gretelai/gretel-math-gsm8k-v1

这是一个由Gretel Navigator和meta-llama/Meta-Llama-3.1-405B生成的合成数据集，灵感来源于GSM8K数据集。它包含了小学级别的推理任务，每个任务都有逐步的思考和解答过程，专注于多步骤推理问题。数据集的特点包括：使用Gretel Navigator进行数据生成，包含自动化输出验证和品质评估；有结构化的推理提示，详细记录了AI的决策过程；输出经过LLM-as-a-judge验证以确保质量和一致性；使用Python sympy库验证计算注解的准确性；数据覆盖多种现实世界情境，提供自然语言推理的逼真场景；应用上下文标签确保数据多样性，帮助模型泛化到不同类型的问题；问题难度分为四个级别，提供比原始gsm8k数据集更高的复杂性；包含训练集和测试集，测试集有1300个示例，按主题和难度分层。

This is a synthetically generated dataset inspired by the GSM8K dataset, created using Gretel Navigator with meta-llama/Meta-Llama-3.1-405B as the agent LLM. It contains Grade School-level reasoning tasks with step-by-step reflections and solutions, focusing on multi-step reasoning problems. Key features include synthetic data generation, structured reasoning prompts, LLM-as-a-judge evaluation, sympy library validation for calculation annotations, diverse real-world contexts, contextual tags for data diversity, four difficulty levels, and train & test sets with stratified topic and difficulty.