The groundtruth answers provided for public test data use multiple unicode encoding. As a result, if the predicted answer use a single unicode encoding, the cider calculation will result in 0. An example is question_id 33615 (answer: "tại khu phố này trời đang nắng") and question_id 16753 (answer: "màu trắng")
Posted by: hoangdta @ Oct. 5, 2023, 3:30 a.m.