NinaCalvi/ultra-rm-truthfulness-1000-Armo

该数据集可能用于评估模型生成的文本质量，包含多个复杂的结构，如annotations、completions等，涉及评分、理由、类型等字段。数据集中的字段包括模型响应、用户输入、评分标准等，支持对模型生成文本的帮助性、诚实性、指令遵循性、真实性等方面进行评估。

The dataset contains multiple fields, primarily used for evaluating and analyzing instructions, responses, ratings, and critiques. The features include source, instruction, completions, critique, custom system prompt, fine-grained score, model, overall score, principle, response, correct answers, incorrect answers, split, annotations, custom system prompt, fine-grained score, model, overall score, principle, assistant response, evaluation criteria, chosen score, rationale, rationale for rating, type, overall critique, original critique, user input, rejected score, world knowledge, instance index, rubric, rubric objective, rubric score descriptions, chosen or rejected, judgement for assembly, synth flag, assembled generation prompt, generation messages, generated response, generated completion, finish reason, parse result, critique, judgement, messages, score, etc. The training set of the dataset contains 3000 samples with a data size of 72277226 bytes.