cluder
cluder
tested and didnt see a problem with it
i made a pull request #4514 to fix the checkpoint caching which will conflict with this one. I think the caching is done wrong, as it adds the weight to...
i tried to test it, but in my tests the resulting model did produce the same images as model A. i don't know about the merging variants, but if it...
i just saw that #4151 did add the message "Select a tertiary model (C) or consider using 'Weighted sum'" when trying without the 3rd model, so i think 4151 would...
what cpu/gpu hardware do you have ? (laptop/cpu model etc)
The problem only happens, if the setting **Checkpoints to cache in RAM** is set to > 0.
> Have you tried if #4283 fixes it for you? It is probably the same problem i didn't see the issue / pr before i created my own pr, its...
it could be related, but i didn't test vae loading yet. i might play around with it later, maybe you could also test it with this change. but i guess...
we can change that to not revert the fix, it just means, if it is set to 1, we actually need to keep 2 models in the cache (current +1...
From my testing it fixes now 4305 and 4448 (caching problem) and keeps the 4036 behavior (setting cache to 1, lets you switch between 2 models).