AI Glossary

Perplexity

What does “Perplexity” mean?

Perplexity: A language-model measure derived from average token prediction loss. Lower values mean the model assigned more probability to the test text under that setup.

Perplexity visual guide: A language-model measure derived from average token prediction loss. Lower values mean the model assigned more probability to the test text under that setup.
Visual guide Open full size (new tab)

What is an example of “Perplexity”?

Researchers compare perplexity on the same text using a shared tokenizer.

Keep learning

See also

EvalsTermBenchmarkTermSWE-benchTermExact match scoreTermBLEUTerm

Same topic

ROUGETermPersonal information redactionTermPass@kTermPrecisionTermOut-of-distribution detectionTerm