GH Repository · hiyouga
LlamaFactory
Unified Efficient Fine-Tuning of 100+ LLMs & VLMs (ACL 2024)
- stars
- 74,783
- 30-day movement
- +8729/day
- Related entries
- 61
- Connections
- 3
agent-appnlpgemmadeepseekllama3moeaiagentinstruction-tuningpythonllmfine-tuningpeftPythonrlhfqlorallamaquantizationqwenmakegpttransformerslarge-language-modelslora
LlamaFactory is a Python toolchain for unified, efficient fine-tuning of over 100 large language models and vision-language models, published as an ACL 2024 paper. It covers parameter-efficient methods such as LoRA, QLoRA, and PEFT, plus RLHF, quantization, and instruction tuning across model families like Llama, Qwen, DeepSeek, Gemma, and MoE architectures.
You reach for it when you want one framework to fine-tune many different open LLMs and VLMs efficiently instead of stitching together separate scripts per model.
Use it to
- Fine-tune Llama, Qwen, or DeepSeek models with LoRA or QLoRA
- Run RLHF or instruction-tuning pipelines
- Quantize models for efficient training or inference
- Adapt vision-language models
- Experiment across 100+ model architectures with one toolchain
For ML engineers and researchers fine-tuning open-source LLMs
- Role
- agent-app
- Language
- Python
- Licence
- Apache-2.0
- Forks
- 9,159
- Open issues
- 998
- Last push
- 2026-09-14
- Latest release
- v0.0.9 · 2023-07-15
- Skills shipped
- 1
topicsfine-tuningloraqlorarlhfllminstruction-tuning