A web platform for running and monitoring LLM fine-tuning workflows using Axolotl framework.

Features

• Supports the Axolotl framework (fine-tuning, evaluation, and inference modes)
• Supports datasets from Hugging Face
• Supports Weights & Biases (wandb) to track and visualize fine-tuning runs

Installation

git clone https://github.com/froyo75/FineP0D.git
cd FineP0D

Backend

Create a new Python virtual environment, activate it, and install the required dependencies:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Frontend

To build the UI, run the following command in the ui directory:

npm install
npm run build

Setup and start the app

Use the provided .env.example file to create your .env configuration, then start the app:

cp .env.example .env

Configure your API keys for Hugging Face and W&B in the config/p0d.json file.

Note: The project currently supports only the Axolotl framework.

Backend

python main.py

Note: A new, unique session key is generated at each startup and written to logs/p0d.log.

Frontend

cd ui
npm run preview

Usage

To add a new FineP0D server, click the + Add button, then enter the server URL and the session key generated by the backend.

Hosting the fineP0D server with Google Colab (Optional)

You can fine-tune models on Google Colab using a free T4 GPU. Here is a sample Google Colab notebook to run the setup with Ngrok. The examples/ngrok.yml file specifies tunnels that expose the backend server and the Gradio server (when inference is running).