Token Perplexity

Problem

Convert an average negative log-likelihood per token measured in nats to perplexity:

perplexity = exp(nll_nats_per_token)

Hint

Perplexity is the effective number of equally likely next tokens.

case 1

in: [0]

expected: 1

case 2

in: [0.6931471805599453]

expected: 2

case 3

in: [1.6094379124341003]

expected: 4.999999999999999

case 4

in: [2.302585092994046]

expected: 10.000000000000002

Starter codePython

import math
def token_perplexity(nll_nats_per_token):
    # Your code here
    pass

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