speed up phi-3 inference?

[CHNtentes]

Hi, I tried your phi-3 example on Android. I wonder if it can run on GPU/Qualcomm HTP to further increase speed? Currently I suppose it only uses CPU.

[salykova]

I am interested in this too

[CHNtentes]

I suppose for now they probably cannnot do it, like executorch only small models can run on gpu/htp.

[varunchariArm]

I am interested too, also may I know what Execution Provider is being used in the provided 'libonnxruntime.so'?
Thans