How can I use this ggml-hsa to enable llama.cpp running on AMD NPU？

[MaoxinYee]

Dear ggml_hsa team,

We noticed that your project serves as a critical backend for accelerating llama.cpp on AMD GPUs/NPUs. We'd like to ask: we want to use ggml-hsa to enable llama.cpp to run on AMD NPU. What efforts do I need to make, and could you provide some tips and assistance?

Thank you!

[riverzhou]

same question

[bil-ash]

@ypapadop-amd Same question here. As of today, can I use this repo instead of the upstream ggml for model inference on my AMD NPU ( 8700GE ) ?
If yes, adding a few examples would be nice.
If no, it would be nice if you add a TODO.md file for the tasks remaining to achieve feature parity with upstream and then go on ticking each point as they get implemented. This will bring transparency for people like me and many others who would just like to perform ML inference on NPU but do not have much idea about the codebase.