htp-llm

Standalone llama.cpp runner that drives the Hexagon HTP NPU in a Snapdragon X Elite directly, with custom HVX/HMX kernels inside its own DSP skel library, aiming to run 27B-class LLMs locally on a laptop with no GPU and no cloud. Also carries a work-in-progress port of Prism ML's ternary Bonsai 2 27B weights onto the same NPU. Status: active engineering log, not yet verified end to end.