Full thread0-_-0·It's the cumulative weighting based on the softmax output? Is it per layer?View on HN