1.1 Update on the gguf dev models?
It looks like all the dev models got updated or re-uploaded today, can somebody tell what does that mean or what happened? On the original LTX2.3 page, the dev model seems untouched.
yes,i wonder too. UD bigger size same as quant stack
Metadata json formatting changed for the non UD. The UD quants are new.
results totally different
Could you clarify? I can double check in the morning.
better，I think.I thought this is better than quantStack, it seems that i am wrong.I should use quantStack at the beginning
There’s are generally larger. We quantize more weights to make the final gguf smaller.
Which ggufs are you comparing?
Metadata json formatting changed for the non UD. The UD quants are new.
Why did it need to be changed? That's the part I didn't understand.
We published new UD quants for dev, distilled, and distilled-1.1. Those are higher quality now. All other quant weights are exactly the same. The only thing that changed was the format of the config field, but otherwise the values are the same. "config" is json and can be written as a string in may ways. The format of the string changed between uploads, hf upload sees a new file hash and decides to reupload it even though the weights are the same. The original gguf's get generated as part of the dynamic quant creation process. Apologies for the confusion!
ud q3ks
ud q3ks
quantstacks q3ks is 14GB our ud q3ks is 11.4GB. It would be more appropriate to use our ud q4ks for comparison.
yours are smaller now ,thanks. almost all my ggufs are yours
by the way,can you upload wan2.2 UD2.0 verson. wan2.2 is better than ltx2.3 @shimmyshimmer . only
QuantStack has wan2.2 gguf, i thinks yours will be better and smaller