Also the 3588 chips can run LLMs on the NPU.

Not quite llama.cpp level easy but definitely doable.

For 7B class models the speed is usable