eval-laion_64GPU_base_32b-55-32B_DCAgent2_swebench-verified-random-100-folders-traces
收藏资源简介:
该数据集是一个结构化对话记录集合,专门用于评估和分析对话系统性能。它包含738个训练样本,每个样本记录了一次多轮对话的完整交互过程及相关元数据。核心数据字段包括:对话内容(conversations),其中包含按角色(role)组织的多条消息内容(content);执行对话的智能体标识(agent);使用的模型信息(model)及其提供商(model_provider);对话发生日期(date);任务类型(task);对话轮次标识(episode);运行ID(run_id);试验名称(trial_name);对话结果(result);验证器输出(verifier_output);以及数据来源追踪(trace_source)。该数据集适用于对话系统评估、多轮对话分析、不同模型或智能体的性能比较、任务导向对话研究等场景。数据以结构化格式存储,便于进行定量分析和实验复现。
This dataset is a structured collection of dialogue records, specifically designed for evaluating and analyzing dialogue system performance. It contains 738 training samples, each recording the complete interaction process of a multi-turn dialogue along with related metadata. Key data fields include: dialogue content (conversations), which consists of multiple message contents (content) organized by role (role); agent identifier (agent) executing the dialogue; model information (model) and its provider (model_provider); date of dialogue (date); task type (task); dialogue episode identifier (episode); run ID (run_id); trial name (trial_name); dialogue result (result); verifier output (verifier_output); and trace source for data provenance (trace_source). The dataset is suitable for scenarios such as dialogue system evaluation, multi-turn dialogue analysis, performance comparison of different models or agents, and task-oriented dialogue research. Data is stored in a structured format, facilitating quantitative analysis and experimental reproducibility.
- 数据集名称: eval-laion_64GPU_base_32b-55-32B_DCAgent2_swebench-verified-random-100-folders-traces
- 数据集地址: https://huggingface.co/datasets/laion/eval-laion_64GPU_base_32b-55-32B_DCAgent2_swebench-verified-random-100-folders-traces
数据集特征:
conversations: 由多条消息组成,每条消息包含content(字符串类型)和role(字符串类型)两个字段。agent: 字符串类型,表示智能体。model: 字符串类型,表示模型。model_provider: 字符串类型,表示模型提供方。date: 字符串类型,表示日期。task: 字符串类型,表示任务。episode: 字符串类型,表示回合。run_id: 字符串类型,表示运行ID。trial_name: 字符串类型,表示试验名称。result: 字符串类型,表示结果。verifier_output: 字符串类型,表示验证器输出。trace_source: 字符串类型,表示追踪来源。
数据规模:
- 仅包含一个
train划分。 - 训练集样本数量: 738 条。
- 数据集总大小: 132,771,609 字节(约 126.6 MB)。
- 下载大小: 73,465,614 字节(约 70.1 MB)。




