Comment on How do we use the computational power of A17 Pro Neural Engine? Correction. Not "by per-tensor quantization", but "by training-time quantization". Machine Learning & AI Core ML May ’24