Unsloth — быстрое дообучение LLM с LoRA

★ 7.7 · ai-tooling

unsloth is a Claude Code skill that provides expert guidance for fast LLM fine-tuning with Unsloth, delivering 2–5x faster training and 50–80% memory reduction through optimized kernels. It covers LoRA and QLoRA workflows on models such as Llama, Mistral, Gemma, and Qwen, relying on dependencies including unsloth, torch, transformers, trl, datasets, and peft. The skill ships with structured documentation in a references/ directory — including llms-txt.md extracted from official sources — with code examples, detailed explanations, and a navigable table of contents. It's aimed at ML practitioners who are implementing memory-efficient fine-tuning pipelines, debugging Unsloth code, or exploring its APIs and best practices.