What is general.quantization_version in the GGUF spec used for?

[qskousen]

Reading the spec file required keys, one of those is general.quantization_version, but I am not clear exactly what this field is for. According to the docs it is unrelated to the quantization scheme. In what case would this be used?

[CISC]

It's used for changes to the quantization formats themselves, see ggml-org/llama.cpp#1508 when the scaling factor in Q4 and Q8 changed from F32 to F16.