swiss-ai/if-rl-singleturn-prompts
收藏资源简介:
该数据集是一个用于评估语言模型在约束条件下性能的数据集,包含79,622个训练样本。每个样本包括唯一标识符(key)、多轮对话消息(messages,含角色和内容)、真实答案(ground_truth)、数据集来源(dataset)、约束类型(constraint_type)、具体约束条件(constraint),以及多个评估字段(如pass_by_constraint、total_pass等,部分字段值为空)。数据集旨在测试模型在遵循特定约束(如安全或格式要求)时的响应能力,适用于自然语言处理任务中的对齐研究和性能基准测试。
This dataset is designed for evaluating the performance of language models under constraints, containing 79,622 training examples. Each example includes a unique identifier (key), multi-turn dialogue messages (messages with role and content), ground truth answers (ground_truth), dataset source (dataset), constraint type (constraint_type), specific constraint conditions (constraint), and multiple evaluation fields (e.g., pass_by_constraint, total_pass, with some fields null). The dataset aims to test model responses while adhering to specific constraints (such as safety or format requirements), suitable for alignment research and performance benchmarking in natural language processing tasks.




