I will not call a pre-trained #LLM / #AI model that you can run on your own machine an "open source" model. It's simply the wrong term.
You effectively download a binary blob. You don't get any of the sources. You can't see how the sausage was made. You can't modify it (yes, I know about LoRA, not what I mean). You can't meaningfully learn how it was made. "open source" as a term sets a pretty low bar, and these models don't even clear that.
I will die on this hill.