HN user
I also asked it to explain what's funny in some newspaper comic strips in Finnish. It misunderstands some words and makes up nutty explanations, but most phrases still get translated correctly and its explanations do fit the drawn scenes once you factor in those misunderstandings. For such a small model that seemed impressive.
I found a losslessly compressed version: https://github.com/LeanModels/Bagel-DFloat11
It works following readme instructions at least on Ubuntu, on my RTX 3090 GPU with 24 gigs of memory, just barely. Have to close most other windows and lower screen resolution to be able to load the model. Then it generates or edits images in 2-3 minutes. I only have this one GPU and am using Chrome to use the browser interface on the same machine.
The original release won't run on this hardware, but the compressed one is supposed to give identical results.
It's a blog post, but you can try / edit the demo. Appropriate for Show HN?
I'm actually working on better tooling for OpenStreetMap, but taking the longer road (to put it mildly) redesigning and documenting everything, starting from pushing pixels on screen.
I know it will take years. Hopefully others can already put some of the techniques to good use in the meantime.
PhosphorJS seems truly amazing. Dragging and resizing things feels like using native controls. The example has a polished look and everything is written in well-commented and easily understandable TypeScript without any nonsense external dependencies like jQuery.