Ross Wightman

@clementpoiret so concern with this, it changes a kernen that's likely a fused addmm into two separate kernels which isn't great. Aren't there more advanced quanitization options these days that...

@clementpoiret going to implement as per that linked PR above, uses a attr to determine path to use, but don't think I want to punch that all the way through...

@gau-nernst yeah, if it's just standard clip vit arch but w/ some custom widths/depths/heads that approach will work fine. If there's a PR with these added that points to url...

@raulcarlomagno I like this model quite a bit, neat ideas, but they've marked both the code and weights as non-commercial. I can deal with the weights, I treat them with...

@chayryali that's great! I think it shouldn't be too had to get it in, the style is pretty much in line with timm already ... just a number of timm...

@chayryali so, been juggling just a few things lately, but do have this model working locally in timm. I've been trying to add support for changing resolution though, either on...

If I hold the patch stride vs img size ratio constant it appears to work, but that constrains the possibilities significantly...

@chayryali nice, I hadn't seen that paper will have a read. I was working through an idea to add different ROPE pos to the window'd and global stages to see...

could make that view less redundant there, but just fiddling :)

@chayryali read the paper, makes sense. Is the updated code/models coming anytime soon? In the comparison tables you have numbers for fine-tune at higher res. Definitely want to see those...