How to debug failed inference on GPU?

[mattiasmar]

I've serialized my model using the OpenVINO API (results attached). But when calling inference (on an Alder Lake GPU) I get the below exception. How can I debug it? On CPU the same code works fine.

...
  File "/root/ws/lightglue-openvino/ALIKE/alike_openvino.py", line 153, in extract_dense_map
    self.infer_request.wait()
RuntimeError: Exception from src/inference/src/infer_request.cpp:249:
Exception from src/inference/src/dev/converter_utils.cpp:706:
Exception from core/infer_request.hpp:54:
Caught exception: clWaitForEvents

Attached zip contains serialized IR xml and bin.
serialized.zip

[mattiasmar]

I suspected the call to torch.sort to be most risky call in this model. Having canceled out that method the inference call on GPU executed too (just not doing what is expected of the model). Will open separate issue for that.