fork from https://github.com/ostris/ai-toolkit with some adjustments to fix issues when training LoRA on Modal

FLUX LoRA Training Setup Guide

Video Tutorial

This guide will help you set up the environment for training LoRA models with FLUX on Modal

Prerequisites

Before starting, make sure you have:

• Administrator privileges on your Windows system (for Windows users)
• Registered accounts on Modal and Hugging Face
• Accepted FLUX.1-dev license on Hugging Face (if using it)

Setup Instructions

For Windows Users:

1. Run setup_modal_training.bat as Administrator
   • Right-click on the script
   • Select "Run as administrator"

For MacOS Users:

1. Open Terminal and navigate to the project directory
2. Make the setup script executable:

   chmod +x setup_modal_training.sh

3. Run the setup script:

   ./setup_modal_training.sh

Common Steps for Both Platforms:

1. Follow the Modal token setup:

   • Go to https://modal.com/settings/tokens
   • Click "New Token"
   • Copy the command that looks like:

     modal token set --token-id ak-xxxx --token-secret as-xxxx
     

   • Paste the command when prompted

2. Prepare required files:

   • Configuration file:
   • Customize settings according to your needs in config/file modal_train_lora_flux.yaml
   • Environment file (.env):
     • Add your Hugging Face token in format:

     HF_TOKEN=hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
     

Installation Process

The setup script will automatically:

1. Install required software (if not already installed):

   • Python 3.10 or higher
   • Git (on MacOS, you can install it via Homebrew if needed)

2. Clone the ai-toolkit repository:

   • Windows: to C:\ai-toolkit (to prevent path length limitations)
   • MacOS: to the current directory

3. Set up virtual environment and dependencies

4. Configure Modal and Hugging Face tokens

Starting Training

Once all files are prepared, the training process will start automatically with:

modal run --detach run_modal.py --config-file-list-str=/root/ai-toolkit/config/modal_train_lora_flux.yaml

You can monitor the training progress and logs at: https://modal.com/logs

Troubleshooting

If you encounter any issues:

1. Make sure you're running the script as Administrator
2. Check that all required tokens are correctly set up
3. Verify that Python and Git are properly installed and added to PATH
4. Ensure all required files are present and correctly formatted

Note

If you need to restart the setup process:

1. Close the current window
2. Open a new Command Prompt or Terminal
3. Navigate back to the installation folder
4. Run the script again as Administrator

Download the content by running

modal volume get flux-lora-models your-model-name