Loading hypernetworks created using the hypernetwork_training branch fails

[bmaltais]

@AUTOMATIC1111 I created an hypernetwork using the training branch and it is failing to load with the following error in the master branch... but load fine in the training branch:

I am trying to figure out how to tame this beast... I have had some success but much more trial is going to be required to figure out how to get some expected results. I have gotten some interesting ones but not quite what I was expecting.

[cmp-nct]

Are you trying to load an embedding as hypernetwork model ?
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Textual-Inversion

[bmaltais]

No, I trained an actual hypernetwork using the hypernetwork-training branch code: https://github.com/AUTOMATIC1111/stable-diffusion-webui/tree/hypernetwork-training

I am trying to use it in the latest master code base... but it fail to load.

It does load just fine under the hypernetwork-training branch ui...

[Konvoname]

Can you point me out how to train with that, I'm very new to this.

[bmaltais]

I suggest to 1st train a textual inversion. Once you have that working you could try training an hypernetwork... steps are pretty similar... but it is bleeding edge and not working super well at the moment. I am still trying to figure it out myself.

[Konvoname]

I see. thanks

[bmaltais]

As of Oct 11 2022 commits this appear to have been fixed. Trained HN from the training branch are now loading. Now the challenge is... making a good HN. So far I only succeeded to make one... all others have diverged toward NaN and awful results. Will try to replicate the conditionf for the one training that worked and see if I can reproduce with a different dataset.