Nemotron-RL-Instruction-Following-Structured-Outputs-v2-prompt-only
收藏资源简介:
Nemotron-RL-Instruction-Following-Structured-Outputs-v2-prompt-only 是一个专门从源数据集 nvidia/Nemotron-RL-Instruction-Following-Structured-Outputs-v2 中提取仅提示(prompt)部分的数据集。该数据集旨在为指令跟随和结构化输出任务提供高质量的提示数据,适用于大型语言模型的后训练和微调场景。数据集包含一个核心数据文件 prompts.csv,其中每条记录对应源数据的一行,并提供了完整的提示信息,包括用户提示、独立的系统提示,以及当源数据定义了可用工具时的结构化工具描述(以JSON格式编码存储在CSV单元格内)。此外,数据集还附带了摘要文件和空值记录文件,用于说明数据提取的统计信息和异常情况。数据规模方面,共成功提取了62,696条提示记录,失败记录数为0,与源数据行数完全一致。该数据集由Nemotron Post-Training v3的提示提取工作流生成,并通过jamesdborin账户发布,是Nemotron系列后训练数据的重要组成部分。
Nemotron-RL-Instruction-Following-Structured-Outputs-v2-prompt-only is a dataset specifically extracted from the source dataset nvidia/Nemotron-RL-Instruction-Following-Structured-Outputs-v2, containing only the prompt portions. It is designed to provide high-quality prompt data for instruction-following and structured-output tasks, suitable for post-training and fine-tuning scenarios of large language models. The dataset includes a core data file prompts.csv, where each record corresponds to a row from the source data and provides complete prompt information, including user prompts, independent system prompts, and structured tool descriptions (when tools are defined in the source data), with these nested structured data encoded in JSON format within CSV cells. Additionally, the dataset comes with summary files (summary.md) and null or empty row records files (null_or_empty_rows.md) to illustrate statistical information and anomalies from the data extraction. In terms of data scale, a total of 62,696 prompt records were successfully extracted, with 0 failed records, fully consistent with the source data row count. This dataset is generated by the Nemotron Post-Training v3 prompt extraction workflow and released via the jamesdborin account, serving as an important component of the Nemotron series post-training data.
数据集概述
该数据集是 nvidia/Nemotron-RL-Instruction-Following-Structured-Outputs-v2 的提示(prompt)提取版本,专注于指令遵循与结构化输出场景。
基本信息
- 数据集名称: Nemotron-RL-Instruction-Following-Structured-Outputs-v2-prompt-only
- 来源数据集: nvidia/Nemotron-RL-Instruction-Following-Structured-Outputs-v2
- 标签: nemotron, prompt-only, post-training
- 上传者: jamesdborin
文件结构
数据集包含以下文件:
- prompts.csv:主数据文件,每条源记录对应一个提示提取记录。字段包括:
prompt(提示内容)system_prompt(分离的系统提示)tools(结构化工具定义,当源行定义了可用工具时存在;嵌套值以 JSON 编码存储在 CSV 单元格中)
- summary.md:包含源行计数、提取行计数、计数差异及失败提示计数。
- null_or_empty_rows.md:列出提取结果为空或空提示的行索引。
数据统计
| 指标 | 数值 |
|---|---|
| 提取行数 | 62696 |
| 失败提示行数 | 0 |
| 行数差异 | 0 |
用途标签
该数据集标注为 post-training,适合用于后训练阶段的指令遵循和结构化输出能力评估与优化。




