Firefly is a one-stop large model training tool that supports full-parameter training, LoRA, and QLoRA efficient training methods. It supports pre-training, supervised fine-tuning (SFT), and Direct Preference Optimization (DPO) across a wide range of open-source LLMs including Qwen, Yi, Llama, Gemma, MiniCPM, InternLM, and others. The project integrates with Unsloth for accelerated training and reduced memory usage, and provides configuration-based training for easy model switching.