ScalingIntelligence/tpt-gemma2-2b-gsm8k-2k

这是一个包含2000个GSM8K风格数学问题的合成示例的数据集，适用于数学推理研究。每个示例包括一个小学水平的数学文字问题、模型的完整推理轨迹和最终答案。数据集分为训练集和测试集，可用于微调或评估数学文字问题模型，研究自我改进的LLM训练循环，分析思维链推理质量，以及合成数据策展和训练动态研究。

This dataset contains 2,000 synthetic examples of GSM8K-style math problems, suitable for research in math reasoning. Each example includes a grade-school level math word problem, the models full reasoning trace, and the final answer. The dataset is split into training and test sets and can be used for fine-tuning or evaluating models on math word problems, studying self-improving LLM training loops, analyzing the quality of chain-of-thought reasoning, and researching synthetic data curation and training dynamics.