Files
flash-attention-prebuild-wh…/.github/workflows/test-arm-self-hosted.yml
T
Junya Morioka 7d12c6d7e4 feat: Add ARM64 support with QEMU in self-hosted runner
- Add container-image input parameter to _build_linux_self_host workflow
- Create test-arm-self-hosted workflow for ARM64 Linux builds
- Add qemu-user-static package to self-hosted runner Dockerfile
- Enable ARM64 container builds using arm64v8/ubuntu:22.04
2025-12-10 23:54:47 +09:00

31 lines
1017 B
YAML

name: Test ARM Linux build (self-hosted)
on:
workflow_dispatch:
jobs:
# #########################################################
# Build wheels with self-hosted runner
# #########################################################
build_wheels_self_hosted:
name: Build wheels and Upload (Linux ARM64, self-hosted runner)
uses: ./.github/workflows/_build_linux_self_host.yml
strategy:
fail-fast: false
matrix:
flash-attn-version: ["2.8.3"]
python-version: ["3.13"]
torch-version: ["2.9.1"]
# https://developer.nvidia.com/cuda-toolkit-archive
cuda-version: ["13.0.1"]
runner: ["self-hosted"]
with:
flash-attn-version: ${{ matrix.flash-attn-version }}
python-version: ${{ matrix.python-version }}
torch-version: ${{ matrix.torch-version }}
cuda-version: ${{ matrix.cuda-version }}
is-upload: false
runner: ${{ matrix.runner}}
use-container: true
container-image: "arm64v8/ubuntu:22.04"