bmaltais
bmaltais
The significant difference between them is the switch to utf-8 vs utf8 file encoding… not sure why this would be an issue…
I created a test branch with utf8 instead of utf-8... can you give it a try: ``` git fetch origin git checkout utf8 git pull ```
Ok, so at least the obvious changes are not the source… I will look for the next best culprit…
> Not sure what the original script did since it had a default alpha value of 0. Essentially it was doing nothing unless you passed a value... but yes, that...
I tried the script and I am not getting the results expected. Expectations: 1. Running compress create a patch representing the dreamboot model that was applied on top of the...
OK... I found the hardcoded value. Your script appears to do the same as mine when applying a --loss such that you have only part of the model extracted... We...
> The point was more or less -- try to figure out which portions of the finetuning contribute most to the weights and then create a sparse tensor patch based...
Is it possible that sparse tensors are less storage efficient than regular tensors? Saving patches can result in larger files than the original ckpt... This should not be possible... I...
Hello diazona, I understand your concerns regarding the source code. I compiled a fresh copy of the application and was able to sign it with a new Dev App ID...
@albertyw I have taken handyLock down from netputing at the moment given the little dev activity I have done on it... but you are welcome to grab the copy from...