This repository contains the dataset for our paper:
Unlocking Reasoning Potential in Large Language Models by Scaling Code-form Planning
In this work, we introduce CodePlan, a scalable framework that empowers LLMs to generate and follow code-form plans---pseudocode that outlines high-level, structured reasoning processes.
By leveraing the structured and versatile nature of code, CodePlan effectively captures the rich semantics and control flows inherent to sophisticated reasoning tasks.
huggingface dataset: link