Unsloth Framework

Unsloth is the Python framework for Large Language Model fine-tuning: achieve 2x to 5x faster training with up to 80% less VRAM.

Unsloth is your accelerator for Large Language Model fine-tuning, built on a foundation of custom, handwritten GPU kernels in Triton. This optimization delivers serious performance: expect training speeds up to 5x faster than standard Hugging Face methods and a VRAM reduction of up to 80%. The framework supports popular open-source models like Llama 3.1, Mistral, and Gemma, making advanced LLM adaptation accessible even on limited hardware, such as a single consumer-grade GPU. It integrates seamlessly with the Hugging Face ecosystem, supporting QLoRA, full-finetuning, and one-line conversion to formats like GGUF for deployment.

https://unsloth.ai/