Added TeaCache Support and node

[ctf05]

It works. I did very little testing. You get a speed-up with a loss of quality but you can get a ~1.4x speedup with little loss.

[DiffuzionDreamer]

I'm getting an error when running it in conjunction with Enhance-A-Video:

But... I'm also able to run them together after running with just TeaCache and then again with both EAV and TC. Love the speed increase with TC but I wouldn't know where to begin to fix the bug.

[morganavr]

Sometimes after I reload workflow from .json I got this error.
I use TeaCache with Enhance-A-Video.

[DarioFT]

@DiffuzionDreamer I've created a small PR that addresses this on @ctf05 's fork.

[ctf05]

@DiffuzionDreamer I've created a small PR that addresses this on @ctf05 's fork.

I just saw this. Thanks for the fix. It fixed some problems I had. I think TeaCache is bug free now. (known bugs)

[DiffuzionDreamer]

Tested the fix and and it is working on my system, and seems to resolve the issue reported by @morganavr as well which had occured for me when aiming for larger/longer outputs.

[morganavr]

TeaCache does not work with BlockSwap node - it throws an exception

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument mat1 in method wrapper_CUDA_addmm)