lasgroup/verifiable-corpus

这是一个用于研究大型语言模型在测试时如何在目标任务上持续改进推理能力的语料库。该数据集通过组装特定任务的测试时课程（TTC-RL）并应用强化学习来继续训练模型以完成其目标任务。数据集包括问题类型、原始数据集名称、问题描述、问题（包括模板）、答案（除代码数据集外）和测试用例（序列化为json格式）。

This is a corpus for studying how large language models (LLMs) can continually improve their reasoning on target tasks at test-time. The dataset involves assembling a task-specific test-time curriculum (TTC-RL) and applying reinforcement learning to further train the model for its target task. It includes fields for problem type, original dataset name, problem description, problem (including boilerplate), answer (for all datasets except code), and test cases (serialized to json).