LLM360/guru-RL-92k-extra-info-compressed

Guru是一个包含六个不同推理密集型领域（数学、编程、科学、逻辑、模拟和表格推理）的精选数据集，旨在为大型语言模型（LLM）提供用于复杂推理任务和强化学习（RL）训练的高质量样本。该数据集通过五阶段的精选流程进行质量保证，并包含特定领域的奖励函数以便于可靠评估。

Guru is a curated dataset across six reasoning-intensive domains (math, coding, science, logic, simulation, and tabular reasoning) designed for large language models (LLM) to perform complex reasoning tasks and reinforcement learning (RL) training. It contains high-quality samples that have gone through a five-stage curation pipeline for quality assurance and includes domain-specific reward functions for reliable evaluation.