terminal_bench_2_a3_rl_DCAgent_selfinstruct_naive_sandboxes_2_verified_70_8B_20260f49e33da
收藏资源简介:
该数据集为多轮对话数据,每条样本包含完整的对话记录(conversations字段,由角色role和内容content组成),以及相关的元信息:使用的代理(agent)、模型(model)及模型提供商(model_provider)、日期(date)、任务(task)、回合(episode)、运行ID(run_id)、试验名称(trial_name)、结果(result)、验证器输出(verifier_output)和追踪来源(trace_source)。训练集共包含5272条样本,数据总量约423MB。该数据集可用于对话系统训练、模型评估、多轮对话行为分析等任务。
This dataset consists of multi-turn dialogue data. Each sample contains a complete conversation record (conversations field, composed of role and content) along with related metadata: agent, model, model_provider, date, task, episode, run_id, trial_name, result, verifier_output, and trace_source. The training set includes 5272 samples, totaling approximately 423 MB. The dataset can be used for dialogue system training, model evaluation, multi-turn dialogue behavior analysis, and other tasks.



