Training classifier on Apple Silicon results in error related to prediction?, but works

[psobolewskiPhD]

This was first posted on image.sc :
https://forum.image.sc/t/napari-object-classification-apoc-failing-on-m2-mac/85733/1
I've hit the same thing: https://forum.image.sc/t/napari-object-classification-apoc-failing-on-m2-mac/85733/7?u=psobolewskiphd
Opening this issue as maybe this is a better place for discussion a workaround or fix.

Anyhow, I can use napari-apoc for segmentation on my M1 Max with no issue. But when training a classifier I get an error:

RuntimeError: clBuildProgram failed: BUILD_PROGRAM_FAILURE - clBuildProgram failed: BUILD_PROGRAM_FAILURE - clBuildProgram failed: BUILD_PROGRAM_FAILURE

Build on <pyopencl.Device 'Apple M1 Max' on 'Apple' at 0x1027f00>:

Compiler encountered an internal error

Despite the error the ObjectClassifier.cl is generated and it looks fine.
If I then use the napari-apoc > apply object classification widget and pass the path of the classifier, everything works just fine.
...but not if I use the tab in the train widget oddly enough.
If I comment out this line:

napari-accelerated-pixel-and-object-classification/napari_accelerated_pixel_and_object_classification/_object_classification_widget.py

Line 379 in a094823

result = clf.predict(labels, image)

Then I don’t get the CL error so training works-- I just get the error related to the return type, but that's obviously easily fixable--so I guess the .predict right after .train seems to trigger the issue?

My env:
Python: 3.10.14 (arm64)
napari: 0.4.19.post1
napari-accelerated-pixel-and-object-classification: 0.14.1