Awesome AI Agents › AI Developer Toolkit

unslothai/unsloth

⭐ 76125 Python added to this list on 2025-07-30 repository created 2023-11-29

Unsloth is a cutting-edge project focused on fine-tuning and reinforcement learning for large language models (LLMs). It enables users to train advanced models such as Qwen3, Llama 4, DeepSeek-R1, Gemma 3, and text-to-speech (TTS) models significantly faster and with much lower VRAM consumption compared to traditional methods. The project boasts performance improvements like training speeds up to 2.2 times faster and VRAM usage reductions up to 80%, making it highly efficient for resource-constrained environments. The project provides beginner-friendly notebooks for free fine-tuning of various popular models, including Gemma 3n, Qwen3, Llama 3.2, Phi-4, Mistral, and Orpheus-TTS. These notebooks allow users to add their datasets, run the training process easily, and export the fine-tuned models in multiple formats such as GGUF, Ollama, vLLM, or Hugging Face. Unsloth supports a wide range of models and tasks, including conversational AI, reasoning with long context windows, vision models, and text-to-speech. Unsloth also introduces innovative features like Dynamic 2.0 quantization, which optimizes model accuracy and VRAM usage, and supports multi-GPU setups. The project is continuously updated with new models and capabilities, including support for Meta's Llama 4, Apple's Cut Cross Entropy for extended context lengths, and advanced reasoning techniques like GRPO. It also offers detailed documentation and community support through Discord and other channels. Overall, Unsloth is a comprehensive toolkit designed to democratize and accelerate the fine-tuning and deployment of large language models, making it accessible for researchers, developers, and enthusiasts who want to customize and optimize LLMs efficiently.

https://github.com/unslothai/unsloth

agentaiconversational-aideepseekdeepseek-r1fine-tuninggemmagemma-3gemma3ggufgrpohugging-facelarge-language-modelsllamallama-4llama3llmllmslong-contextloramistralmodel-trainingmulti-gpunotebooksollamaopenaipytorchquantizationqwenqwen3reasoningreinforcement-learningtext-to-speechttsunslothvision-modelsvllmvram-optimization