Axolotl

Axolotl is an open-source fine-tuning tool whose defining choice is the configuration file: training runs are described in YAML rather than assembled in Python, and the project ships recipes, cookbooks and notebooks for the setups people actually run — LoRA, full fine-tuning, multi-GPU, and specific models.

That choice is the whole value proposition and the whole limitation. Describing a run in a config file makes it reproducible and comparable: two runs differ in a diff you can read, and a recipe that worked can be handed to someone else. It also means the tool covers the well-trodden paths well and gets awkward when your training loop needs something nobody has needed before — at which point you either extend the tool or drop to a framework like PyTorch and lose the convenience. As with any fine-tuning pipeline, the tool is not where quality comes from; the data is, and no configuration file fixes a bad dataset.