[bug]: InvokeAI crashes on Mac when using Lora

[slavikshen]

Is there an existing issue for this?

• [X] I have searched the existing issues

OS

macOS

GPU

mps

VRAM

No response

What version did you experience this issue on?

2.3.4

What happened?

The python process crashes when using lora.

/AppleInternal/Library/BuildRoots/f0468ab4-4115-11ed-8edc-7ef33c48bc85/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShaders/MPSCore/Types/MPSNDArray.mm:705: failed assertion `[MPSNDArray initWithDevice:descriptor:] Error: product of dimension sizes > 2**31'
./start.sh: line 27:  6399 Abort trap: 6           invokeai --web $@
slavik@SlaviksacStudio InvokeAI % /opt/homebrew/Cellar/python@3.10/3.10.10_1/Frameworks/Python.framework/Versions/3.10/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 '

The prompt I'm using

withLora(Shenhe_Hard,0.75) masterpiece (ultra-detailed) illustration (mature_female)++ 1girl stands under sakura tree blossom,(school_uniform,bustier).blend(1,1)  (mini frilled skirt) underboob++++ (detailed-embroidery) (full body) (detailed face) (\:3)++++ (seductive smile)   (hair_ornament) (detailed eyes) (falling_petals)++:3 

(detailed background) (insanely detailed and intricate) (hyper maximalist) (light from above):1 

Screenshots

No response

Additional context

No response

Contact Details

No response

[slavikshen]

Updates The txt2img can pass when the size is under 512x768. However, the result is far from correct

This png contains parameters.

[lstein]

Duplicate of #3157

[slavikshen]

Duplicate of #3157

That's a different one. Actually I find the InvokeAI crashes without withLora as well. The only requirement to reproduce this issue is to set the output size larger than 512x768

[lstein]

I can't reproduce this one. It happens with short prompts as well?

[slavikshen]

I'm not sure if the length of the prompt affects the result. I'm using a MacStudio M1Max 64G RAM The model is https://civitai.com/models/9942/abyssorangemix3-aom3 The VAE is kl-f8-anime2.ckpt The params in the log file is

[Generated] "/Volumes/Work/AIArt/007480.9689e9e3.1212659739.png": ""SamDoesArt-8000":1

masterpiece (ultra-detailed) illustration (mature_female)++ 1girl stands under sakura tree blossom,(school_uniform,bustier).blend(1,1)  (mini frilled skirt) underboob++++ (detailed-embroidery) (full body) (detailed face) (\:3)++++ (seductive smile)   (hair_ornament) (detailed eyes) (falling_petals)++:3 

(detailed background) (insanely detailed and intricate) (hyper maximalist) (light from above):1 [(flat_shading) loli+++ child+++ (low quality)+++ (worst quality)+++ (out of frame)++ (band hand) (fused fingers) "bad_prompt_version2"- "NG_DeepNegative_V1_T75"++ "bad-hands-5"++]" -s 30 -S 1212659739 -W 512 -H 512 -C 7 -A k_dpm_2_a --hires -f 0.8

The same setup works in the previous release.

[slavikshen]

@lstein https://github.com/invoke-ai/InvokeAI/issues/2444

This ticket looks very close to the bug.

[slavikshen]

The crash is gone after the latest MacOS update. However, the txt2img result is significantly worse than the previous release.