axolotl-ai-cloud/axolotl is an open-source framework that runs LLM training and fine-tuning via a configuration file (YAML). It is written in Python and published under the Apache-2.0 license. The latest release currently confirmed is v0.21.0, but because release notes for this version were not provided, the changed features and fix list cannot be pinned down.
The README describes this project as a free open-source LLM finetuning tool, and documents training configurations including LoRA and QLoRA together with distributed training support. LoRA is a way to fine-tune that learns only additional weights without directly updating the existing weights of a huge model, while QLoRA layers LoRA on top of training with weights compressed to a small number of bits. Distributed training, which splits training across several devices, is also covered in the YAML recipes. A YAML recipe is a configuration file that records the model, the data and the training method in text, so the training setup is gathered in one place. The repository has 12603 stars.
That said, the given records contain no benchmark figures comparing performance. Of the score items, only momentum is filled in at 3.2, while the novelty, technology and quality scores are empty, so it is premature to judge how far it differs from other fine-tuning tools. The repository was created on April 14, 2023, and the verifiable evidence goes no further than the repository description and the latest release number. Taken together, the Apache-2.0 license and the YAML-recipe-centric structure put it within reach of consideration for teams that already have a training pipeline.