jayshah5696/humanize-rl-tasks

humanize-rl-tasks 是一个用于 humanize-rl 项目的单轮强化学习任务数据集。每行代表一个写作任务，模型接收提示（指令加源文本），生成补全内容，环境使用50/50奖励公式进行评分：奖励 = 0.50 × ridge_rubric_mean + 0.50 × deterministic_mean + 惩罚。数据集总共有612个任务，分为13个家族，包括模板生成和真实来源（如电子邮件、创意文本等）。输入长度平均约为103个单词，最长可达221个单词。数据集分为训练集（489行）、验证集（62行）和测试集（61行）。每个任务包含ID、家族、领域、模式、寄存器、指令、输入文本、约束、奖励配置、必需事实、禁止事实、禁止短语、陷阱标签、分割和来源等字段。humanize-rl-tasks is a single-turn RL task dataset for the humanize-rl project. Each row is one writing task. A model receives the prompt (instruction + source text), produces a completion, and the environment scores it with the 50/50 reward formula: reward = 0.50 × ridge_rubric_mean + 0.50 × deterministic_mean + penalties. The dataset consists of 612 tasks across 13 families, including template-generated and real sources (e.g., email, creative, general). Input length averages ~103 words, up to 221 words. It is split into train (489 rows), validation (62 rows), and test (61 rows). Each task includes fields such as id, family, domain, mode, register, instruction, input_text, constraints, reward_profile, required_facts, forbidden_facts, forbidden_phrases, trap_tags, split, and source.