GLOSSARY
GGUF
The file format local LLMs actually load — quantized weights in one portable blob that llama.cpp, Ollama and Jan speak.
GGUF (GPT-Generated Unified Format) is the packaging that made local LLMs practical: one file holds tokenizer, metadata and quantized weights, so a laptop can load a 7B or 13B model without a Python stack. llama.cpp defined it; Ollama, LM Studio and Jan consume it.
The number after the name — Q4_K_M, Q8_0 — is the quantization recipe, a trade of size and quality. GGUF is not a training format and not Hugging Face's default safetensors; it is the last-mile format for inference on machines you own. If you run models offline, you are almost certainly loading GGUF.