Nemotron-RL-Agentic-SWE-Pivot-v1-prompt-only
收藏资源简介:
本数据集名为Nemotron-RL-Agentic-SWE-Pivot-v1-prompt-only,是从源数据集nvidia/Nemotron-RL-Agentic-SWE-Pivot-v1中提取的仅包含提示(prompt-only)的版本。数据以CSV文件(prompts.csv)形式存储,每条记录对应源数据的一行,包含核心提示(prompt)、分离的系统提示(system_prompt)以及当源行定义可用工具时的结构化工具描述(tools),其中嵌套值以JSON格式编码在CSV单元格内。数据集还附带了两个摘要文件:summary.md(记录源行数、提取行数、行数差值和失败提示计数)和null_or_empty_rows.md(列出提示提取结果为null或空值的行索引)。数据规模方面,共提取了50,661行数据,失败行数为0,行数差值为0。数据集标签表明其与Nemotron、仅提示(prompt-only)和后训练(post-training)相关,上传信息显示它来自Nemotron Post-Training v3的提示提取工作流。该数据集适用于软件工程(SWE)相关的代理或强化学习(RL)场景中的提示工程、模型后训练或任务规划等应用。
This dataset, named Nemotron-RL-Agentic-SWE-Pivot-v1-prompt-only, is a prompt-only version extracted from the source dataset nvidia/Nemotron-RL-Agentic-SWE-Pivot-v1. The data is stored in a CSV file (prompts.csv), with each record corresponding to a row from the source data, containing the core prompt, separated system prompt, and structured tool descriptions (tools) when tools are defined in the source row, where nested values are encoded in JSON format within CSV cells. The dataset also includes two summary files: summary.md (recording source row count, extracted row count, row difference, and failed prompt count) and null_or_empty_rows.md (listing row indices where prompt extraction resulted in null or empty values). In terms of data scale, a total of 50,661 rows were extracted, with 0 failed rows and a row difference of 0. The dataset tags indicate its relevance to Nemotron, prompt-only, and post-training, and upload information shows it originates from the Nemotron Post-Training v3 prompt extraction workflow. This dataset is suitable for applications such as prompt engineering, model post-training, or task planning in software engineering (SWE)-related agent or reinforcement learning (RL) scenarios.





