failed assertion `[MPSNDArray initWithDevice:descriptor:] Error: total bytes of NDArray

[alexcc4]

workflow: powerpaint remove object device: Mac arm64, mps log:

model_type EPS
Using split attention in VAE
Using split attention in VAE
Requested to load SD1ClipModel
Loading 1 new model
Base model type: SD1.5
BrushNet image.shape = torch.Size([1, 1707, 1280, 3]) mask.shape = torch.Size([1, 1707, 1280])
Requested to load AutoencoderKL
Loading 1 new model
BrushNet CL: image_latents shape = torch.Size([1, 4, 213, 160]) interpolated_mask shape = torch.Size([1, 1, 213, 160])
You should add to positive prompt: "empty scene blur"
Requested to load SD1ClipModel
Loading 1 new model
Requested to load BaseModel
Loading 1 new model
  0%|                                                                                          | 0/20 [00:00<?, ?it/s]BrushNet inference: do_classifier_free_guidance is True False
BrushNet inference, step = 0: image batch = 1, got 1 latents, starting from 0
BrushNet inference: sample torch.Size([1, 4, 213, 160]) , CL torch.Size([1, 5, 213, 160])
/AppleInternal/Library/BuildRoots/ce725a5f-c761-11ee-a4ec-b6ef2fd8d87b/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShaders/MPSCore/Types/MPSNDArray.mm:788: failed assertion `[MPSNDArray initWithDevice:descriptor:] Error: total bytes of NDArray > 2**32'
[1]    20151 abort      python main.py
/Users/liangbinsi/.pyenv/versions/3.10.0/lib/python3.10/multiprocessing/resource_tracker.py:224: UserWarning: resource_tracker: There appear to be 1 leaked semaphore objects to clean up at shutdown
  warnings.warn('resource_tracker: There appear to be %d '

[nullquant]

The issue is connected with pytorch mac implementation.

1. What mac OS version do you use?
2. What pytorch version in your ComfyUI installation do you use?

[alexcc4]

• mac OS version: 14.4.1

• pytorch-lightning==1.9.4

[nullquant]

That is quite old pytorch version. It seems that bug was quite common in previous version of mac OS for some sizes of images, but was gone a year ago. Pytorch 1.9.4 is also a version about a year ago old. So, may be it is time to move to 2.x? :)

[alexcc4]

I upgraded my dependencies: pytorch-lightning==2.2.4 Still the same error occurs

[nullquant]

This problem is related to the size of the input image and attention calculation in PowerPaint. I added an option to PowerPaint node. Update nodes and try to set save_memory to max. If the problem remains you should use CutForInpaint or resize image.