Unsloth is an open-source Python framework that enables fast, memory-efficient fine-tuning and reinforcement learning for major LLMs—including Llama, Mistral, Gemma, Qwen, and DeepSeek—on consumer hardware and platforms like Google Colab and Kaggle.

It supports full, 4-bit, 8-bit, and 16-bit training, is highly customizable for tasks like vision, speech, and text, and integrates with popular inference engines such as Ollama and vLLM. Unsloth’s workflow covers model loading, quantizing, training, evaluation, saving, and exporting, with multi-GPU support coming soon.