Axolotl

Axolotl is an open source tool that makes fine-tuning language models easier to set up and repeat. It is built around configuration files, so a training run is described in one place rather than in custom code.

Fine-tuning adapts a pretrained model to new data or a particular task. Axolotl takes a YAML file that names the model, the dataset and the training settings, then runs the job. It supports a wide range of model architectures and ships with recipes and notebooks for common setups.

• Fine-tune many model families from YAML config files
• Supports Llama, Mistral, Qwen, Gemma and other families
• Run training in your own cloud or on Docker and Kubernetes
• Ready-made recipes and notebooks to start from
• Guides for efficient multi-GPU training

Machine learning engineers and small AI teams that fine-tune open models often and want a repeatable setup. Researchers who want to test techniques across several model families will also find it useful.

The project is free and released under the Apache-2.0 licence. Compute is the real cost, because training large models needs GPUs, whether you rent them from a cloud provider or run them on your own servers. The project also runs a Discord server where users ask questions.