Unsloth

If you're fine-tuning LLMs and want things to run faster without burning through VRAM, this is the skill you need. Unsloth optimizes training by making models like Llama and Mistral run 2-5x faster with significantly lower memory usage. The skill pulls from official documentation to help you set up training loops, configure quantization, and debug common issues when working with the library. It's passed security audits from Gen Agent Trust Hub, Socket, and Snyk, so it's safe to add to your workflow. Worth having around if you're doing any serious model training work locally or want to experiment without cloud GPU bills.