MarkovInequality

@d8ahazard I have this problem when training negative prompts as well on 10GB of VRAM. Turning off "train with reconstruction" stops the OOM errors, but I can't get training to...

The refresh button spans across a large chunk of the quick settings bar. Pic below.  You can fix this by changing the following in style.css ``` #refresh_sd_model_checkpoint, #refresh_sd_hypernetwork, #refresh_train_hypernetwork_name,...

I just tested the latest build without transformers using the doggettx, and invokeAI optimizations and I can't seem to reproduce the problem. I also tested training with VAE selected as...

I think I found the problem, moving VAE to the CPU doesn't play nicely with "Show image creation progress every N sampling steps". Set it 0 to disable. Can you...

Can you please try to train a hypernetwork with "Show image creation progress every N sampling steps" set to 0, with my optimization with the "Move VAE and CLIP to...

Just checked the code for HNs, and if it's happening for TI, it should theoretically happen for HN as well. After running hypernetwork training to double check, I see that...

The reason is because both TI and HN attempts to move the VAE to the CPU to save VRAM. However, in a parallel thread, the progress bar is calling sd_samplers.sample_to_image(self.current_latent)....