Unsloth

Unsloth is a desktop app to run and train AI models on your own hardware. It works on Windows, Linux, WSL and macOS, with multi-GPU, NVIDIA, AMD, Intel, CPU and Vulkan support, and installs through a script, Docker, Homebrew or Winget.

Unsloth fine-tunes LLMs, diffusion, TTS and embedding models about 2x faster with 70% less VRAM, according to the project, and supports LoRA, QLoRA, full fine-tuning, pretraining and reinforcement learning. It also runs models locally with web search and RAG, serves them through an OpenAI-compatible API, connects them to tools like Claude Code and Codex, and exports to GGUF.

Unsloth made fine-tuning open models practical on a single GPU, and now wraps running and training in one desktop app. It is Apache-licensed.

Stars

0

Forks

0

Commits

0

Repository

unslothai/unsloth

• LM StudioDesktop app for running local language models.
• OpenAI APIOpenAI's hosted API, with fine-tuning.
• Together AIHosted inference and fine-tuning for open models.