unslothai/notebooks

What it solves

This repository provides a collection of ready-to-use Google Colab notebooks designed to simplify the fine-tuning, training, and inference of large language models (LLMs) and multimodal models. It removes the technical friction of setting up complex training environments for various open-source architectures.

How it works

Each notebook is a self-contained environment that guides users through data preparation, model training, and inference. The notebooks leverage the Unsloth framework to optimize memory usage and speed during the fine-tuning process. They cover a wide range of techniques, including standard fine-tuning, Reinforcement Learning (GRPO, PPO, DPO, ORPO), and specialized tasks like vision, audio, and embedding generation.

Who it’s for

This is for developers, researchers, and AI hobbyists who want to fine-tune or experiment with state-of-the-art open-source models (such as Llama, Qwen, Gemma, and Mistral) without managing complex infrastructure.

Highlights

• Extensive library of notebooks for popular model architectures including Llama 3.2, Qwen 3, and Gemma 4.
• Support for advanced training methods like Group Relative Policy Optimization (GRPO) and Reinforcement Learning for reasoning tasks.
• Specialized workflows for multimodal capabilities, including vision-language models and text-to-speech (TTS) fine-tuning.
• One-click deployment to Google Colab for immediate experimentation with data prep and model training.