# hiyouga/LlamaFactory

Unified Efficient Fine-Tuning of 100+ LLMs & VLMs (ACL 2024)

Repository: https://github.com/hiyouga/LlamaFactory
Canonical: https://ross.abutalabs.com/products/llamafactory
Homepage: https://llamafactory.readthedocs.io
Language: Python
License: Apache-2.0
License Family: permissive
Topics: fine-tuning, llama, llm, peft, transformers, rlhf, qlora, quantization, qwen, instruction-tuning, gpt, lora, large-language-models, agent, ai, moe, llama3, deepseek, gemma, nlp
Last push: 2026-08-26T21:14:26+00:00

## Health v2 (maintenance only)
Score: 83/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 62, longevity 85
- inputs: {"age_days": 1193, "days_push": 7, "days_rel": 95, "gap_med": 106, "n_releases_24m": 6}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 74382, forks 9101 (observed 2026-08-28T04:12:21.442791+00:00)

## What it is
LlamaFactory is a unified, efficient fine-tuning framework for 100+ large language models and vision-language models, supporting LoRA, QLoRA, full-parameter tuning, and RLHF methods like DPO and PPO. It offers a no-code WebUI (LLaMA Board), CLI, distributed training backends (DeepSpeed, FSDP, Ray), and quantization/inference/export tooling.

## Use cases
- fine-tune llama on my own data
- train a lora adapter for qwen
- run qlora fine-tuning on a single gpu
- do dpo or rlhf training for a chat model
- fine-tune a multimodal vision-language model
- merge and quantize a fine-tuned model
- fine-tune an llm without writing code
- distributed multi-gpu llm training with deepspeed

## When to choose
- you want to fine-tune or align many open LLMs/VLMs with one unified tool
- you need LoRA/QLoRA, RLHF (DPO, PPO, KTO, ORPO), and quantization in a single framework
- you prefer a no-code WebUI or simple CLI over writing training scripts
- you need distributed training (DeepSpeed, FSDP, Ray, Megatron) and experiment tracking

## When to avoid
- you need pretraining from scratch at massive scale rather than fine-tuning
- you want a minimal, hackable training script instead of a full framework
- your model architecture is not among the supported templates
- you only need inference/serving without any training

## Facets
- artifact type: framework
- maturity: active
- function: llm-training, machine-learning, deep-learning, cli, gui
- domain: large-language-models, machine-learning, deep-learning, artificial-intelligence
- platform: python
- tags: fine-tuning, lora, qlora, rlhf, dpo, peft, quantization, instruction-tuning, webui, multimodal, distributed-training, deepspeed, vllm, natural-language-processing, linux, docker, gpu, web-server

## Member repositories
- hiyouga/LlamaFactory (main) score 83

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:12:21.442791+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T16:14:59.061032+00:00, confidence not recorded.
  - readme: https://github.com/hiyouga/LlamaFactory (fetched 2026-08-28T04:12:21.442791+00:00, sha 9500fce77fea)
  - homepage: https://llamafactory.readthedocs.io (fetched 2026-08-28T17:46:07.565499+00:00, sha 63d872c4b64d)
- Data as of 2026-08-30T08:39:29.467469+00:00.