Someone just created a pull request in llama.cpp for Qwen2VL support!

Replying to the deleted comment on llama.cpp devs rejecting PRs that people don't commit to maintaining: That is their prerogative, but I think it's tricky to strike a good balance. On one hand you don't wanna have unmaintainable code, on the other hand having buggy support for a cool arch is better than *not* having support at all. After all this is FOSS and **other** developers can always collaborate to fix issues - but they can't if you don't even entertain them in the first place. I think it's slightly unfair to pin the expectation of eternal maintenance on a prospective contributor forever for free. Because while the devs work on what they like to work on, ultimately I think most people have a goal of running exciting new models instead of witnessing yet another backend refactor or another 2% cuda speedup (doubling the library size)