Nerogar

I've updated the example files to reflect the new format. unet attention keys are not saved in diffusers format anymore.

Sorry for all this back and forth, but the comfy dev found another issue with my lora keys. I updated the example file again. The only thing that changes is...

@ricor07 done. Most of the discussions about new features happen on the discord server. So if you have any questions, feel free to join. https://discord.com/invite/KwgcQd5scF

> as some keyboards come with a nice rotary encoder for the media volume like the Logitech G815/G915 which makes changing the music volume a breeze. That's exactly the reason...

I've done some experiments with this myself. Here is my implementation of a stochastic copy function. It expects a target tensor in bf16 format, and a source tensor in fp32...

> main_horse mentioned this once: > > > I can't remember where I brought this up in the past, but you can't use stochastic rounding efficiently on anything other than...

Without seeing your config I can't really help. But to me these looks like VAE artifacts. For some reason the VAE has created a different distribution at the left and...

It's not an issue in this case. Can you check if replacing the line with this removes the warning? `self.orig_median_norm = torch.norm(self.orig_module.weight, dim=1).median().detach().item()`

>Training robustness is improved by instead sampling a new sample from the latent distribution each time we want that image Do you have evidence for that? I've only implemented mean...

`variations_group_in_name` is the correct parameter. Re-caching if augmentations change is currently enabled by passing the concept.image key to this parameter. The cache directory name is a hash of all the...