poolside-laguna-hackathon/py-bug-trace-laguna-xs-2-l1-rollouts
收藏资源简介:
该数据集是一个用于评估AI代理或强化学习模型性能的数据集,包含任务执行过程中的追踪信息。数据特征包括trace_id(追踪标识)、example_id(示例标识)、rollout_number(滚动编号)、task(任务)、prompt(提示,包含角色和内容)、completion(完成响应,包含角色、内容和推理内容)、answer(答案)、reward(奖励分数)、score(得分)、correct(正确性)、num_steps(步数)、total_time(总时间)、latency_ms(延迟毫秒)、created_at(创建时间)以及info(信息,包括任务难度、类别、时间统计、令牌使用量、完成状态、截断状态、停止条件和评估指标如精确匹配奖励和轮次数量)。数据集有35个训练样本,旨在分析模型在复杂任务中的表现、效率和准确性。
This dataset is designed for evaluating the performance of AI agents or reinforcement learning models, containing trace information during task execution. Features include trace_id, example_id, rollout_number, task, prompt (with role and content), completion (with role, content, and reasoning content), answer, reward, score, correct, num_steps, total_time, latency_ms, created_at, and info (including task difficulty, category, timing statistics, token usage, completion status, truncation status, stop condition, and evaluation metrics such as exact match reward and number of turns). The dataset consists of 35 training examples and aims to analyze model performance, efficiency, and accuracy in complex tasks.




