Icon for LLaMA-Factory

LLaMA-Factory

Open Source Free Trial

Open-source fine-tuning framework for 100+ LLMs with a web UI

LLaMA-Factory is a popular open-source framework for fine-tuning large language models. It supports over 100 model architectures with LoRA, QLoRA, full fine-tuning, RLHF, and DPO methods. Features a web-based UI (LLaMA Board) for no-code training configuration. Includes built-in dataset preprocessing and evaluation tools.

Pricing: Free

HQ 🇺🇸 United States
GitHub 71,596 stars
Screenshot of LLaMA-Factory webpage

What is LLaMA-Factory?

LLaMA-Factory is an open-source framework for fine-tuning large language and vision-language models. It wraps the full training stack, data preprocessing, training, and evaluation, behind a single CLI (llamafactory-cli) and an optional Gradio web UI called LLaMA Board, so you can fine-tune a model without writing the training loop yourself. It supports 100+ model families including LLaMA, Mistral, Qwen, Gemma, DeepSeek, GLM, Phi, and Falcon. The project is Apache-2.0 licensed and widely adopted (around 72k GitHub stars as of mid-2026, with reported use at Amazon, NVIDIA, and Aliyun).

Methods and hardware efficiency

It covers the main training approaches in one place: full-parameter and freeze tuning, LoRA and QLoRA, plus pretraining, supervised fine-tuning, reward modeling, and preference optimization (PPO, DPO, KTO, ORPO, SimPO). For fitting larger models onto limited GPUs it supports 2/3/4/5/6/8-bit quantization via AQLM, AWQ, GPTQ, LLM.int8, HQQ, and EETQ. Performance integrations include FlashAttention-2, Unsloth, DeepSpeed, and faster inference through vLLM and SGLang.

Pricing

LLaMA-Factory is free and open-source (Apache-2.0). There is no hosted service or paid tier; you run it on your own hardware or rented GPUs, so the real cost is the compute you train on rather than the framework.

Who it's for

It suits developers and researchers who want to fine-tune open models on their own infrastructure with broad method and model coverage, and a UI for configuring runs without code. Compared with Unsloth (focused on single-GPU speed and memory) or Axolotl (config-file driven), LLaMA-Factory's draw is breadth: many models, many training methods, and both CLI and web workflows in one tool. For a fully managed alternative that removes the infrastructure entirely, hosted fine-tuning platforms are the trade-off in the other direction.

Work on LLaMA-Factory? Feature it at the top of Fine-tuning.

Is your product missing?

Add it here →