ngotrnghia1811/

mSCORE

mSCORE is a multilingual, skill-annotated benchmark for evaluating commonsense reasoning capabilities of large language models. The benchmark features:

Two sub-benchmarks: mSCoRe-G (General) across 5 languages (en, de, fr, zh, ja) and mSCoRe-S (Social/Cultural) with tiktok and reddit splits. Four complexity levels (L0–L3) created through a structured context-expansion and implicitation pipeline, progressively hiding… See the full description on the dataset page: https://huggingface.co/datasets/ngotrnghia1811/mSCORE.