GSM-Plus-BN

GSM-Plus-BN is a Bangla dataset designed to evaluate robust mathematical reasoning in large language models. It contains 10,553 samples, each representing a Bangla multi-step arithmetic word problem with corresponding solutions and perturbed variations. The dataset includes 7 columns — Bangla_Question, Bangla_Solution, perturbation_type, Bangla_Seed_Question, Bangla_Seed_Solution, and seed_answer. The dataset provides both original and modified problem–solution pairs for robustness analysis. GSM-Plus-BN supports research on multilingual reasoning, linguistic variation, and generalization in low-resource language settings.