# LLaMA Factory(Nvidia GPU)

LLaMA Factory(Nvidia GPU) is a self-hosted ai app. LLaMA Factory is an open-source, self-hosted framework for fine-tuning and serving large language models on NVIDIA GPU hardware. It is open source and self-hosted, installable on Docker Hub and ZimaOS.

Canonical page: https://ourlocalhost.com/app/3803/llama-factory-nvidia-gpu

## Description

LLaMA Factory is an open-source, self-hosted framework for fine-tuning and serving large language models on NVIDIA GPU hardware. It provides a web interface for more than 100 models and supports LoRA, QLoRA, full-parameter and freeze training, quantization, multi-GPU training, and an integrated inference API for self-hosters who want a unified model-customization workflow.

## Platform availability

Available on 1 of 10 platforms: ZimaOS.

## Health score

85/100 (Excellent), based on 4 of 5 factors.
- Maintained: 100 (last commit 15 days ago)
- Popular: 98 (75,942 GitHub stars)
- Easy to install: 35 (1 install platform, install notes, Docker image)
- Security check: 89 (no serious findings, 1 minor, CI, tests)

## Links

- GitHub: https://github.com/hiyouga/LLaMA-Factory
- Documentation: https://llamafactory.readthedocs.io/
- Docker Hub: https://hub.docker.com/r/hiyouga/llamafactory
- ZimaOS: https://github.com/IceWhaleTech/CasaOS-AppStore/tree/main/Apps/LLaMA-Factory_Nvidia
- Official website: https://llamafactory.readthedocs.io

## Common questions

### Can installing LLaMA Factory cause a PyTorch version conflict?

Yes. On a DGX Spark, the setup playbook can hit a conflict between the Docker environment and the LLaMA installation at step 4.

### What should I set before starting the LLaMA Factory API server?

Set API_PORT to a free port such as 8000 and CUDA_VISIBLE_DEVICES to the GPU you want to use. Open that port if another machine will call it.

### Which SFT config values must I adjust before training?

Use the sample SFT config as a starting checklist and change model_name_or_path, dataset, and output_dir. Wrong paths or a batch size that is too large can crash training, exhaust memory, or save the wrong files.

### Can I launch a first training run without editing YAML files?

Yes. Launch the WebUI with llamafactory-cli webui and use its forms to choose the model, dataset, and LoRA settings.

## Alternatives

- Unsloth: Unsloth is a cross-platform desktop app (Windows, macOS, Linux) that lets you download, run, and fine-tune open-source large language model…

---
Source: Our Local Host (https://ourlocalhost.com/app/3803/llama-factory-nvidia-gpu), the self-hosted and open-source app directory.
