Pytorch Cuda Out of Memory Error

[Katipo1]

I have a dual-boot laptop with an RTX4060. Running Windows 11 and POP_OS linux.
Under Windows 11, I followed the install instructions and everything worked out of the box.
Under POP_OS linux, I followed the install instructions and I get the following error:

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 30.00 MiB. GPU 0 has a total capacty of 7.63 GiB of which 8.69 MiB is free. Including non-PyTorch memory, this process has 7.61 GiB memory in use. Of the allocated memory 7.34 GiB is allocated by PyTorch, and 167.80 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
Error: Failed to run MagicQuill.

It seems like it thinks the VRAM has already been reserved. I have no idea why.
Any ideas on what I can do?

[julianCast]

Same thing but with a 2070 Super.
It was shown when I unloaded the first image, load another one and tried to add an item (spider).

[Katipo1]

Managed to find a work-around for this:

Edit MagicQuill/comfy/model_management.py

Change:
vram_state = VRAMState.NORMAL_VRAM
set_vram_to = VRAMState.NORMAL_VRAM

to

vram_state = VRAMState.NO_VRAM
set_vram_to = VRAMState.NO_VRAM

Then edit gradio_run.py
Change:
llavaModel = LLaVAModel()

to

llavaModel = None

Magicquill should now start. When it does, click the magic-wand icon to turn off automated suggestions.
Works really well.