Unsloth

If you're fine-tuning LLMs and tired of waiting hours or running out of memory, this skill gives you expert guidance on Unsloth's optimization techniques. It covers LoRA and QLoRA implementations with 2-5x faster training speeds and 50-80% memory reduction across popular models like Llama, Mistral, Gemma, and Qwen. The documentation here is pulled from official sources, so you get the real patterns and examples without hunting through GitHub repos. Good for both getting started quickly and debugging when your training runs go sideways. Honestly, if you're doing any serious fine-tuning work, knowing these optimization tricks will save you real time and compute costs.