Awesome AI Agents › LLM-Powered Platforms

hiyouga/LlamaFactory

⭐ 74750 Python added to this list on 2026-01-04 repository created 2023-05-28

LLaMA Factory is a comprehensive and unified platform designed for efficient fine-tuning of over 100 large language models (LLMs) and vision-language models (VLMs). It supports a wide range of models including LLaMA, LLaVA, Mistral, Mixtral-MoE, Qwen, DeepSeek, ChatGLM, Phi, and many others, making it a versatile tool for researchers and developers working with state-of-the-art AI models. The platform integrates various training methods such as continuous pre-training, multimodal supervised fine-tuning, reward modeling, and advanced reinforcement learning techniques like PPO, DPO, KTO, and ORPO. It also offers scalable resource management with options for 16-bit full-tuning, freeze-tuning, and low-bit quantization techniques like LoRA and QLoRA, supported by advanced quantization algorithms including AQLM, AWQ, GPTQ, and others. LLaMA Factory incorporates cutting-edge optimization algorithms such as GaLore, BAdam, APOLLO, Adam-mini, Muon, and OFT, alongside practical performance enhancements like FlashAttention-2, Unsloth, Liger Kernel, and RoPE scaling. The platform is designed to handle a variety of tasks including multi-turn dialogue, tool usage, image and video understanding, visual grounding, and audio processing. It provides robust experiment monitoring through integrations with LlamaBoard, TensorBoard, Wandb, MLflow, and SwanLab, facilitating comprehensive tracking and analysis of training processes. For deployment and inference, LLaMA Factory supports OpenAI-style APIs, Gradio UI, and CLI interfaces, enhanced by vLLM and SGLang workers for faster performance. The platform is accessible via local setups, cloud services like Colab and PAI-DSW, and offers Docker support for easy deployment. It is widely adopted by major industry players such as Amazon, NVIDIA, and Aliyun, reflecting its reliability and effectiveness in real-world applications. Extensive documentation, community support through WeChat, Discord, and other channels, and a rich ecosystem of tutorials and blogs further enhance its usability and learning curve. Overall, LLaMA Factory stands out as a powerful, flexible, and user-friendly solution for fine-tuning and deploying large-scale language and multimodal models, catering to both academic research and industrial applications.

https://github.com/hiyouga/LlamaFactory

16-bit-tuningadam-miniagentaialiyunamazonapolloaqlmaudio-understandingawqbadamchatglmclicloud-trainingcolabcontinuous-pre-trainingdeepseekdockerdpofine-tuningflashattention-2freeze-tuninggaloregemmagptgptqgradio-uiimage-understandinginstruction-tuningktolarge-language-modelsliger-kernelllamallama3llamaboardllavallmloramistralmixtral-moemlflowmoemulti-turn-dialoguemuonnlpnvidiaoftopenai-style-apiorpopai-dswpeftphippoqloraquantizationqwenreward-modelingrlhfrope-scalingsglangsupervised-fine-tuningswanlabtensorboardtransformersunslothvideo-recognitionvision-language-modelsvllmwandb