Running LLMs only on the O6N's NPU

Hi,
This is a experiment/PoC where I demonstrate running dense networks on the NPU:

there is still room for improvement and I’ll probably add a couple more models once I’m “satisfied" with the current performance.

My NPU memclk is locked either at 600MHz or 750MHz although the old docs say it should be able to do 1GHz so also take the current numbers with a grain of salt. It might already perform faster on your boards.

Give it a go and tell me about it :slight_smile: