AI & ML Workloads: GPU and Fooocus

Introduction

We show how to use a ThreeFold GPU node on the grid to deploy AI workloads for graphics and image generation. We will be using Fooocus and a full virtual machine with a GPU card.

Prerequisites

• A TFChain account
• TFT in your TFChain account
  • Buy TFT
  • Send TFT to TFChain
• Reserve a dedicated GPU node on the ThreeFold Grid
• Deploy a full VM running Ubuntu 22.04
• SSH into the node with Wireguard

Prepare the System

• Update the system

  dpkg --add-architecture i386
  apt-get update
  apt-get dist-upgrade
  reboot
  

• Check the GPU info

  lspci | grep VGA
  lshw -c video
  

Install the GPU Driver

• Download the latest Nvidia driver
  • Check which driver is recommended

    apt install ubuntu-drivers-common
    ubuntu-drivers devices
    

  • Install the recommended driver (e.g. with 535)

    apt install nvidia-driver-535
    

• Check the GPU status

  nvidia-smi
  

Now that the GPU node is set, let's install and launch Fooocus.

Install the Prerequisites and Launch Fooocus

We install the prerequisites, including Miniconda, clone the repository, download the models and launch Fooocus.

• Install the prerequisites

  apt update
  apt install python3-pip python3-dev
  pip3 install --upgrade pip
  wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
  bash Miniconda3-latest-Linux-x86_64.sh
  

• Reload the shell to enable Conda
• Clone the Fooocus directory and install the Python requirements with Miniconda

  git clone https://github.com/lllyasviel/Fooocus.git
  cd Fooocus
  conda env create -f environment.yaml
  conda activate fooocus
  pip install -r requirements_versions.txt
  

• Download the models with conda and deploy Fooocus

  conda activate fooocus
  python entry_with_update.py
  

• Create an SSH tunnel to the VM

  ssh -4 -L 7865:127.0.0.1:7865 root@10.20.4.2
  

Use Fooocus

You can then access Fooocus:

127.0.0.1:7865

Here are some image generated with Fooocus: