.1.9" "tile-kernels==1.0.0" "sgl-deep-ep==0.1.2"
|
|
python3 -m pip install dist/*.whl
|
|
# Torch pins NCCL 2.29.7, but this DeepEP wheel requires 2.30.7.
|
|
# Override it after every dependency-resolving pip command.
|
|
python3 -m pip install "nvidia-nccl-cu13==2.30.7" --force-reinstall --no-deps
|
|
nccl_lib=$(python3 -c "from importlib.metadata import distribution; print(distribution('nvidia-nccl-cu13').locate_file('nvidia/nccl/lib'))")
|
|
export LD_LIBRARY_PATH="${nccl_lib}${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}"
|
|
echo "LD_LIBRARY_PATH=${LD_LIBRARY_PATH}" >> "${GITHUB_ENV}"
|
|
python3 -c "import deep_gemm; print('deep_gemm:', deep_gemm.__file__)"
|
|
python3 - <<'PY'
|
|
import ctypes
|
|
import sys
|
|
import torch
|
|
import deep_ep
|
|
from tile_kernels.moe import top2_sum_gate
|
|
|
|
# torch.cuda.nccl.version() reports Torch's compiled version, not the loaded library.
|
|
nccl_version = ctypes.c_int()
|
|
assert ctypes.CDLL("libnccl.so.2").ncclGetVersion(ctypes.byref(nccl_version)) == 0
|
|
assert nccl_version.value == 23007, nccl_version.value
|
|
assert hasattr(deep_ep, "ElasticBuffer"), "DeepEP ElasticBuffer is required"
|
|
sys.path.insert(0, "DeepGEMM/third-party")
|
|
from tilelang_ops import ref_mhc
|
|
assert ref_mhc.has_baseline(), "Mega mHC reference operators are unavailable"
|
|
print("DeepGEMM test dependencies ready; NCCL:", nccl_version.value)
|
|
PY
|
|
|
|
- name: Install Compute Sanitizer 2025.4.1
|
|
env:
|
|
WHEEL_ARCH: ${{ matrix.wheel_arch }}
|
|
run: |
|
|
set -euo pipefail
|
|
# CUDA 13.1.1's standalone sanitizer fixes Python tensor retention in
|
|
# host backtraces. Keep the wheel and runtime on CUDA 13.0.
|
|
case "${WHEEL_ARCH}" in
|
|
x86_64)
|
|
sanitizer_arch=linux-x86_64
|
|
sanitizer_sha256=b9637777a31cd0f0ffe1e965523e8b0d382019b8496855fd688c886988583cf4
|
|
;;
|
|
aarch64)
|
|
sanitizer_arch=linux-sbsa
|
|
sanitizer_sha256=1dc68ccb146032bcdc9d932569c865d68e3966ecf136940ca38fd7d5abfda4ac
|
|
;;
|
|
*) echo "Unsupported sanitizer architecture: ${WHEEL_ARCH}" >&2; exit 1 ;;
|
|
esac
|
|
archive="cuda_sanitizer_api-${sanitizer_arch}-13.1.118-archive"
|
|
install_dir="${RUNNER_TEMP}/deepgemm-compute-sanitizer"
|
|
mkdir -p "${install_dir}"
|
|
curl -fsSL --retry 3 \
|
|
"https://developer.download.nvidia.com/compute/cuda/redist/cuda_sanitizer_api/${sanitizer_arch}/${archive}.tar.xz" \
|
|
-o "${install_dir}/${archive}.tar.xz"
|
|
echo "${sanitizer_sha256} ${install_dir}/${archive}.tar.xz" | sha256sum -c -
|
|
tar -xJf "${install_dir}/${archive}.tar.xz" -C "${install_dir}"
|
|
sanitizer="${install_dir}/${archive}/bin/compute-sanitizer"
|
|
"${sanitizer}" --version
|
|
echo "COMPUTE_SANITIZER=${sanitizer}" >> "${GITHUB_ENV}"
|
|
|
|
- name: Run DeepGEMM test suite
|
|
env:
|
|
# Each runner is a single NVLink domain; DeepEP references do not need RDMA/GIN.
|
|
EP_DISABLE_GIN: "1"
|
|
run: |
|
|
chmod +x "${{ github.workspace }}/DeepGEMM/sgl_deep_gemm/run_tests.sh"
|
|
"${{ github.workspace }}/DeepGEMM/sgl_deep_gemm/run_tests.sh" --release
|
|
|
|
release-cu130:
|
|
if: ${{ !inputs.validation_only }}
|
|
needs: [build-cu130-matrix, test-cu130]
|
|
runs-on: ubuntu-latest
|
|
steps:
|
|
- uses: actions/checkout@v4
|
|
|
|
- name: Download artifacts
|
|
uses: actions/download-artifact@v4
|
|
with:
|
|
path: dist/
|
|
merge-multiple: true
|
|
pattern: deepgemm-wheel-cuda13.0-*
|
|
|
|
- name: Download PyPI artifacts
|
|
uses: actions/download-artifact@v4
|
|
with:
|
|
path: dist-pypi/
|
|
merge-multiple: true
|
|
pattern: deepgemm-pypi-cuda13.0-*
|
|
|
|
- name: Upload to PyPI
|
|
run: |
|
|
pip install --upgrade twine "packaging>=24.2"
|
|
python3 -m twine upload --skip-existing dist-pypi/* -u __token__ -p ${{ secrets.SGL_DEEP_GEMM_PYPI_TOKEN }}
|
|
|
|
- name: Release
|
|
uses: softprops/action-gh-release@v2
|
|
with:
|
|
tag_name: v${{ inputs.version }}
|
|
repository: sgl-project/whl
|
|
token: ${{ secrets.GH_PAT_FOR_WHL_RELEASE }}
|
|
files: |
|
|
dist/*
|
|
|
|
- name: Clone wheel index
|
|
run: git clone https://oauth2:${WHL_TOKEN}@github.com/sgl-project/whl.git sgl-whl
|
|
env:
|
|
WHL_TOKEN: ${{ secrets.GH_PAT_FOR_WHL_RELEASE }}
|
|
|
|
- name: Update wheel index
|
|
run: python3 scripts/update_deepgemm_whl_index.py --cuda 130
|
|
|
|
- name: Push wheel index
|
|
run: |
|
|
cd sgl-whl
|
|
git config --local user.name "sglang-bot"
|
|
git config --local user.email "sglangbot@gmail.com"
|
|
git add -A
|
|
git commit -m "update sgl-deep-gemm whl index for v${{ inputs.version }}"
|
|
git push
|