Weights and GGUF Files

What is inside a model file?

Model weights are the numeric parameter values learned during training. GGUF originally meant GPT-Generated Unified Format. It is now the common single-file format used by llama.cpp and related local-model programs. A GGUF file packages weights with the model's architecture, tokenizer (the rules that split text into tokens), and special control tokens, so another computer knows how to load it.

Ollama does not show one friendly .gguf filename. It stores blobs, files named by the hash of their content, and manifests, small records listing the blobs and settings that make one model. A hash is a repeatable digital fingerprint, so changed or damaged content is detectable.

Analogy: A GGUF is like a sealed appliance package containing the machine, its control labels, and setup card. A manifest is the warehouse inventory telling the runtime which sealed boxes form a model.