LLaMA-Factory

LLaMA-Factory

LLaMA-Factory is a unified, efficient framework for fine-tuning large language models. It supports 100+ model architectures and all major training methods through a clean interface — from the llamafactory-cli command line to a fully-featured web UI (LlamaBoard). It's become the go-to toolkit for researchers and engineers fine-tuning open-source LLMs.

Key Features

• 100+ LLMs supported: Llama, Mistral, Qwen, Baichuan, Gemma, Phi, Yi, InternLM, DeepSeek
• Training methods: SFT, reward modeling, PPO, DPO, KTO, ORPO, SimPO
• Efficient fine-tuning: LoRA, QLoRA (4-bit/8-bit), full fine-tuning
• LlamaBoard: drag-and-drop web UI for dataset prep, training, and evaluation
• FlashAttention-2 and DeepSpeed ZeRO integration
• Supports multi-turn conversation datasets and custom templates

Quick Start

pip install llamafactory

# Launch web UI
llamafactory-cli webui

# Or train via CLI
llamafactory-cli train \
  --model_name_or_path meta-llama/Llama-3.1-8B \
  --finetuning_type lora \
  --dataset alpaca_en \
  --output_dir ./output

Install via ai-supply

npx ai-supply add llamafactory-unified-finetuning

Curated mirror of the open-source LLaMA-Factory (Apache-2.0). Get it from the source.