mLLAMA 3.2 Output Issue and Gradio Error on Intel SPR Machine

[saranyabalakumar]

I'm running mLLAMA 3.2 on two machines:

Intel 4th Gen SPR server NVIDIA GPU machine On the NVIDIA GPU machine, everything works fine. But on the Intel SPR machine, the output is strange (!!!!!!!!!!!!!!!!!!), and Gradio gives errors when I try to open it on both the server and cloud setups.

It seems like the issue is specific to the Intel SPR machine. Any ideas on what might be causing this?

[brmarkus]

Can you provide more details, please? On the server machine, are you using Jupyter-notebook directly (the server with a monitor connected), or remotely (e.g. screen-forwarding?

Which OS, which OS-version, which Python version, which version of OpenVINO-Notebooks?

You asked the question "To provide all the information in csv format from the images?". Which output issue do you mean, seeing all those "!!!"?

Do you see the same issues&errors when using another picture, another question?

[saranyabalakumar]

I am using the server machine remotely. I convert the Jupyter notebook code to Python scripts and use them for inference.

OS: Ubuntu Linux
Kernel Version: 6.5.0-41-generic Ubuntu version : 22.04.2 Python version: 3.10.15 OpenVINO-Notebooks Version: latest

I tried using one or more images with prompts, but the output response took a long time and all I got was '!!!'.

[brmarkus]

For consistency check, would you be able to install the prerequisites and the packages mentioned in this notebook on another (non-SPR-)machine - and run your ported Python script? It could be related to the CPU-instruction-sets mentioend in the log-messages, could also be due to prerequisites, or due to the ported Python script...

Maybe the dev-team anyway will ask you to share the ported Python-script for reproduction and deeper analysis as well as the output of pip freeze.

[saranyabalakumar]

Sure, I can install the prerequisites and packages mentioned in the notebook on a non-SPR machine freeze.txt