Attention Weights

Each row of the post-softmax attention weight matrix corresponds to which side of the QK product?

Spot the error: 'I apply the causal mask by setting masked positions to 0 in the attention scores before softmax.'

Predict the softmax output for a causal masked attention row.