Separate file for the tokenizer files and config json for the gguf files

[Nexesenex]

Would it be possible to have an option to quantize a model, either from its HF weight, either from an already existing gguf, with the config files of the model put in a separate (from the tensor weights) file, which would be easily editable? (like a -00000.gguf file, for example)

It could even be specified aside an existing .gguf or even .ggml to supersede its parameters, and easily bring old quants to new format features, not disturbing the existing pool of quantized models but offering a way to bring them up to date and correct what's eventually wrong with their config.

That way, the mistakes in the config.json, or in the tokenizer files and chat templates could be easily corrected without remaking a whole conversion of quantization, which is a real hassle for big models, and a waste of time, compute, SSD writings and uploading bandwidth in any case.