@PokeRageUS I’m actually was making a mobile/web app that allows you to create virtual binders reflecting actual real binders… wanna me send an invite once beta access opens?
@norpadon@afrogmaen You go into wrong direction with him there, it was just enough to say that total size of input tokens is much larger then weights size, then take gpt-oss-20b or qwen-3.5-9b as examples with MUCH smaller weight size than input to fully prove no way for train data to be output