CodeInsights Correct Code

HELM scenario that asks a model to fill a C++ course template so the generated body passes the problem's unit tests.

Also known as: CodeInsightsCorrectCodeScenario, codeinsights correct code

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 C++ unit-test pass rate on a CodeInsight question sample
Page statusunknown
Metricfunctional_correctness
Directionhigher_is_better
Dataset size84
PublisherStanford CRFM (HELM scenario); CodeInsight dataset from VNU-HCM University of Technology CS

What it measures

codeinsights_correct_code is a Stanford HELM run spec on sampled CodeInsight course logs. It groups Kazchoko/my_dataset Scenario1_2_data.csv by question_unittest_id, takes the first row of each question, and asks the model to write only the C++ that replaces {{ STUDENT_ANSWER }} in the course template. Scoring compiles with g++ and measures the fraction of parsed unit tests that match expected stdout. It is functional correctness on undergraduate C++ assignments, not the CodeInsight paper's next-attempt predictor, not student-style imitation, and not [ci_mcqa](ci_mcqa.md).

Task format

English HELM instruction plus a Vietnamese/English problem prompt; model returns a fenced C++ fragment. Default temperature 0, max_tokens 4000, num_testcases=1 (only the first parsed unit test is kept).

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub