EasyNLPnlp-365

Attention Padding Mask Scores

Problem

Padding positions receive a large negative score so their softmax weight vanishes. Use -1e9 for masked entries.

Given the scores and a 1/0 keep mask, return the masked scores.

Hint

Keep scores where the mask is one.

case 1

in: [[1,2,3],[1,0,1]]

expected: [1,-1000000000,3]

case 2

in: [[0.5,0.5],[0,0]]

expected: [-1000000000,-1000000000]

case 3

in: [[1],[1]]

expected: [1]

Starter codePython

def attention_padding_mask_scores(scores, mask):
    # Your code here
    pass

3 test cases · run your code against them in the editor