Augmentoolkit
A tool that turns a folder of documents into a domain-expert training dataset and a fully fine-tuned custom LLM, optionally running entirely offline with no external API key.
Augmentoolkit creates domain-expert training datasets from a user's own documents, updating a model's knowledge so it becomes an expert in a chosen area. A user uploads documents and runs the tool to get a fully trained custom LLM, and it can optionally work entirely offline on the user's own computer for dataset generation on most hardware, without an external API key.
It also automatically creates a RAG-ready dataset from the same documents and can start up an inference server if traditional retrieval-based grounding is wanted alongside the fine-tuned model. Interfaces are provided for macOS and Linux, plus scripts for local inference that support multi-GPU setups through tensor parallelism, and Windows users are directed to run it under WSL.
It requires Python 3.11 and is installed by cloning the repository and running a setup script for the target platform.