ah, 3.5 large is a new model, lol
ah, 3.5 large is a new model, lol
1
1API I2V was impressive
Fthere is a client requested
Fyes but code missing :d
Foh 9 min ago published 
Fworking on installer
Fdmn still no torchvision for rtx 5000 series
1I didn't expect the 14b to be released too
Fthey released both amazing
Fmaking the gradio app installer
1
1I'd be interested in helping to test
Ftrying to make installer
Fdmn it uses flash attention forcedly
Fi installed weird testing again
Fok downloading models fingers crossed
Fgenerating video
F
1looks promising
Fafter prompt enhance it doesnt deload model lol
F
Fvae decoding taking forever
Fexcellent
F1ohh damn, that's only the 1.3B model
Fyes
Fdownloading 14b model for image to video
Fbut i dont know if will fit vram
Fprobably not
Fand i can use 5090 yet 
Fbut i am gonna make installer for runpod and massed compute too
1I think it's 50+Gb unquantised.
Fdmn probably wont fit
Fi dont see quant option on the repo
1but that is FP32. FP16 should be managable
Fdid you check their pipeline?
Fwhere it is cast :d
1No, haven't had a chance to take a look, stuck in work meetings atm
1I've just sold my 4090 too so will need to wait for torch to work with 5090
Fi kept 3090 just for this casue
1I've still got a 3090 but no space in the case. 5090 is taking too many slots 
Fthis is how i use
Fimage to video failed 13b
1reminds me of my mining days
C1that 5090 performance
CAdded L40
CWe had one server causing this to happen frequently over the last week, I suspect you ran into it. We stopped new rentals on it, and refunded a ton of VM rental attempts that went to it.
CWe are working on some improvements around this. I know connecting to the VM before its fully ready can cause a problem, but if you continue to run into this and have error messages we can look at we'd appreciate if you send them them to us so we can be certain its something on our radar.
