GGUF
Universal binary format for quantized models in llama.cpp
Formats
In plain English
A neat single file package that has all the model's knowledge, tokenizer, and settings packed together.
Technically
GPT-Generated Unified Format. The binary format developed by Georgi Gerganov and the llama.cpp community. Supports memory-mapped (mmap) fast loading, embedded metadata, tokenizer configurations, and modern k-quants (Q4_K_M, Q5_K_M).