unslothai/unsloth
A desktop app that lets you run and train AI models on your own computer, no coding required
Unsloth is a desktop app for running and fine-tuning a wide range of models, including LLMs like Qwen and DeepSeek and image generators like FLUX, directly on your own machine. It works on Windows, macOS, and Linux, and can be started with a single install script or a downloadable installer. It also offers a web UI (Unsloth Studio) and a code-based package (Unsloth Core) for more technical use.
What it does
- Available in three forms: a no-setup desktop app, a web UI (Studio), and a code-based package (Core)
- Supports fine-tuning methods including LoRA, QLoRA, full fine-tuning, reinforcement learning (RL), GRPO, and DPO, claimed to run 2x faster and use 70% less VRAM than standard setups
- Connects local models to AI agent tools like Claude Code and Codex with a single command, letting them act as the agent's brain
- Runs across NVIDIA, AMD, Intel, Apple Silicon, and CPU hardware, with remote access support via Cloudflare tunnels
- Trained models can be exported or deployed in formats like GGUF, NVFP4, and FP8
Why it matters
It lowers the barrier for developers and researchers who want to run and customize large AI models locally instead of relying on cloud APIs. This means working with the latest open-source models without ongoing API costs or sending data outside your own machine.
Terms in this repo
- LoRA/QLoRA · lightweight fine-tuning methods that adjust only part of a model, requiring far less compute
- GGUF · a compressed model file format designed to run on ordinary computers
- GRPO/DPO · reinforcement-learning-based methods for improving a model's responses using feedback or rules
- VRAM · the memory on a graphics card; running out of it can prevent a model from training or running
- MCP (Model Context Protocol) · a standard that lets AI models connect to files, apps, and external tools
Repository description (English)
Local UI to run and train LLMs and diffusion models, including Qwen3.8, Kimi K3, MiniMax-H3, Gemma 4, DeepSeek-V4, FLUX and more.
Open on GitHubTrending repos
- cathrynlavery/diagram-designA skill that makes AI coding tools draw magazine-quality diagrams instead of generic rounded boxes
- public-apis/public-apisA giant crowd-curated directory of free APIs for developers
- semantica-agi/semanticaAn open-source graph infrastructure that lets AI agents show their work, not just their answers
- cactus-compute/needleA 14MB AI model small enough to run tool-calling on a phone or watch, without internet
- macro-inc/macroAn all-in-one workspace where email, chat, docs, tasks, and CRM are cross-linked and share one AI memory
- harry0703/MoneyPrinterTurboAn open-source tool that turns a single topic or keyword into a finished short video, complete with script, footage, subtitles, and music
- basecamp/omarchyA ready-made, opinionated Linux setup built by DHH
- mattpocock/skillsA pack of reusable prompt playbooks that keep AI coding agents like Claude Code from going off the rails
Latest from METAL LAB
- Apple Keeps 2027 Target for Camera AirPods, Qwen3.8-27B Makes 80 Tool Calls Locally
- Binance allows AI agents to trade automatically, limits left up to users
- Meta AI Mac App Now Pulls in Instagram, Facebook, and Ad Accounts
- OpenAI patches Codex file-deletion bug caused by temp-folder cleanup command
- NVIDIA's 300 Verified Skills Lift Correctness by 41 Points