[bug]: ControlNet `zoe` processor does not free VRAM

[psychedelicious]

Is there an existing issue for this?

• [X] I have searched the existing issues

OS

Linux

GPU

cuda

VRAM

No response

What version did you experience this issue on?

Main

What happened?

The zoe processor does not free VRAM when it finishes processing. Each time it runs, it uses more VRAM:

(Yellow line is VRAM, the 3 steps are running the ZoeImageProcessor node once)

It appears to be the only control net processor that uses any VRAM - I wonder if the other processors would have the same issue if they used GPU.

I tried a couple things to unload the model, but it doesn't work:

torch.cuda.empty_cache()
torch.clear_autocast_cache()
gc.collect()

I suppose the issue is the repeated loading of the model with from_pretrained, which does not check if the model is already loaded.

Screenshots

No response

Additional context

No response

Contact Details

No response