unsloth — Unsloth is a desktop application and framework for running and training large language models (LLMs) and diffusion models locally across Windows, macOS, and Linux.

Unsloth is a desktop application and framework for running and training large language models (LLMs) and diffusion models locally across Windows, macOS, and Linux.

About this project

Unsloth provides a comprehensive ecosystem for local AI, offering three primary interfaces: Unsloth Desktop (native app), Unsloth Studio (web UI), and Unsloth Core (code-based). It supports a wide range of hardware including NVIDIA, AMD, Intel GPUs, CPUs, and the Vulkan backend. Key capabilities include: - Model Execution & Training: Support for running and fine-tuning LLMs, diffusion, embedding, and audio models (e.g., Qwen, DeepSeek, Gemma 4). It features optimized training that claims to be 2x faster with 70% less VRAM usage. - Training Techniques: Supports LoRA, QLoRA, full fine-tuning, pretraining, Reinforcement Learning (RL), GRPO, DPO, and FP8. - Agent Integration: The 'unsloth start' command connects local models to agents like Claude Code, Codex, and others via OpenAI-compatible APIs. - RAG & Search: Includes private web search, deep research, and Retrieval-Augmented Generation (RAG) capabilities. - Deployment: Ability to export models in various formats, including GGUF, NVFP4, and FP8. - Data Management: Tools to build datasets from PDFs, CSVs, and DOCX files using Data Recipes. - Remote Access: Supports serving local models over LAN or securely via Cloudflare HTTPS.