fix: [GPU] ProgramBuilder build failed!

[intel-ohara]

Summary

I installed physicalai in uv virtual python environment and typed commands below.

from physicalai.inference import InferenceModel
policy = InferenceModel("./SmolVLA/exports/openvino", device="GPU")

Then the following error happened.

Traceback (most recent call last):
File "", line 1, in
File "/home/intel/openvino-physical-ai/.venv/lib/python3.12/site-packages/physicalai/inference/model.py", line 127, in init
self.adapter.load(model_path)
File "/home/intel/openvino-physical-ai/.venv/lib/python3.12/site-packages/physicalai/inference/adapters/openvino.py", line 70, in load
self.compiled_model = core.compile_model(model=model, device_name=self.device, config=self.config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/intel/openvino-physical-ai/.venv/lib/python3.12/site-packages/openvino/_ov_api.py", line 646, in compile_model
super().compile_model(model, device_name, {} if config is None else config),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Exception from src/inference/src/cpp/core.cpp:118:
Exception from src/inference/src/dev/plugin.cpp:54:
Check 'false' failed at src/plugins/intel_gpu/src/plugin/program_builder.cpp:163:
[GPU] ProgramBuilder build failed!
clBuildProgram

Without choosing GPU, it looks normally done.

policy = InferenceModel("./SmolVLA/exports/openvino", device="CPU")
policy = InferenceModel("./SmolVLA/exports/openvino")

Steps to reproduce

1. Create uv venv environment
2. Install physicalai in the venv following the instruction
3. install GPU runtime (ocl-icd-libopencl1 and intel-opencl-icd)
   • clinfo -l doesn't show anything
4. install drivers following to the instructions
5. Open python and type commands below.

   from physicalai.inference import InferenceModel
   policy = InferenceModel("./SmolVLA/exports/openvino", device="GPU")

Logs and traceback

Environment

OS: Ubuntu 24.04.4 LTS
Python: 3.12.3
physicalai: 0.1.1
Device: AAEON CEXD-INTRBL
CPU: Intel(R) Core(TM) Ultra X7 358H
Memory: 64GB
Disk: 256.1GB disk

lib.txt

Additional context

No response

[intel-ohara]

I didn't see this issue with ACT like below.

policy = InferenceModel("./ACT/exports/openvino", device="GPU")

[intel-ohara]

The issue persists after upgrade compute-runtime to 26.22.38646.4

[Radwan-Ibrahim]

@intel-ohara This issue has been fixed by the OpenVINO team in the PR referenced below. While upgrading to the latest OpenVINO (26.2) version resolves this problem, it may introduce another regression affecting PI0.5. Upgrading to OpenVINO 2026.3 when it is released is recommended, as both issues have been addressed in that release.

Fix SmolVLA
openvinotoolkit/openvino#35406
Fix PI0.5 regression
openvinotoolkit/openvino#36029

[intel-ohara]

Radwan-Ibrahim, Thanks for the answers!