# GGUF tools

A C library and command-line tool for inspecting and manipulating GGUF model files, showing metadata and tensors, comparing two models for similarity, and inspecting raw tensor weights.

GGUF tools is a library and command-line program for working with GGUF files, the format used by llama.cpp and other local inference tools. Its show subcommand displays a file's key-value metadata, including arrays, and detailed information about every tensor, with offsets given relative to the start of the file rather than the data section.

Its compare subcommand checks whether two GGUF models are related, for instance whether one is a fine-tune of another, by computing the average weight difference for each matching tensor by name and parameter count, useful for seeing how much a model diverged and which layers were frozen during fine-tuning. Its inspect-tensor subcommand prints some or all of a named tensor's raw weight values, useful for checking quantization behavior or fingerprinting a model. An experimental subcommand extracts a smaller model from a mixture-of-experts file for experimentation, though the author notes the result does not perform useful work on its own.

It is written in C and licensed under MIT.

Website: https://github.com/antirez/gguf-tools