poolside-laguna-hackathon/py-bug-trace-laguna-xs-2-l2-rollouts
收藏资源简介:
该数据集包含用于AI模型评估的交互轨迹数据,每个示例包括唯一的trace_id和example_id,以及rollout_number表示迭代次数。任务字段暂未定义,prompt和completion字段分别存储用户输入和模型响应,其中completion包含推理内容。奖励(reward)和准确匹配奖励(exact_match_reward)用于量化模型性能,同时记录延迟(latency_ms)和总时间(total_time)。info字段提供元数据,如任务难度、类别、时间统计和令牌使用情况,支持模型效率分析。数据集分为训练集,包含15个示例,总大小约38KB,适用于强化学习或对话系统的评估场景。
This dataset contains interaction trajectory data for AI model evaluation, with each example including unique trace_id and example_id, along with rollout_number indicating iteration counts. The task field is undefined, while prompt and completion fields store user inputs and model responses respectively, with completion containing reasoning content. Rewards (reward and exact_match_reward) quantify model performance, and metrics such as latency_ms and total_time are recorded. The info field provides metadata like task difficulty, category, timing statistics, and token usage, enabling model efficiency analysis. The dataset is split into a training set with 15 examples, totaling approximately 38KB, suitable for evaluation scenarios in reinforcement learning or dialogue systems.




