laion/nemotron-gym-math-stack-overflow

该数据集是nvidia/Nemotron-RL-math-stack_overflow的Harbor格式转换版本，专注于数学问题（源自Stack Overflow），用于强化学习任务。每个数据行包含一个确定性短ID路径和一个gzipped tar文件，该文件封装了完整的Harbor任务，包括指令提示、Docker环境、验证器脚本和元数据等。数据集设计安全，内容不插值到代码中，使用固定基础镜像，并经过严格验证。它适用于在Harbor环境中运行和评估强化学习代理，特别针对数学问题解决任务。

This dataset is a Harbor-format conversion of nvidia/Nemotron-RL-math-stack_overflow, focusing on mathematical problems (sourced from Stack Overflow) for reinforcement learning tasks. Each row contains a deterministic short ID path and a gzipped tar file that encapsulates a full Harbor task, including instruction prompts, Docker environment, verifier scripts, and metadata. The dataset is designed with security in mind, avoiding content interpolation into code, using a pinned base image, and undergoing rigorous validation. It is suitable for running and evaluating reinforcement learning agents in Harbor environments, particularly for mathematical problem-solving tasks.