AI Blog

AI Blog

by Michele Laurelli

Perplexity

Metric
Definition

Measurement of how well a probability model predicts a sample, for evaluating language models.

Perplexity = exp(cross_entropy). Lower perplexity indicates better prediction. Commonly used for LM evaluation.

Examples

1

Language model evaluation

2

GPT performance metric

3

Text generation quality

Michele Laurelli - AI Research & Engineering