👋Welcome
New to Unsloth?
Last updated
Was this helpful?
New to Unsloth?
Last updated
Was this helpful?
Unsloth makes finetuning large language models like Llama-3, Mistral, Phi-4 and Gemma 2x faster, use 70% less memory, and with no degradation in accuracy! Our docs will guide you through training your own custom model. It covers the essentials of installing & updating Unsloth, creating datasets, running & deploying your model.
Unsloth makes it super easy for you to train models like Llama 3 locally or on platforms such as Google Colab and Kaggle. We streamline the entire training workflow, including model loading, quantizing, training, evaluating, running, saving, exporting, and integrations with inference engines like Ollama, llama.cpp, and vLLM.
The key reason to use Unsloth is because of our deep involvement in fixing critical bugs across major models. We’ve collaborated directly with teams behind Mistral (Devstral), Qwen3, Meta (Llama 4), Google (Gemma 1–3), and Microsoft (Phi-3/4), often contributing essential fixes that improved accuracy, stability, and prompt handling. Our bug fixes include a +2% accuracy boost for Llama 4, critical RoPE and chat template corrections, and the identification of uninitialized weights in Gemma 3. And we also actively maintain and update models whenever they have updated/fixed chat templates/tokenizers etc.
Unsloth is highly customizable as we allow you to alter things like chat templates or dataset formatting. We also support and have pre-built notebooks for vision, text-to-speech (TTS), BERT, reinforcement learning (RL) and more! We also support all training methods and all transformer-based models.
Install locally with pip (recommended) for Linux devices:
For Windows install instructions, see here.
Fine-tuning an LLM customizes its behavior, enhances domain knowledge, and optimizes performance for specific tasks. Finetuning is the process of updating the actual "brains" of the language model through some process called back-propagation.
By fine-tuning a pre-trained model (e.g. Llama-3.1-8B) on a specialized dataset, you can:
Update Knowledge: Introduce new domain-specific information.
Customize Behavior: Adjust the model’s tone, personality, or response style.
Optimize for Tasks: Improve accuracy and relevance for specific use cases.
Example usecases:
Train LLM to predict if a headline impacts a company positively or negatively.
Use historical customer interactions for more accurate and custom responses.
Fine-tune LLM on legal texts for contract analysis, case law research, and compliance.
You can think of a fine-tuned model as a specialized agent designed to do specific tasks more effectively and efficiently. Fine-tuning can replicate all of RAG's capabilities, but not vice versa.
Unsloth can be installed locally via Linux, Windows, Kaggle, or another GPU service like Google Colab. Most use Unsloth through the interface Google Colab which provides a free GPU to train with.
