toksuite/toksuite_math

Dataset Card for Tokenization Robustness (Math) TokSuite Benchmark (Math Collection) Dataset Description This dataset is part of TokSuite, a comprehensive benchmark designed to measure how different tokenization strategies affect language model behavior under controlled conditions. This specific subset focuses on mathematical text completion, containing multiple-choice math questions with a variety of surface-form perturbations that stress… See the full description on the dataset page: https://huggingface.co/datasets/toksuite/toksuite_math.