LlamaFactory
Fine-tune 100+ LLMs with a point-and-click web UI or CLI.
About
LlamaFactory is the unified fine-tuning framework: SFT, RLHF, DPO, and LoRA/QLoRA across 100+ models (Llama, Qwen, Mistral, Gemma...) via CLI, API, or the LLaMA Board web UI — no training code required. 75.4k stars, Apache-2.0, used by Amazon, NVIDIA, and Aliyun, cited 1000+ times. (Renamed from hiyouga/LLaMA-Factory; old URLs redirect.) The UI makes fine-tuning deceptively easy — remember that dataset quality dominates: garbage in, garbage out. For programmatic training pipelines, pair with trl/axolotl.
What it can do
- **100+ models**: Llama, Qwen, Mistral, Gemma, and more
- **All methods**: SFT, RLHF, DPO, ORPO, LoRA/QLoRA
- **LLaMA Board**: web UI for training, eval, and inference
- **Multimodal**: vision-language fine-tuning support
- **Export**: merge adapters, quantize (GGUF), serve
- **Cited 1000+ times**: the research community's fine-tuning default
Setup
```bash
pip install llamafactory
llamafactory-cli webui # opens LLaMA Board
```
Or llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml. Docker images available. Docs in the repo.
Prerequisites
Python 3.10+; NVIDIA GPU (VRAM depends on model + LoRA vs full).
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: 75,428 GitHub stars (log-scaled vs catalog peers); blended with ~16,241 monthly package downloads; last push 3d ago; 2 releases/12mo; 1184 open issues. Manual review: unified fine-tuning workbench (100+ models, SFT/RLHF/DPO/LoRA, LLaMA Board UI), 75.4k stars, used by Amazon/NVIDIA/Aliyun; UI invites casual tuning — data quality dominates. License: Apache-2.0.
★ 75,428 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.2/10
- Maintenance · recency, releases, issue load
- 7.6/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · 75,428 GitHub stars (log-scaled vs catalog peers)
- · blended with ~16,241 monthly package downloads
- · last push 3d ago; 2 releases/12mo; 1184 open issues
- · manual review: unified fine-tuning workbench (100+ models, SFT/RLHF/DPO/LoRA, LLaMA Board UI), 75.4k stars, used by Amazon/NVIDIA/Aliyun; UI invites casual tuning — data quality dominates
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.