About ggufmodels.com

ggufmodels.com is an independent catalog of GGUF models. It exists to answer one question before you download a multi-gigabyte file: will this run on my hardware?

What the site does

It lists GGUF repositories from Hugging Face (215,124 models at the moment), reads the details of each file, and shows them in the same layout on every page. You can filter by family, size, quant, license and the amount of VRAM you have, compare quants, and see which ones fit.

Where the data comes from

The GGUF file header. For each repository we read the first part of one GGUF file with a range request, without downloading the model. The header gives the architecture, the number of layers, the trained context length and the attention layout.

The Hugging Face API. The file list and sizes, the license, the base model, download counts and the parameter count Hugging Face displays on the repository page come from Hugging Face.

The model cards. We read the repository’s own README and the base model’s README as background for the written descriptions.

How the memory estimates work

For every file we add the file size, an fp16 KV cache for an 8K context window, and 0.5 GB of runtime overhead. The KV cache is computed from the model’s own header, using the layer count, KV head count and head size. The model pages let you change the context length and your VRAM. These are estimates and not measurements: real usage depends on your runtime, your settings and your hardware.

How often it is updated

New uploads are checked every half hour, and the full list is refreshed daily, so file sizes, download counts and new quants stay current.

AI-assisted text

The short descriptions, pros and cons and answers on model pages are written by an AI model from the data shown on the page, then checked by automated rules that reject invented benchmark scores and unsupported claims. They can still contain mistakes, so treat the numbers in the tables as the reference. Wherever a number matters, the table is the source.

What this site is not

It does not host or distribute model files: every download button opens the file on Hugging Face, where the license and terms apply. It is not affiliated with Hugging Face, llama.cpp or any model publisher. Benchmark results are not shown yet, and we will mark them clearly once they are.

Found a mistake?

If a number looks wrong, compare it with the repository on Hugging Face, which is the source. Our data is read automatically, so a repository with unusual file names or an unusual layout can occasionally be misread.