3 ms·Inference at the edge: Efficient transformer model inference on-device3 points by lioeters 3y ago