Nemotron-RL-ReasoningGym-v1-prompt-only
收藏资源简介:
Nemotron-RL-ReasoningGym-v1-prompt-only 是一个从源数据集 nvidia/Nemotron-RL-ReasoningGym-v1 中提取出的、仅包含提示(prompt)部分的数据集,专为强化学习后训练(post-training)场景设计,属于 Nemotron 项目的一部分。数据集的核心文件是 prompts.csv,包含 15000 条提取记录,每条记录对应源数据的一行,并包含提取出的用户提示(prompt)、独立的系统提示(system_prompt),以及当源数据定义了可用工具时的结构化工具描述(tools),这些嵌套值以 JSON 格式存储在 CSV 单元格内。数据集还提供了总结文件(summary.md)和空值行索引文件(null_or_empty_rows.md),用于记录提取过程的统计信息和失败情况。根据统计,所有 15000 行提示均成功提取,无失败记录。该数据集适用于需要高质量、结构化提示文本的任务,例如大语言模型的指令微调、提示工程或强化学习中的环境交互模拟。
Nemotron-RL-ReasoningGym-v1-prompt-only is a prompt-only dataset extracted from the source dataset nvidia/Nemotron-RL-ReasoningGym-v1. It is designed specifically for reinforcement learning post-training scenarios and is part of the Nemotron project. The core file of this dataset is prompts.csv, which contains 15,000 extracted records. Each record corresponds to one row of the source data, and includes the extracted user prompt, standalone system prompt, and structured tool description (tools) when the source data defines available tools. These nested values are stored in JSON format within the CSV cells. The dataset also provides a summary file (summary.md) and a null or empty row index file (null_or_empty_rows.md) for recording statistical information and failure cases during the extraction process. According to statistics, all 15,000 prompt rows were successfully extracted with no failed records. This dataset is suitable for tasks requiring high-quality, structured prompt text, such as instruction fine-tuning of large language models (LLMs), prompt engineering, or environment interaction simulation in reinforcement learning.




