hiyouga/LlamaFactory

Project overview

A framework that provides zero-code fine-tuning for over 100 large language models and vision-language models through command-line and web interfaces. It enables continuous pre-training, supervised fine-tuning, and preference alignment using scalable methods like LoRA and QLoRA.

Project type

Model Development · Model Runtime · Data Processing

Deployment

Refer to project documentation

License

Apache-2.0

Best for

• AI engineers, developers, and researchers who need a unified environment to apply multiple training algorithms and tuning methods to large language models or vision-language models.
• Teams looking for local execution and zero-code configuration to adapt models without writing custom training scripts.

Key capabilities

• Enables the adaptation of over 100 large language models via command-line and web interfaces without writing code.
• Supports continuous pre-training, supervised fine-tuning, reward modeling, PPO, DPO, KTO, ORPO, and SimPO training.
• Offers 16-bit full-tuning, freeze-tuning, LoRA, and 2/3/4/5/6/8-bit QLoRA via AQLM/AWQ/GPTQ/LLM.int8/HQQ/EETQ.
• Provides an OpenAI-style API, Gradio UI, and CLI with a vLLM worker or SGLang worker.
• Provides a web interface powered by Gradio to configure and execute fine-tuning tasks.

Limitations and risks

• Full fine-tuning of large models demands significant VRAM; for example, pure_bf16 training of a 70B model requires approximately 600GB of VRAM.
• On Windows, users must manually install specific GPU versions of PyTorch and dependencies such as bitsandbytes.

Getting started

• Install the framework via pip or Docker. Launch the web interface by executing the command line interface, which provides access to zero-code configuration for model training.

Evidence and sources

• GitHub project description: Unified Efficient Fine-Tuning of 100+ LLMs & VLMs (ACL 2024)
• README: Easily fine-tune 100+ large language models with zero-code [CLI](#quickstart) and [Web UI](#fine-tuning-with-llama-board-gui-powered-by-gradio)
• README: Used by [Amazon](https://aws.amazon.com/cn/blogs/machine-learning/how-apoidea-group-enhances-visual-information-extraction-from-banking-documents-with-multimodal-models-using-llam…
• README: []
• README: Integrated methods: (Continuous) pre-training, (multimodal) supervised fine-tuning, reward modeling, PPO, DPO, KTO, ORPO, etc.

Reviewed at: 2026-07-17T05:02:50Z