Function should_use_gemv 

pub fn should_use_gemv(_x: &Tensor, _w: &Tensor) -> bool

Fallback for non-CUDA builds