ollama - 💡(How to fix) Fix compiling ollama in debian 12. seems to try to run yum from redhat using build_linux.sh [6 comments, 2 participants]

=> ERROR [linux/arm64 base-arm64 2/2] RUN yum install -y yum-utils epel-release && dnf install -y clang ccache git && yum-config-manager --add-repo https://develop 4.4s => CANCELED [linux/arm64 jetpack-6 2/5] RUN apt-get update && apt-get install -y curl ccache unzip && curl -fsSL https://github.com/Kitware/CMake/releases/download/v3. 0.0s => CANCELED [linux/arm64 jetpack-5 2/5] RUN apt-get update && apt-get install -y curl ccache unzip && curl -fsSL https://github.com/Kitware/CMake/releases/download/v3. 0.0s => CANCELED [linux/amd64 base-amd64 2/2] RUN dnf install -y yum-utils ccache gcc-toolset-11-gcc gcc-toolset-11-gcc-c++ gcc-toolset-11-binutils && yum-config-manager -- 0.0s

[linux/arm64 base-arm64 2/2] RUN yum install -y yum-utils epel-release && dnf install -y clang ccache git && yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/sbsa/cuda-rhel8.repo: 4.249 exec /bin/sh: exec format error

2 warnings found (use docker --debug to expand):

• FromPlatformFlagConstDisallowed: FROM --platform flag should not use constant value "linux/arm64" (line 95)
• FromPlatformFlagConstDisallowed: FROM --platform flag should not use constant value "linux/arm64" (line 111) Dockerfile:24

23 | # install epel-release for ccache 24 | >>> RUN yum install -y yum-utils epel-release
25 | >>> && dnf install -y clang ccache git
26 | >>> && yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/sbsa/cuda-rhel8.repo 27 | ENV CC=clang CXX=clang++

ERROR: failed to build: failed to solve: process "/bin/sh -c yum install -y yum-utils epel-release && dnf install -y clang ccache git && yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/sbsa/cuda-rhel8.repo" did not complete successfully: exit code: 255

Fix Plan

The fix involves modifying the Dockerfile to correctly install dependencies for the target architecture.

• Update the base image to match the target architecture.
• Install necessary dependencies using the correct package manager for the base image.
• Configure the repository for CUDA installation.

Here are the steps:

1. Update the base image: Change the FROM instruction to use an image that matches the target architecture. For example, use nvidia/cuda:11.8.0-cudnn8-devel-ubuntu20.04 for Ubuntu-based images.
2. Install dependencies: Update the RUN instruction to install dependencies using apt-get for Ubuntu-based images or dnf/yum for RHEL-based images.
3. Configure CUDA repository: Update the yum-config-manager instruction to use the correct repository URL for the target architecture.

Example code snippet:

# Use an Ubuntu-based image for arm64 architecture
FROM nvidia/cuda:11.8.0-cudnn8-devel-ubuntu20.04

# Install dependencies using apt-get
RUN apt-get update && apt-get install -y curl ccache unzip

# Configure CUDA repository
RUN curl -fsSL https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-ubuntu2004.pin | tee /etc/apt/preferences.d/cuda-repository-pin-600
RUN apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/7fa2af80.pub
RUN add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/ /"
RUN apt-get update && apt-get install -y cuda

Note: The above example is for Ubuntu-based images. For RHEL-based images, use dnf/yum instead of apt-get.

Verification

To verify the fix, rebuild the Docker image and check if the dependencies are installed correctly. You can do this by running the following command:

docker build -t my-image .

Then, run a container from the image and check if the dependencies are installed:

docker run -it my-image /bin/bash

Inside the container, you can check if the dependencies are installed by running commands like clang --version or nvcc --version.