Cenv

ghcr.io: ghcr.io/eoeair/cenv:TAG

Platform

• ARCH: x86_64
• OS:
  • debian testing
  • alpine edge
• ROCM: 7.1.0
• CUDA: >= 12.3

Container Usage

1. oci: podman run -it, See the runArgs section of the devcontainer config file for more options.
2. devcontainer(VScode): cp -r .devcontainer WORKDIR

Note

1. Denv: git may not be loaded on the first run. Please Reload Window.
   1. Press Ctrl + Shift + P (Windows/Linux) or Cmd + Shift + P (macOS) to open the Command Palette.
   2. Type Reload Window in the search bar.
   3. Select the Reload Window command.
2. C/C++ environment: .vscode config may not be loaded on the first run. Please Reload Window to ensure the configuration is loaded.
   1. if you need source, please run sed -i 's/Types: deb/Types: deb deb-src/' /etc/apt/sources.list.d/debian.sources
3. upython(micropython):
   1. run sudo usermod -aG dialout $USER before use container, add youself to dialout
   2. hardware -> tty:
      • raspberry pico: /dev/ttyACM0
      • raspberry pi: /dev/ttyUSB0
   3. stub(support pylance): pip install micropython-XXX-stubs, search on pypi.org.
4. cuda:
   1. Make sure you have installed the NVIDIA driver for your Linux Distribution
   2. Note that you do not need to install the CUDA Toolkit on the host system, but the NVIDIA driver needs to be installed
   3. For instructions on getting started with the NVIDIA Container Toolkit, refer to the installation guide
   4. for china,maybe need a mirror site: run sed -i 's/developer.download.nvidia.com/developer.download.nvidia.cn/g' /etc/apt/sources.list.d/cuda-debian12-x86_64.list
5. rocm:
   1. run sudo apt install rocminfo
   2. run sudo usermod -aG video,render $USER before use container, add youself to video,render
   3. not include any package, please install what you want(apt install rocm)
   4. how to run jax/flax:
      1. apt install hipsolver hipfft miopen-hip rccl rocm-llvm rocprofiler-sdk hsa-amd-aqlprofile libamd-comgr2 libdw1 (I just want to minimize runtime; you can absolutely do a full ROCm installation.)
      2. see rocm-jax
   5. for china,maybe need a mirror site:
      1. run sed -i 's/repo.radeon.com/radeon.geekery.cn/g' /etc/apt/sources.list.d/amdgpu.list
      2. run sed -i 's/repo.radeon.com/radeon.geekery.cn/g' /etc/apt/sources.list.d/rocm.list
   6. how to run pytorch: PyTorch doesn’t require user-space ROCm support;
      1. run pip install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/rocm7.0

Image dependencies

• Denv : development environment
• Renv: runtime environment
• Gpu: gpu environment

graph LR
Debian-->B{Base}
Alpine-->B

B-->D{Denv}
D-->DA(C)
D-->DB(Cpp)
D-->DF(Debforge)
D-->DC(Python)
D-->DD(Zig)
D-->DE(Upython)

DC-->G{Gpu}-->GA(Rocm)
G-->GB(CUDA)

B-->R{RENV}
R-->RA(Zine)
R-->RB(Steam)
R-->RC(JRE_21)
R-->RD(Novnc)
R-->RE(Crane)

Loading

Mirror source

• debian ustc：https://mirrors.ustc.edu.cn/help/debian.html