Skip to content

GGUF

Short answer

GGUF is the single-file format that llama.cpp and the tools built on it use to store a language model's weights, its tokenizer and its metadata together.

Before GGUF, running a model locally meant collecting a config file, a tokenizer, and weights split across several shards, then hoping the loader agreed with you about all three. GGUF puts everything in one file, so moving a model between machines is copying a file.

It matters in Nightjar because the model picker reads GGUF metadata directly: the file says how many parameters it has, how it was quantised and what context length it was built for, which is how the browser can tell you whether a model will fit before it downloads it. Any GGUF file works, including ones you built yourself.

Related terms

Other ways people ask this

  • what is gguf

Related