Lincoln Stein
Lincoln Stein
This is ready for review.
The problem is that an embedding trained on a stable diffusion 1.5 model won't load onto stable diffusion 2.1. I think it's because the image dimensions are incompatible. I'll see...
This might not be necessary. The new model manager that I'm working on uses the context manager to free GPU memory when a model is no longer in context: ```...
Wouldn't be hard to do this on a single node. In a distributed environment you'd need to propagate a message throughout the whole system. Is there a framework for doing...
Currently the way the `latents` node seems to work is that it retrieves the model and then configures seamless for the unet and vae just before using it for generation....
#6132 does seem to fix the issue. I've done a little testing with this PR, and indeed the VRAM used by the upscaler model is now released either when upscaling...
I have added a migration script that tidies up the `models/core` directory and removes unused models such as GFPGAN. In addition, I have renamed `models/.cache` to `models/.convert_cache` to distinguish it...
@psychedelicious @RyanJDick I think I've responded to all comments and suggestions. Thanks for the reviews!
> I'll do a full review later, we still have too much coupling between the context/model manager and some of the utility classes. > > I've just pushed changes that...
> > There just doesn't seem to be a good way for utility classes to access the model manager service in the running app without going through the context. One...