simple_arithmetic_multiple_targets_json (BIG-bench multi-target JSON template)

A 10-item BIG-bench JSON template of one-digit addition that accepts either a numeral, an English word, or both as a correct target.

unassessed

This page is a discovery lead. Nobody has yet assessed it against the catalogue contract, so it carries no disposition. Absence of evidence here is not evidence of staleness.
Categorymath
SubcategoryBIG-bench JSON template; addition with multiple acceptable targets
Page statusunknown
Metricexact_str_match
Directionhigher_is_better
Unit%
Dataset size10
Dataset licenceApache-2.0
PublisherGoogle (BIG-bench organizers)

What it measures

simple_arithmetic_multiple_targets_json shows a one-digit addition such as "1 + 4 = " and treats the answer as correct if the model produces any of one or more listed targets. Targets mix Arabic numerals ("5") and English number words ("five"), and the list length is not the same on every item: some items have two targets, some have one, and two items list only a word ("thirteen", "four") with no numeral. The organizers wrote it to show JSON multiple-target scoring. The README notes that ROUGE is not supported with multiple targets. Tagged "example task".

Task format

Free-text completion of a one-digit addition equation; any listed target (numeral and/or English word) is accepted as correct under exact_str_match.

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub