NinaCalvi/ultra-rm-truthfulness-1000
收藏资源简介:
该数据集包含多个复杂的特征字段,主要用于评估模型的回答质量。数据集的主要特征包括来源(source)、指令(instruction)、完成情况(completions)等。完成情况字段下包含多个子字段,如注释(annotations)、批评(critique)、模型(model)等。注释字段进一步细分为多个评分和理由字段,如帮助性(helpfulness)、诚实性(honesty)、指令遵循(instruction_following)等。此外,数据集还包含正确答案(correct_answers)、错误答案(incorrect_answers)、分割(split)等字段。数据集的分割信息显示,训练集包含1000个样本,总大小为24281992字节。
This dataset contains multiple complex feature fields, primarily used to evaluate the quality of model responses. The main features of the dataset include source, instruction, completions, etc. The completions field contains multiple subfields such as annotations, critique, model, etc. The annotations field is further subdivided into multiple rating and rationale fields, such as helpfulness, honesty, instruction_following, etc. Additionally, the dataset includes correct_answers, incorrect_answers, split, and other fields. The split information shows that the training set contains 1000 samples with a total size of 24281992 bytes.



