← AI Terminology

GGUF

GGUF is a binary file format for quantized ML models (successor spirit to GGML formats) widely used by llama.cpp and the local-LLM ecosystem.

It packages tensors, metadata, and quant types for portable inference.
Why It Matters in AI
Local AI exploded via single-file GGUF models shared on Hugging Face. Understanding GGUF quant levels (Q4_K_M, Q5_K_S, …) is required to run LLMs on laptops and phones via llama.cpp/Ollama.
Key Points
Aspect Description
Pro Portable, simple distribution
Quants Many GGML-family quant schemes in one format
Convert HF → GGUF conversion scripts
Related llama.cpp, AWQ/GPTQ (different stacks)
Contains Weights, tokenizer info, architecture metadata
Ecosystem llama.cpp, Ollama, LM Studio, GPT4All
Simple Analogy
A USB stick format everyone agreed on for pocket-sized brains — plug into many apps without re-encoding chaos.
Common Usage Examples
  • Download model-Q4_K_M.gguf
  • llama.cpp load GGUF
  • Ollama Modelfile FROM gguf
  • Pick quant by VRAM/RAM budget
Summary
In short: GGUF is the portable quantized model file format powering most local LLM tools built around llama.cpp.