Determining if AMX is in use by ollama

[js333031]

Hello, I used latest steps to install ipex-llm into a venv on a 5th Gen Xeon system. I don't think AMX is being utilized based on screenshot below. Should AMX show up in list of CPU features in output below - last line? lscpu shows AMX instructions are present. Can you please show a verification step for ipex-llm confirming that CPU optimizations are installed in ipex-llm venv?

Thanks

[sgwhat]

Hi @js333031 , ipex-llm ollama is a GPU-optimized version, so we currently do not plan to work on this.

[js333031]

Has there been any change to the plan for optimizing for Xeon/AMX?

[glorysdj]

Sorry, we don't have any plan for optimizing ollama for Xeon/AMX yet.

[endomorphosis]

I am also querying about this, but and I have noticed that its supported by the llama_cpp package, so it just depends I think on how downstream they are from changes made to the base package.

[tjwebb]

any update on this?

[js333031]

I have not yet tried but this is one way to trace: https://www.intel.com/content/www/us/en/docs/onednn/developer-guide-reference/2025-0/verbose-mode.html

For PyTorch apps using oneDNN: enable env variable as shown in https://docs.pytorch.org/tutorials/recipes/amx.html