eval-fsr-a1-unitsyn-python-swe-r388-rf0711-traces
收藏资源简介:
该数据集是一个结构化记录集,核心内容围绕多轮对话交互与任务执行过程。每个数据样本包含多个关键字段:1)一个对话列表(conversations),其中每条记录由内容(content)和发言者角色(role)构成,记录了完整的交互序列;2)标识字段,包括执行任务的智能体(agent)、所使用的模型(model)及其提供商(model_provider)、日期(date)、任务描述(task)、情景编号(episode)、运行ID(run_id)和试验名称(trial_name);3)结果与验证字段,包括任务执行结果(result)、验证器输出(verifier_output)以及数据追踪来源(trace_source)。数据集规模为2809个样本,全部划分为训练集。从字段命名推断,该数据集可能用于记录和评估不同模型或智能体在特定任务上的多轮交互表现、输出结果及其验证情况,适用于对话系统评估、智能体行为分析或模型性能比较等研究场景。
This dataset is a structured collection of records centered on multi-turn dialogue interactions and task execution processes. Each data sample includes multiple key fields: 1) A conversation list (conversations), where each entry consists of a content field and a role field, documenting the complete interaction sequence; 2) Identification fields including the task-executing AI agent (agent), the utilized model (model) and its provider (model_provider), date (date), task description (task), episode number (episode), run ID (run_id), and trial name (trial_name); 3) Result and verification fields including task execution result (result), verifier output (verifier_output), and data trace source (trace_source). The dataset contains 2809 samples in total, all partitioned into the training set. Based on the field naming, this dataset can be used to record and evaluate the multi-turn interaction performance, output results and their verification status of different models or AI agents on specific tasks, and is applicable to research scenarios such as dialogue system evaluation, AI agent behavior analysis, or model performance comparison.
- 数据集名称:laion/eval-fsr-a1-unitsyn-python-swe-r388-rf0711-traces
- 数据集大小:下载大小为355,194,885字节,数据集总大小为485,080,553字节。
- 数据集分割:仅包含训练集(train),共2,809个样本。
- 特征字段:
- conversations:对话列表,包含content(字符串)和role(字符串)字段。
- agent:字符串,表示智能体。
- model:字符串,表示模型。
- model_provider:字符串,表示模型提供者。
- date:字符串,表示日期。
- task:字符串,表示任务。
- episode:字符串,表示回合。
- run_id:字符串,表示运行ID。
- trial_name:字符串,表示试验名称。
- result:字符串,表示结果。
- verifier_output:字符串,表示验证器输出。
- trace_source:字符串,表示追踪来源。
- 配置文件:
- config名称:default
- 数据文件路径:训练集对应的文件位于
data/train-*。



