GGUF

The single-file model format used by llama.cpp, carrying weights plus all metadata needed to run them.

GGUF replaced GGML as llama.cpp's container. A file begins with a header: a magic number, version, tensor count, a key-value metadata block, then one record per tensor giving its name, dimensions and quantization type.

Because the header is self-delimiting and sits at the start of the file, it can be read with a single HTTP range request without downloading the weights — which is how this site inspects quantizations it has never downloaded. Headers run from around 54 KB to 13 MB, the large ones because tokenizer vocabularies are stored inline.

See it in the data