Unsloth — дообучение LLM на своём железе

★ 67.7K

Unsloth is an open-source tool for fine-tuning language models to your own task on your own hardware. Reach for it when a general model answers the wrong way: it does not know your domain, will not hold the required style and format, and stumbles over your terminology. Fine-tuning on your own examples fixes that more reliably than ever-longer prompts. Its main value is efficiency: training runs markedly faster and needs less video memory, so a job that previously demanded expensive rented compute fits on a single consumer GPU. It supports current models — Qwen3, DeepSeek-V4, Gemma 4, Kimi K3 — as well as diffusion, embedding and audio models. It comes as a desktop app for Windows, macOS and Linux, a web UI and a library, and the finished model can be run locally straight away.