Besides the "culture"/licensing/FOSS issue already mentioned, I just wanted to be able to reuse model weights across various applications, but Ollama decided to ship their own way of storing things on disk + with their own registry. I'm guessing it's because they want to eventually be able to monetize this somehow, maybe "private" weights hosted on their registry or something. I don't get why they thought splitting up files into "blobs" made sense for LLM weights, seems they wanted to reduce duplication (ala Docker) but instead it just makes things more complicated for no gains.

End result for users like me though, is to have to duplicate +30GB large files just because I wanted to use the weights in Ollama and the rest of the ecosystem. So instead I use everything else that largely just works the same way, and not Ollama.