LLaMA-Factory (now spelled LlamaFactory) is an open-source toolkit for fine-tuning many open language and vision models through a command line or a web interface.
In plain words
LLaMA-Factory gathers many training methods in one place: continued pretraining, supervised fine-tuning, reward modelling and preference methods such as PPO and DPO. You can train the full model, freeze parts of it, or use LoRA and low-bit QLoRA, which need far less GPU memory. A browser interface called LLaMA Board lets you set up training, evaluation and inference without writing code.
A quick example
Imagine a student with one consumer GPU who wants a chat model tuned on their dataset. They open the LlamaBoard web UI, choose a base model, pick 4-bit QLoRA to fit in memory and start training from the browser.
Where you'll meet it
- Fine-tuning open models such as Qwen, Llama and Gemma.
- Teaching and experimenting with fine-tuning methods.
Related
Sources
- DocsLlamaFactory README, hiyouga (GitHub) · read 29 Sept 2026
- DocsLlamaFactory repository, hiyouga (GitHub) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.