eval-fsr-a1-softwareheritage-swe-r366-rf0710-traces
收藏资源简介:
该数据集是一个专为多轮对话任务设计的结构化数据集,包含3607个训练样本,数据规模约为574.9MB。它以对话交互为核心,每条记录由对话列表(conversations)组成,每个对话回合包含内容(content)和角色(role)字段。此外,数据集提供了丰富的元数据,如代理标识(agent)、模型信息(模型名称model、模型提供商model_provider)、日期(date)、任务类型(task)、实验追踪信息(episode、run_id、trial_name)、结果(result)、验证输出(verifier_output)以及数据溯源来源(trace_source)。这些元数据表明数据集可能应用于AI代理的交互实验、对话生成评估或任务执行追踪等场景。数据集仅包含训练分片,适用于自然语言处理和机器学习研究。
This dataset is a structured collection designed for multi-turn dialogue tasks, containing 3607 training samples with a data size of approximately 574.9MB. It centers around conversational interactions, where each record consists of a dialogue list (conversations), with each turn including content and role fields. Additionally, the dataset provides extensive metadata, such as agent identifiers, model information (model name and model provider), date, task type, experimental tracking details (episode, run_id, trial_name), results, verifier output, and trace source. This metadata suggests potential applications in AI agent interaction experiments, dialogue generation evaluation, or task execution tracking. The dataset only includes training shards and is suitable for natural language processing and machine learning research.




