Unsloth

Unsloth is an open-source fine-tuning framework that dramatically accelerates LLM training while reducing memory usage. It supports popular models including Llama, Mistral, Phi, Gemma, and Qwen, with full compatibility with Hugging Face's transformers and TRL libraries. Unsloth achieves its speed gains through custom Triton kernels and intelligent memory optimization, requiring no changes to existing training code. Free and Pro tiers are available, with enterprise support for multi-node deployments.