Use a Hugging Face model in LU

Model pages on Hugging Face have a "Use this model" button. With Locally Uncensored switched on there, one click opens the model in LU, shows the file, its quant and its size, and downloads it after you confirm. Written for LU 3.0.9.

Switch LU on at Hugging Face

Hugging Face only lists the apps you picked. Sign in, open Settings, Local Apps and tick Locally Uncensored. LU has to be installed on the computer you click from: the installer registers the locallyuncensored:// links with Windows, macOS and Linux (.deb and .rpm). The AppImage registers them itself the first time it starts.

Open a model in LU

1. Open a model page that offers GGUF files, for example a repository whose name ends in -GGUF.
2. Click "Use this model" and pick Locally Uncensored. To get one exact quant, open the file list first and use the button next to the file.
3. Your browser asks whether it may open Locally Uncensored. Allow it. LU starts if it was closed, or comes to the front.
4. LU opens Models, Chat, Discover and shows the repository, the file, the quant and the download size. If the link named no file, pick one from the list; Q4_K_M is preselected when the repository has it.
5. Click Download. The file appears in the download list like any other model, and under Installed when it is done.

If LU is in Cloud mode, it first asks to switch to Local, because the model runs on your computer.

What LU checks first

A link can be fired by any web page, so LU treats it as a question, never as an order:

• The link carries a repository name and a file path, nothing else. The download address is built by LU and always points at huggingface.co. A link with any other parameter, a second repository, a web address or an unusual path is refused.
• Only GGUF model files. Other formats and image parts (mmproj) on their own are refused; an image part that belongs to the model comes along with it.
• The file has to exist in the repository under exactly the name the link gives. LU does not guess a neighbour.
• Nothing is downloaded or started before you click Download, and the button only becomes active a second after the dialog appears.

The model card and the license of the model are the uploader's. Read them before you use the model.

Where the file lands

Exactly where a download from Models, Discover lands: in the LU Engine folder by default, in Ollama when Ollama is your chat backend, in the LM Studio folder when LM Studio is. LU and your models on another drive explains the folders.

When it goes wrong

• Locally Uncensored is missing in the "Use this model" list. Switch it on under Settings, Local Apps, and make sure the repository has GGUF files.
• The browser does nothing. Start LU once after installing, then try again. On Linux with the AppImage, start it once from its file.
• "Link not opened". The link failed one of the checks above. Nothing was downloaded.
• "Hugging Face could not be reached, or the repository is private or gated." Gated models need you to accept their terms on Hugging Face and are not supported by the button yet.