Mac OS - hardware denoising fails on 3.1.0rc2

[cafuego]

I just downloaded graxpert 3.1.0rc2 and found that denoising no longer works with ai hardware acceleration enabled - it worked fine on graxpert 3.0.2. Denoising does run OK with the hardware acceleration option disabled.

M3 (Pro) Macbook Pro, OS X 15.5.

The logs are:

2025-05-28 09:16:25,447 MainProcess root INFO     Starting denoising
2025-05-28 09:16:26,203 MainProcess root ERROR    [ONNXRuntimeError] : 1 : FAIL : Failed to create MLModel, error: Failed to build the model execution plan using a model architecture file '/private/var/folders/t8/gd4mgy0562x2ysvt3kpmmccm0000gn/T/onnxruntime-C059B122-C8AF-41B1-9763-13406E0190C5-13774-000016079C0BD503.mlmodelc/model.mil' with error code: -7.
Traceback (most recent call last):
  File "graxpert/application/app.py", line 451, in on_denoise_request
  File "graxpert/denoising.py", line 70, in denoise
  File "onnxruntime/capi/onnxruntime_inference_collection.py", line 465, in __init__
  File "onnxruntime/capi/onnxruntime_inference_collection.py", line 537, in _create_inference_session
onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : Failed to create MLModel, error: Failed to build the model execution plan using a model architecture file '/private/var/folders/t8/gd4mgy0562x2ysvt3kpmmccm0000gn/T/onnxruntime-C059B122-C8AF-41B1-9763-13406E0190C5-13774-000016079C0BD503.mlmodelc/model.mil' with error code: -7.

[dlveach]

Getting same issue on Mac Mini M4 Pro, OSX 15.5.

[geeksville]

I think I have a fix for this. I'll be putting out a test build in the next few days. geeksville@e6c9329?fbclid=IwY2xjawMxR9lleHRuA2FlbQIxMABicmlkETF4dTMzZHRnR0JXcmFZREZ6AR7jBhMqaWakEbTni6YfoUYCJQjWDVF-ZYkH6dSqLOLOrAUzcUrj5_9OQbz04Q_aem_-0E7xUqpMLYVzfhb2uymOA

[Steffenhir]

@geeksville Did you already find a fix for the issue?

[geeksville]

@Steffenhir yes I think possibly, I've done a lot of work to update graxpert for current os-x. Alas - I don't have a mac to test with - I've just been testing in a VM mac. can you try one of my test builds and report back? I'll push out a build sometime this week (I just moved Taiwan for the next few months - so won't have real internet till monday).

[CarCarlo1977]

Hi, can you send me a test build? I have a mac with M2 Pro and I could test it on this model. thanks

[geeksville]

@CarCarlo1977 (and @Steffenhir ) thank you so much. the DMG build is still having issues but the pypi (standard python 'appstore') installer version seems to work well for me on OS-X. Even in my VM mac I get very good acceleration.

For early WIP install instructions see https://github.com/geeksville/GraXpert/blob/main/README.md#cross-platform-installation (no PR sent yet).

But in short for a Mac:

• Install Homebrew
• brew install pipx python-tk@3.13; pipx install --python python3.13 graxpert\[cpuonly\] - Use this if you are on OS-X - it is poorly named, but it will still use the CoreML Apple GPU support
• run "graxpert" from the shell or the Finder

[geeksville]

please let me know what you find (for good or bad ;-) )

[CarCarlo1977]

@geeksville there is some problem with AI model 3.0.2 in denoising. I'm using macos Sequoia with Apple A2 Pro cpu. This is the error message:

2025-10-10 09:45:41,992 MainProcess root INFO Starting denoising
2025-10-10 09:45:42.286 Python[9351:263868] 2025-10-10 09:45:42.286235 [W:onnxruntime:, coreml_execution_provider.cc:113 GetCapability] CoreMLExecutionProvider::GetCapability, number of partitions supported by CoreML: 151 number of nodes in the graph: 2527 number of nodes supported by CoreML: 1729
2025-10-10 09:45:42,749 MainProcess root ERROR [ONNXRuntimeError] : 1 : FAIL : Failed to create MLModel, error: Failed to build the model execution plan using a model architecture file '/private/var/folders/t0/nt7rw34155g1t4b7zdq5d8cm0000gn/T/onnxruntime-715D61B3-1547-4396-8F0F-DD71A87B133D-9351-0000006ACD995897.mlmodelc/model.mil' with error code: -7.
Traceback (most recent call last):
File "/Users/carlo/.local/pipx/venvs/graxpert/lib/python3.13/site-packages/graxpert/application/app.py", line 451, in on_denoise_request
imarray = denoise(
img_array_to_be_processed,
...<4 lines>...
ai_gpu_acceleration=self.prefs.ai_gpu_acceleration,
)
File "/Users/carlo/.local/pipx/venvs/graxpert/lib/python3.13/site-packages/graxpert/denoising.py", line 69, in denoise
session = ort.InferenceSession(ai_path, providers=providers)
File "/Users/carlo/.local/pipx/venvs/graxpert/lib/python3.13/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 485, in init
self._create_inference_session(providers, provider_options, disabled_optimizers)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/carlo/.local/pipx/venvs/graxpert/lib/python3.13/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 584, in _create_inference_session
sess.initialize_session(providers, provider_options, disabled_optimizers)
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : Failed to create MLModel, error: Failed to build the model execution plan using a model architecture file '/private/var/folders/t0/nt7rw34155g1t4b7zdq5d8cm0000gn/T/onnxruntime-715D61B3-1547-4396-8F0F-DD71A87B133D-9351-0000006ACD995897.mlmodelc/model.mil' with error code: -7.

[geeksville]

ok thanks @CarCarlo1977 - I'll try to repro that in my VM!

[geeksville]

okay I took a look. Alas, the onnx.model file for denoise is using some operator that is not supported on the CoreML runtime (https://onnxruntime.ai/docs/execution-providers/CoreML-ExecutionProvider.html#mlprogram). The 'real' fix is to rerun pytorch and regen the model files but with whatever badopcode not used. Alas, I don't think I can sign up for that right now (and I haven't even looked to see if the original inputs to the model are on github somewhere?).

But I'll put a change in tomorrow to catch this and fall back to CPU only as needed when this occurs (it seems like only denoise is using the mystery 'bad' opcode).

[geeksville]

@CarCarlo1977 I think I have a fix in now that at least works well on my macvm. Would you mind trying another test build?

If you run the following your machine should update to the dev7 build and then denoise should work (alas, by falling back to the CPU for just that step).

pipx uninstall graxpert
pipx install --python python3.13 graxpert\[cpuonly\]

thank you for your help!