search

AsahiLinux / docs

Hardware and software docs / wiki
Other
1.77k stars 48 forks source link

Docker image #95

Open johnnynunez opened 10 months ago

johnnynunez commented 10 months ago

Is it possible to use asahi linux as docker image for ML/DL? I mean, apple not distribute macos images in docker,(only linux images are distributed) so it's not possible to use MPS or MLX. Is it possible to use this asahi linux and run with gpu support?

angelbirth commented 10 months ago

AFAIK asahi linux only runs on bare metal, so either you install it on your mac, or use generic arm64 linux image for docker

johnnynunez commented 10 months ago

AFAIK asahi linux only runs on bare metal, so either you install it on your mac, or use generic arm64 linux image for docker

thanks! for example I would like to run this type of algorithms:

ARG PYTORCH="1.9.1"
ARG CUDA="11.1"
ARG CUDNN="8"

FROM pytorch/pytorch:${PYTORCH}-cuda${CUDA}-cudnn${CUDNN}-devel

RUN export TORCH_CUDA_ARCH_LIST="6.0 6.1 7.0 8.0+PTX" \
    && export TORCH_NVCC_FLAGS="-Xfatbin -compress-all" \
    && export CMAKE_PREFIX_PATH="$(dirname $(which conda))/../" 

# INIT OF FIX FOR GPG KEY -- Hopefully they will fix this soon
RUN apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/3bf863cc.pub
# END OF FIX FOR GPG KEY

# apt and apt-get update
RUN apt-get update && apt update && apt upgrade -y && apt-get clean
# apt and apt-get installations
RUN apt install -y vim git unzip \
    && apt-get install -y libosmesa6-dev build-essential libgl1-mesa-dev libglu1-mesa-dev freeglut3-dev

RUN mkdir /software
ENV SW_PATH=/software
WORKDIR $SW_PATH

ENV CONDA_DIR /opt/conda
ENV PATH $CONDA_DIR/bin:$PATH

RUN echo "Installing basic conda environment..." \
    && conda create --name main python=3.9.5 \
    && . /opt/conda/etc/profile.d/conda.sh \
    && conda activate main \
    && conda install numpy \
    && pip install torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html \
    && pip install tensorflow==2.8.0 \
    && pip install opencv-contrib-python pyyaml yacs tqdm colorama matplotlib cython tensorboardX imageio imageio-ffmpeg munkres pandas seaborn h5py sklearn

but using asahi as base, because use ubuntu vainilla doesn’t have mps acceleration

ziw-liu commented 7 months ago

MPS is Apple's proprietary API and is not available on Linux unless someone writes a translation layer (like dxvk).