CodeInsights Edge Case

HELM scenario that asks a model to name the one unit-test index a given C++ student is most likely to fail.

Also known as: CodeInsightsEdgeCaseScenario, codeinsights edge case

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.
Categorycoding
SubcategoryHELM prediction of which C++ unit test a CodeInsight student fails
Page statusunknown
Metricunittest_alignment
Directionhigher_is_better
Dataset size23
PublisherStanford CRFM (HELM scenario); CodeInsight dataset from VNU-HCM University of Technology CS

What it measures

codeinsights_edge_case is a Stanford HELM run spec on sampled CodeInsight logs. For each student it loads a topic pass-rate profile and one target C++ problem with parsed unit tests, then asks the model to output a single 0-based unit-test index the student would fail. Scoring extracts the first integer in the completion and checks it against the unique failing bit in that student's `pass` pattern. It is failure-index prediction, not code generation, and not the CodeInsight paper's next-attempt predictor.

Task format

English HELM instruction plus a Vietnamese/English problem and unit-test list; model should return an integer. Default temperature 0, max_tokens 4000, num_testcases=1 (only the first parsed tests are kept in extra_data). The generation adapter still uses output noun "Your code".

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub