Any plan to support Qualcomm Hexagon NPU / QNN backend? #41

Hi team,

vla.cpp looks very promising for running VLA models outside of large GPU servers.

Many robotics platforms are moving toward power-efficient edge AI hardware, especially Qualcomm Snapdragon-based systems used in robotics.

Would the project consider adding support for:

Qualcomm QNN backend
Hexagon NPU acceleration
Snapdragon X Elite/X Plus
Qualcomm Robotics SoCs (QCS9075, QCS8797 etc.)
A QNN backend could make it possible to deploy models such as OpenVLA, Pi0, SmolVLA, and other VLA models with significantly better power efficiency on embedded devices.

Is this something on the roadmap, or would the maintainers be interested in community contributions in this area?

Thanks!