alialp207/TR-DataAnalystBench

TR-DataAnalystBench A Turkish-language benchmark for evaluating whether language models can perform data-analyst style reasoning over tables and charts: reading a value, finding the maximum/minimum, comparing two years, computing an average or a (signed) percentage change, ranking, summarizing a trend, and — importantly — abstaining when the data does not contain the answer. Gold answers are computed and verified with Python (not produced by a language model), so the benchmark… See the full description on the dataset page: https://huggingface.co/datasets/alialp207/TR-DataAnalystBench.