axolotl-ai-cloud/axolotl ? reverse-engineered prompt

Build me a Python tool that makes it easy to fine tune open source language models from a simple config file.

I want something that helps me take a model, point it at my training data, and run common training styles like supervised fine tuning and preference training without having to stitch everything together myself. It should work well on a single GPU but also support bigger setups when I need them, and it should be easy to launch from the command line or inside Docker.

Please include a few ready to use example configs and datasets, clear docs for getting started, and sensible defaults so a beginner can run a first training job quickly. I also want support for common efficiency tricks like LoRA and quantized training if possible. If there are current best practices or library docs that matter, look them up online while you build it.