[Bug]: Flux DoRas do not work with full layer preset
What happened?
I've described it in this issue for ComfyUI (https://github.com/comfyanonymous/ComfyUI/issues/4695#issuecomment-2326417622), but it seems as though DoRas do not work properly in Comfy and they've implied it might have to do with the format of the file.
What did you expect would happen?
I would expect there to be no pink noise introduced when creating and using a DoRa instead of a LoRa. LoRas look fine.
Relevant log output
No response
Output of pip freeze
No response
It seems to be that a DoRa made using only the attn preset works, but using the full layers preset does not work, causing the pink artifacts.
I'd like to help fix this issue, is it something to be changed on Comfy or on OneTrainer? No idea where to start lol
According to comfy this should be fixed now. Can you confirm, or do you still see the problem?
I have tested a full Flux Dora in Comfy and it works now. It still creates the pink coloration in Forge.
Great, thanks. That confirms it's not a OneTrainer issue