DCAgent3/terminal_bench_2_a3_rl_DCAgent2_nl2bash_tasks_cleaned_oracle_40_8B_20260526_1825851fc8c
收藏资源简介:
该数据集是一个多轮对话数据集,包含426个训练样本。每条数据包含对话历史(conversations,由角色和内容组成)、代理信息(agent)、模型名称(model)、模型提供者(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)以及追踪来源(trace_source)。数据集可用于评估或训练对话代理,记录任务执行过程中的对话、模型输出和验证结果。
This dataset is a multi-turn dialogue dataset containing 426 training samples. Each data entry includes conversation history (conversations, composed of role and content), agent information (agent), model name (model), model provider (model_provider), date (date), task (task), episode (episode), run ID (run_id), trial name (trial_name), result (result), verifier output (verifier_output), and trace source (trace_source). The dataset can be used for evaluating or training dialogue agents, recording dialogues, model outputs, and verification results during task execution.




