In search of wasted bits: how much information do LLM weights carry?

An empirical investigation into the byte-level entropy of model weights
across numeric formats and model families.