MediumNLPnlp-385

Repetition Penalty Apply

Problem

The repetition penalty divides positive logits by the penalty and multiplies negative logits by it, pushing repeated tokens down either way.

Given the logits and the penalty > 0, return the adjusted logits.

Hint

The penalty acts in the direction that lowers the token.

case 1

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

expected: [0.5,-2]

case 2

in: [[0,0.5],1.5]

expected: [0,0.3333333333333333]

case 3

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

expected: [2,-2]

Starter codePython

def repetition_penalty_apply(logits, penalty):
    # Your code here
    pass

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