Compile BiseNetv2 model for NPU

[Maick]

Hello,

I try to use the benchmark_app on a BiseNetv2 model and on NPU device, but the compilation process never ends.

If I choose CPU or NPU every think works fine. I also try with a mobilenet model and it works for all devices.

It seems related to this model on this device...

How could I get more information why the compilation failed ?

Here are my configuration: 

System : Intel Core Ultra7 155H

Operating System : Windows 11 (I also try on Ubuntu 22.04)

OpenVino 2024.5.0

NPU driver : 2024.5.0

Thanks for your answer.

[Aznie_Intel]

Hi Maick,

Thanks for reaching out. Can you reconfirm again which plugin is working fine with the BiseNetv2 model? For further checking, you may share the model files (.xml and .bin).

Regards,

Aznie

[Maick]

Hi Azine,

Thanks for your answer.

Sorry for the typo, it works fine with CPU and GPU.

I attached the .xml and .bin of the model.

Regards,

Maïck

[Aznie_Intel]

Hi Maick,

Thanks for the model files. We are checking this with the engineering team and will get back to you once the information is available.

Regards,

Aznie

[Witold_Intel]

Hi, Maick,

Could you please share your NPU driver version with us? OpenVINO developers have reproduced the case without any compilation or inference problems, neither for fp16 nor for quantized (fp16-int8) model version. So our recommendation would be to upgrade to the latest available driver version ( https://github.com/intel/linux-npu-driver/releases/tag/v1.10.0 ). By the way, is your setup MTL or LNL?

[Maick]

Hi,

Thanks for your answer !

Here is what I just tried : 

• Remove and install NPU driver (https://github.com/intel/linux-npu-driver/releases/tag/v1.10.0)
• Create new python environment
• Install openvino 2024.05
• Launch benchmark_app

I still have the same behavior,  the script is blocked on step 7, it cannot build the model.

I don't understand what you mean by MTL or LNL ?

[Witold_Intel]

Thank you for your feedback. MTL is short for Meteor Lake and LNL for Lunar Lake chip family.

[Witold_Intel]

Please bear with us. I relayed your results to the developers but there has been no response yet.

[Witold_Intel]

Thank you for your patience. I haven't received any follow-up from the developers yet. I will try to obtain more information before our next planned meeting on January 15th.

[Witold_Intel]

I don't have any response from the developers yet. I will try to contact them directly.

[Witold_Intel]

As of now, developer response is still pending. Please bear with us.

[Witold_Intel]

Hello, thank you for your patience. There's a new developer looking at your issue. Just a small follow-up question - are NPU driver versions identical in your Windows and Ubuntu instances?

[Maick]

Hello,

Thanks for your feedback, yes I try with 2024.5 on Ubuntu and Windows.

Best regards,

Maïck

[Witold_Intel]

Thank you very much for your response, I will relay this to the developer.

[Witold_Intel]

Hi Maick,

Thank you for your patience. I don't have a response from the developers yet but i will highlight your issue during our sync meeting on Wednesday. Hopefully, I will be able to provide more assistance by then.

[Witold_Intel]

Hi Maick,

in a meeting with developers I learnt that your OpenVINO version is a bit special. 2024.5 requires NPU driver 32.0.100.3052/3053 on Windows but doesn't have a Linux equivalent driver. One option would be to upgrade to OpenVINO 2024.6 or 2025.0 (compatible with 1.10.1 Linux driver). Another one to downgrade to 2024.4 (Linux driver 1.8). Would one of those solve your issue?

[Maick]

Hi,

I try with OpenVIno 2024.6 and 1.10.1 linux driver but the issue is still the same.

I also try with the same model but with a smaller input shape and it works fine. Maybe it could be helpful for you.

Best regards,

Maïck

[Witold_Intel]

Thank you for your feedback. Can you share the smaller model's name and source? I would consult it with the developers then.

[Maick]

Please find attached the smaller model.

[Witold_Intel]

Thank you for the smaller model, we will try it out and get back to you.

[Witold_Intel]

I would just like to check up if the issue still occurs on your side perhaps? Maybe this case can be closed?