Evaluation fails for YOLOX-X with "could not execute a primitive"

[piotrgrubicki]

Discovered by @vibhubithar

1. Create a project, add several video files, annotate images.
2. start YOLOX-X training.
   Training fails at the last stage. The log reveals:

2026-01-13 19:40:07,508 [ERROR   ] [job.tasks.evaluate_and_infer.evaluate_and_infer:296]  [organization_id=04af51bd-7c4c-4240-a153-c0f0d8d5891e workspace_id=16bdc975-2a09-49f3-ab80-61ceecf4a492]: Error occurred during model evaluation
Traceback (most recent call last):
  File "/interactive_ai/workflows/workflow/job/tasks/evaluate_and_infer/evaluate_and_infer.py", line 279, in evaluate_and_infer
    is_model_accepted, train_inference_subset_id = evaluate(
  File "/interactive_ai/workflows/workflow/job/tasks/evaluate_and_infer/evaluate.py", line 143, in evaluate
    infer_and_evaluate(
  File "/interactive_ai/workflows/workflow/.venv/lib/python3.10/site-packages/jobs_common_extras/evaluation/tasks/infer_and_evaluate.py", line 97, in infer_and_evaluate
    batch_inference.run(use_async=True)
  File "/interactive_ai/workflows/workflow/.venv/lib/python3.10/site-packages/jobs_common_extras/evaluation/services/batch_inference.py", line 140, in run
    self.infer_dataset(batch_dataset, use_async)
  File "/interactive_ai/workflows/workflow/.venv/lib/python3.10/site-packages/jobs_common_extras/evaluation/services/batch_inference.py", line 203, in infer_dataset
    predicted_ann_scene, metadata = self.inferencer.predict(
  File "/interactive_ai/workflows/workflow/.venv/lib/python3.10/site-packages/jobs_common_extras/evaluation/services/inferencer.py", line 201, in predict
    result, metadata = self._predict_raw(numpy_image)
  File "/interactive_ai/workflows/workflow/.venv/lib/python3.10/site-packages/jobs_common_extras/evaluation/services/inferencer.py", line 364, in _predict_raw
    return super()._predict_raw(image)
  File "/interactive_ai/workflows/workflow/.venv/lib/python3.10/site-packages/jobs_common_extras/evaluation/services/inferencer.py", line 172, in _predict_raw
    raw_result = self.model.infer_sync(processed_image)
  File "/interactive_ai/workflows/workflow/.venv/lib/python3.10/site-packages/model_api/models/model.py", line 463, in infer_sync
    return self.inference_adapter.infer_sync(dict_data)
  File "/interactive_ai/workflows/workflow/.venv/lib/python3.10/site-packages/model_api/adapters/openvino_adapter.py", line 313, in infer_sync
    self.infer_request.infer(dict_data)
  File "/interactive_ai/workflows/workflow/.venv/lib/python3.10/site-packages/openvino/runtime/ie_api.py", line 132, in infer
    return OVDict(super().infer(_data_dispatch(
RuntimeError: Exception from src/inference/src/cpp/infer_request.cpp:223:
Exception from src/plugins/intel_cpu/src/graph.cpp:1365:
Node /neck/out_convs.0/conv/Conv/WithoutBiases of type Convolution
could not execute a primitive

I can share the exported project on request. It's quite big. I am not able, though, to reproduce the problem after importing it.