Perplexity

Appears in 1 paper · 1 tutorial

A metric for language models derived from cross-entropy loss.

As used in Paper 12 — Language Models are Few-Shot Learners →

A metric for language models derived from cross-entropy loss. Measures how "surprised" the model is by the test data. Lower perplexity = better. It's the exponential of the cross-entropy loss.

As used in Fine-Tuning & Model Customization →

A friendlier transform of loss: roughly "how many tokens was the model wavering between." Lower is better. (M03)

Paper 12 — Language Models are Few-Shot Learners → Fine-Tuning & Model Customization →