Mastermath2024v1 (OpenCompass)

OpenCompass four-option Chinese kaoyan math MCQ loader; item count is not published in the configs.

Also known as: Mastermath2024v1, mastermath2024v1_gen

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
SubcategoryChinese kaoyan Mathematics Paper 1 four-option multiple choice
Page statusunknown
Metricaccuracy (exact match of extracted A/B/C/D)
Directionhigher_is_better
Unit%
PublisherOpenCompass (Shanghai AI Laboratory)

What it measures

mastermath2024v1, in this id, is OpenCompass's MastermathDatasetv1: Chinese four-option multiple-choice items loaded from a local CSV named kaoyan_math_1_mcq_Sheet1.csv. The filename points at Mathematics Paper 1 of China's postgraduate entrance examination (kaoyan), not the Gaokao papers in [gaokaobench](gaokaobench.md) and not the mixed-licence KaoshiDataset in [kaoshi](kaoshi.md). The model sees the stem plus options A–D and must emit a letter. The skill is exam-style Chinese math choice, not free-response contest math.

Task format

Zero-shot generation. OpenCompass ZeroRetriever + GenInferencer. Chinese prompt appends 选项 (A)–(D) and asks for the form 正确答案是 (insert answer). first_option_postprocess with options ABCD extracts the first A/B/C/D. MastermathDatasetv1Evaluator then exact-matches the letter.

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub