eval-laion_symclip-30-8B_DCAgent2_swebench-verified-random-100-folders-traces
收藏资源简介:
该数据集是一个专为AI模型评估与分析设计的结构化对话数据集。其核心由多轮对话构成,每条记录包含一个对话序列(conversations),其中每个对话回合均标明了发言角色(role)和具体内容(content)。数据集还包含了丰富的元数据,用于记录对话运行的上下文信息,包括:使用的智能体(agent)、模型名称(model)、模型提供商(model_provider)、任务类型(task)、运行日期(date)、运行标识(run_id)、试验名称(trial_name)、关键的结果指标(result)以及验证器输出(verifier_output)。数据来源通过trace_source字段标注。数据集共包含1775个训练样本,总大小约为298 MB。它适用于研究AI模型在多轮对话任务中的表现、进行模型行为分析、结果验证以及任务特定的性能评估。
This is a structured conversational dataset specifically designed for AI model evaluation and analysis. Its core consists of multi-turn dialogues, where each record contains a conversation sequence (conversations), and each dialogue turn is annotated with the speaker role (role) and corresponding content (content). The dataset also includes rich metadata for recording contextual information of dialogue runs, including: the employed AI Agent (agent), model name (model), model provider (model_provider), task type (task), run date (date), run ID (run_id), trial name (trial_name), key result metrics (result), and verifier output (verifier_output). The data source is labeled via the trace_source field. The dataset comprises a total of 1,775 training samples, with an overall size of approximately 298 MB. It is applicable for researching the performance of AI models in multi-turn dialogue tasks, conducting model behavior analysis, result verification, and task-specific performance evaluation.
- 数据集名称: eval-laion_symclip-30-8B_DCAgent2_swebench-verified-random-100-folders-traces
- 来源: LAION (Large-scale Artificial Intelligence Open Network)
- 数据集大小: 约285 MB (298,483,934 字节)
- 下载大小: 约152 MB (158,991,158 字节)
- 数据规模: 包含1,775个样本,所有样本均位于训练集(train split)
- 数据字段:
conversations: 对话内容列表,每条包含content(字符串型) 和role(字符串型)agent: 智能体标识 (字符串型)model: 使用的模型名称 (字符串型)model_provider: 模型提供方 (字符串型)date: 日期 (字符串型)task: 任务描述 (字符串型)episode: 剧集/回合标识 (字符串型)run_id: 运行ID (字符串型)trial_name: 试验名称 (字符串型)result: 结果 (字符串型)verifier_output: 验证器输出 (字符串型)trace_source: 追踪来源 (字符串型)
- 数据配置: 默认配置(config_name: default),数据文件路径为
data/train-*




