Length-Normalized Sequence Log Prob

Problem

Return the length-normalized log-probability of every sequence, i.e. the mean token log-probability of that sequence. An empty sequence contributes 0.0, and an empty batch returns [].

Hint

Length normalization is what lets you compare log-probabilities across different sequence lengths.

Test cases

case 1

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

expected: [-1.5,0,0]

case 2

in: [[[-0.5,-0.5,-0.5]]]

expected: [-0.5]

case 3

in: [[[]]]

expected: [0]

Starter codePython

def mean_sequence_logps(sequences):
    # Your code here
    pass

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