Augmentoolkit
An open-source LLM training toolkit that turns documents into specialist datasets, with optional offline data generation and rented training compute.
Augmentoolkit turns documents into datasets for training an LLM on a particular subject. It's for people building custom models around material such as research papers or fictional lore.
- Document-based training: Generate subject-specific training data and carry the workflow through to a trained custom LLM.
- Offline data generation: Create datasets on your own computer without an external API key, using its dedicated generation model; speed depends on your hardware.
- Model and API choice: Generate data with DeepSeek or Llama, or use an external service such as DeepInfra when you want faster generation.
- Retrieval and inference: Produce a dataset for retrieval-augmented generation and start an inference server for the resulting model.
- Discord bots: Share custom models through Discord bots whose hosting code runs on your computer.
The toolkit is open source under the MIT license and offers graphical and command-line interfaces. It supports Linux and macOS; WSL is the recommended route on Windows. Local data generation is optional. Using an external model API sends that generation work to the provider.
Training requires a powerful machine or rented compute. Augmentoolkit can arrange the rental automatically if you choose that route.
Logotyper
