We read only the safetensors headers of thousands of Hugging Face repositories (names, shapes and storage of every tensor, never the weights) and recovered each model's architecture from them. Here is what the models people actually use are made of, how that changed year by year, and where the metadata around them is wrong.
Type any model from the census. See what it is made of, where it sits among popular models, and which devices it fits on.
Models submitted by their authors and users. Each is read from its headers and placed by the memory it needs at 4-bit with an 8k context, with five checks on its metadata. Submit yours and get a badge for the model card.
Popular repos: the most downloaded, the most downloaded original releases, and recent releases from leading providers. Test fixtures (tiny random-weight repos) are left out.
The most downloaded readable models in the census. Click a name to open its full anatomy.
Params: logical parameters (quantized storage unpacked). Hub: ✓ when the Hub's own count agrees within 0.1%, otherwise our count ÷ the Hub's. KV/token: bytes of KV cache per token at 16 bit.
Every repo: status, popularity, the architecture recovered from the weights, the values its config.json declares, the Hub's own parameter count and the raw manifest, so the results can be re-scored. Code and the frozen recovery rules are public.