Unsloth AI is an open-source toolkit for faster training and fine-tuning of large language models.
Faster fine-tuning and RL for LLMs
Unsloth AI focuses on compute optimization to reduce training time and memory usage. It supports modern training setups and model families, including:
- gpt-oss
- Llama 4
- DeepSeek-R1
- Qwen3
- TTS models
- BERT-style models
Key capabilities include:
- Reinforcement learning support (RL, GRPO)
- Long-context training up to hundreds of thousands of tokens
- Support for modern numeric formats such as FP8
Built for developers and teams
Unsloth AI is aimed at engineers who want to quickly adapt a model for a specific task, such as a chatbot, assistant, or internal tool. To help with deployment and collaboration, it provides:
- Docker images
- Documentation and a blog with updates
- An active Discord community
For experiments and production
The project can be used for research experiments as well as production workloads. Open-source code makes it easier to audit, customize, and adapt the training workflow to your company’s infrastructure.
