Cannot create ClientRunner - Ubuntu 24.04, RTX 5060Ti

General Discussion
1

As the title suggests, I am trying to optimize a conversion of an existing model to Hailo’s .hef format. However I cannot utilize the GPU during conversion, as I run into the below error when creating a ClientRunner object.

tensorflow.python.framework.errors_impl.NotFoundError: /local/workspace/hailo_virtualenv/lib/python3.10/site-packages/hailo_sdk_client/emulator/emulator/lib/HSim.so: undefined symbol: _ZN3tsl8internal15LogMessageFatalC1EPKci

I am running the Hailo AI Software Suite inside a docker container (software suite is version 2025_10).
Other related information are as follows:
Nvidia driver version 570
CUDA version 12.8
tf-nightly version 2.21
Ubuntu 24.04

The GPU is detected inside the docker container when running nvidia-smi and also when running tensorflow.test.is_gpu_available()

How do I fix this?

2

Hi @user869,

The error is caused by a TensorFlow version mismatch. The Hailo SDK’s HSim.so was built against the TensorFlow version shipped with the SW Suite, but you’re using tf-nightly 2.21 which has incompatible internal symbols.

Please try to revert to the TensorFlow version bundled with the Hailo Docker container:
pip install tensorflow==<version bundled with suite 2025_10>
Check the expected version with:
pip show hailo-sdk-client | grep Requires

Thanks,