Nova AI. Инструкция по установке GPU драйверов

#!/usr/bin/env bash
set -Eeuo pipefail

# Vars
ML_REPOSITORY_URL="https://s3.k2.cloud/downloads-nova-platform-io"
NVIDIA_CONTAINER_TOOLKIT_VERSION=1.17.8-1
UNIVERSE_INSTALLATION=0

check_user_permissions() {
  if [[ $EUID -ne 0 ]]; then
    echo "Error: This script must be run as root"
    exit 1
  fi
}

verify_drivers_installed(){
  if command -v nvidia-smi >/dev/null 2>&1; then
  if nvidia-smi; then
    echo
    echo "GPU driver verification completed"
  else
    echo
    echo "nvidia-smi found, but GPU is not fouded"
  fi
else
  echo "nvidia-smi not found"
fi
}

install_drivers() {
  if [[ -f /etc/astra_version ]]; then
    OS_TYPE="astralinux"
    OS_VERSION=$(grep VERSION_ID /etc/os-release | awk -F "=" '{ print $2}' | tr -d '"')
    if [[ "$OS_VERSION" == "1.7_x86-64" ]]; then
      echo "Detected OS: $OS_TYPE $OS_VERSION"
      echo "deb [trusted=yes] $ML_REPOSITORY_URL/deb/nova-ml/astra/1.7.6.uu2/ nova main" > /etc/apt/sources.list.d/nova-ml.list
      echo "deb [trusted=yes] $ML_REPOSITORY_URL/deb/nova-platform/astra/1.7.6/ nova main" > /etc/apt/sources.list.d/nova-platform.list
      apt-get update
      echo blacklist nouveau >> /etc/modprobe.d/blacklist.conf
      echo options nouveau modeset=0 >> /etc/modprobe.d/blacklist.conf
      sed -i -e "s/^[[:space:]]*nouveau[[:space:]]\+modeset[[:space:]]*=[[:space:]]*1/#nouveau modeset=1/" /etc/initramfs-tools/modules
      update-initramfs -u -k all
      wget $ML_REPOSITORY_URL/nova-ml/ALSE17-NVidia-all-240923-1.7.6.tar.gz
      tar xf ALSE1?-NVidia-all-*.tar.gz -C /srv
      echo "deb file:/srv/ALSE17-NVidia-all-240923-1.7.6 1.7_x86-64 non-free" >> /etc/apt/sources.list
      apt update
      apt install -y nvidia-driver-555
      apt-get install -y \
        nvidia-container-toolkit=${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        nvidia-container-toolkit-base=${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        libnvidia-container-tools=${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        libnvidia-container1=${NVIDIA_CONTAINER_TOOLKIT_VERSION}
      reboot
    elif [[ "$OS_VERSION" == "1.8_x86-64" ]]; then
      echo "Detected OS: $OS_TYPE $OS_VERSION"
      echo "deb [trusted=yes] $ML_REPOSITORY_URL/deb/nova-ml/astra/1.8.1.uu1/ nova main" > /etc/apt/sources.list.d/nova-ml.list
      echo "deb [trusted=yes] $ML_REPOSITORY_URL/deb/nova-platform/astra/1.8.1/ nova main" > /etc/apt/sources.list.d/nova-platform.list
      apt-get update
      apt update
      apt install -y nvidia-driver-550
      apt-get install -y \
        nvidia-container-toolkit=${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        nvidia-container-toolkit-base=${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        libnvidia-container-tools=${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        libnvidia-container1=${NVIDIA_CONTAINER_TOOLKIT_VERSION}
      reboot
    else
      echo "Error: Unsupported version $OS_TYPE $OS_VERSION"
      exit 1
    fi
  elif [[ -f /etc/redos-release ]]; then
    OS_TYPE="red"
    OS_VERSION=$(grep VERSION_ID /etc/os-release | awk -F "=" '{ print $2}' | tr -d '"')
    if [[ "$OS_VERSION" == "7.3.4" ]]; then
      echo "Detected OS: $OS_TYPE $OS_VERSION"
      tee /etc/yum.repos.d/nova-ml.repo > /dev/null <<EOF
[nova-ml]
name=nova-ml
baseurl=$ML_REPOSITORY_URL/rpm/nova-ml/redos/7/
gpgcheck=False
module_hotfixes=1
priority=1
EOF
      tee /etc/yum.repos.d/nova-platform.repo > /dev/null <<EOF
[nova-platform]
name=nova-platform
baseurl=$ML_REPOSITORY_URL/rpm/nova-platform/redos/7/
gpgcheck=False
module_hotfixes=1
priority=1
EOF
      dnf makecache
      dnf install -y nvidia-drivers-570.144 nvidia-settings-570.144 nvidia-kmod-570.144 nvidia-persistenced-570.144 nvidia-xconfig-570.144 nvidia-modprobe-570.144 xorg-x11-drv-nvidia-570.144 xorg-x11-drv-nvidia-cuda-libs-570.144
      dnf install -y \
        nvidia-container-toolkit-${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        nvidia-container-toolkit-base-${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        libnvidia-container-tools-${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        libnvidia-container1-${NVIDIA_CONTAINER_TOOLKIT_VERSION}
      reboot
    elif [[ "$OS_VERSION" == "8.0" ]]; then
      echo "Detected OS: $OS_TYPE $OS_VERSION"
      tee /etc/yum.repos.d/nova-ml.repo > /dev/null <<EOF
[nova-ml]
name=nova-ml
baseurl=$ML_REPOSITORY_URL/rpm/nova-ml/redos/8/
gpgcheck=False
module_hotfixes=1
priority=1
EOF
      tee /etc/yum.repos.d/nova-platform.repo > /dev/null <<EOF
[nova-platform]
name=nova-platform
baseurl=$ML_REPOSITORY_URL/rpm/nova-platform/redos/8/
gpgcheck=False
module_hotfixes=1
priority=1
EOF
      dnf makecache
      sudo -E dnf install nvidia-settings-580.76.05 nvidia-kmod-580.76.05 nvidia-persistenced-580.76.05 nvidia-xconfig-580.76.05 nvidia-modprobe-580.76.05 xorg-x11-drv-nvidia-580.76.05 xorg-x11-drv-nvidia-cuda-libs-580.76.05 nvidia-drivers-580.76.05
      dnf install -y \
        nvidia-container-toolkit-${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        nvidia-container-toolkit-base-${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        libnvidia-container-tools-${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        libnvidia-container1-${NVIDIA_CONTAINER_TOOLKIT_VERSION}
      reboot
    else
      echo "Error: Unsupported version $OS_TYPE $OS_VERSION"
      exit 1
    fi
  elif [[ -f /etc/almalinux-release ]]; then
    OS_TYPE="almalinux"
    OS_VERSION=$(grep VERSION_ID /etc/os-release | awk -F "=" '{ print $2}' | tr -d '"')
    if [[ "$OS_VERSION" == "8.10" ]]; then
      echo "Detected OS: $OS_TYPE $OS_VERSION"
      tee /etc/yum.repos.d/nova-ml.repo > /dev/null <<EOF
[nova-ml]
name=nova-ml
baseurl=$ML_REPOSITORY_URL/rpm/nova-ml/rhel/8/
gpgcheck=0
module_hotfixes=1
priority=1
EOF
      tee /etc/yum.repos.d/nova-platform.repo > /dev/null <<EOF
[nova-platform]
name=nova-platform
baseurl=$ML_REPOSITORY_URL/rpm/nova-platform/rhel/8/
gpgcheck=False
module_hotfixes=1
priority=1
EOF
      dnf makecache
      dnf install -y kmod-nvidia-580.82.07-4.18.0-553.75.1-580.82.07 libnvidia-cfg-580.82.07 libnvidia-fbc-580.82.07 libnvidia-gpucomp-580.82.07 libnvidia-ml-580.82.07 nvidia-driver-580.82.07 nvidia-driver-libs-580.82.07 nvidia-modprobe-580.82.07 nvidia-persistenced-580.82.07 nvidia-settings-580.82.07 nvidia-xconfig-580.82.07 nvidia-driver-cuda-580.82.07 nvidia-driver-cuda-libs-580.82.07
      dnf install -y \
        nvidia-container-toolkit-${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        nvidia-container-toolkit-base-${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        libnvidia-container-tools-${NVIDIA_CONTAINER_TOOLKIT_VERSION} \
        libnvidia-container1-${NVIDIA_CONTAINER_TOOLKIT_VERSION}
      reboot
    else
      echo "Error: Unsupported version $OS_TYPE $OS_VERSION"
      exit 1
    fi
  else
    echo "Error: Unsupported operating system"
    exit 1
  fi
}

prepare_for_setup(){
  if [[ $UNIVERSE_INSTALLATION -ne 0 ]]; then
    echo "Nova is installed from Universe, enter the FQDN of the repository server - for example: https://repo.universe.mycompany.local"
    read ML_REPOSITORY_URL
    echo "Universe FQDN - $ML_REPOSITORY_URL"
  else
    echo "Nova is installed from public"
  fi
}

ask_for_repos(){
  echo "Please prepare OS repository with drivers dependencies, enter y - if repositories if configured"
  echo "[y/N]"
  read REPOS_READY
  if [ "$REPOS_READY" != "y" ]; then
    exit 1
  fi
}

usage() {
  cat <<EOF
Использование:
  $0 опции

Опции:
  --universe                 Включить при установке Nova через Universe
  -h, --help                 Показать возможные опции
EOF
}

main(){
  check_user_permissions
  ask_for_repos
  prepare_for_setup
  install_drivers
  verify_drivers_installed
}

# Parse options
while [ $# -gt 0 ]; do
  case "$1" in
    -h|--help)
      usage; exit 0
      ;;
    --universe)
      UNIVERSE_INSTALLATION=1
      ;;
  esac
  shift
done

main