build-infra

This repo, currently is experiment for

base image

pipeline as code, and code as document.

xPU	driver version	python	ubuntu	torch	file	image name
nvidia	cuda-drivers=590.48.01-0ubuntu1 cuda-toolkit=13.1.1-1 libnccl2=2.29.2-1+cuda13.1 libnccl-dev=2.29.2-1+cuda13.1	latest	24.04	latest	container/containerfile.nvidia	harbor.baai.ac.cn/flagbase/flagbase-nvidia:latest
nvidia	cuda-drivers=590.48.01-0ubuntu1 cuda-toolkit=13.1.1-1 libnccl2=2.29.2-1+cuda13.1 libnccl-dev=2.29.2-1+cuda13.1	3.12	24.04	2.8	container/containerfile.nvidia	harbor.baai.ac.cn/flagbase/flagbase-nvidia:py312torch2.8
metax	maca_sdk=3.3.0.15	latest	22.04	latest	container/containerfile.metax	harbor.baai.ac.cn/flagbase/flagbase-metax:latest
amd	rocm-hip-libraries=7.1.1.70101-3822.04 rocm-hip-runtime=7.1.1.70101-3822.04 rocm-language-runtime=7.1.1.70101-3822.04 rocm-hip-sdk=7.1.1.70101-3822.04 rocm-opencl-runtime=7.1.1.70101-3822.04 rocm-developer-tools=7.1.1.70101-3822.04 rocm-opencl-sdk=7.1.1.70101-38~22.04	latest	24.04	latest	container/containerfile.amd	harbor.baai.ac.cn/flagbase/flagbase-amd:latest
ascend	cann:8.2.rc2-910(from base image)	3.11	22.04	2.6	container/containerfile.ascend	harbor.baai.ac.cn/flagbase/flagbase-ascend:py311torch2.6

pipeline as code interface

Image and code build, basic CI to ensure the code works.

ref https://docs.github.com/zh/actions/how-tos/reuse-automations/reuse-workflows

jobs:
  call-workflow-passing-data:
    uses: flagos-ai/build-infra/.github/workflows/imagebuild.yml@main
    with:
      push: ${{ matrix.push }}
      containerfile: ${{ matrix.containerfile }}
      image_name: ${{ matrix.image_name }}
      tag: ${{ matrix.tag }}
      runson: ${{ matrix.runson }}
      build-args: ${{ matrix.build-args }}
      image_prefix: 'flagbase'
      no-cache: ${{ matrix.no-cache }}
    secrets:
      REGISTRY_USERNAME: ${{ secrets.REGISTRY_USERNAME }}
      CONTAINER_REGISTRY: ${{ secrets.CONTAINER_REGISTRY }}