Axolotl

Axolotl is an open-source fine-tuning toolkit that uses a single YAML configuration file across dataset preprocessing, training, evaluation, quantization, and inference. It supports LoRA, QLoRA, full fine-tuning, RLHF (GRPO, reward modeling), and multimodal fine-tuning. Works with LLaMA, Mistral, Mixtral, and other Hugging Face models. Performance features include multipacking, Flash Attention, multi-GPU (FSDP, DeepSpeed), and multi-node training. Designed to make fine-tuning accessible while supporting advanced configurations.