Forkoz

and now XL also downloads a 10gb ip_pytorch_model.bin to an unknown folder. If I let it finish it will probably load but that's a lot to download through unstable PT...

The 10gb model is a CLIP model. The errors that cause it come from repositories/stable-diffusion. I will try to disable the clip extension and see if the error goes away.

I found the bug! https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/master/modules/sd_disable_initialization.py ``` def CLIPTextModel_from_pretrained(pretrained_model_name_or_path, *model_args, **kwargs): res = self.CLIPTextModel_from_pretrained(None, *model_args, config=pretrained_model_name_or_path, state_dict={}, **kwargs) res.name_or_path = pretrained_model_name_or_path return res ``` It is supposed to disable loading the...

Be me.. using multiple "custom" openAI servers from providers and my own open source. Switch connection profiles.. huh.. the key doesn't work. Did I get banned? Literally everything is openAI...

As an idea. How about adding proxy support to generic openAI api.. that way the creds can just be stored how the proxy passwords are but we still get additional...

Same issue for 7.0. Browsers open but never load web pages. Android system has stopped. Zygisk version doesn't have this problem but fails when adding experimental hooks for theming. When...

I like this idea because my dud gens fill up my outputs folder and have to get deleted every once in a while. edit: does PreviewImage work? I heard that...

>Be me, see gemma come out. _People say it's coal._ >Screw it, I'll try it. >wot backend? no exllama, llama.cpp has no pictchas, hey what about vllm? >It supports image...

The PR is in the pipeline. Just merge it and compile. I didn't see anyone saying it doesn't work. Wonder if the visual portion is accounted for.

Not sure that would work here since it's per request.