AS-ES Learning: Towards Efficient CoT Learning in Small Models
This addresses the challenge of enabling efficient CoT learning in small models for logical reasoning applications, representing an incremental improvement over existing distillation methods.
The paper tackles the problem of inefficiently using existing Chain-of-Thought (CoT) data to train small models for logical reasoning tasks, proposing AS-ES learning to surpass direct seq2seq training on tasks like MWP and PET summarization without data augmentation.
Chain-of-Thought (CoT) serves as a critical emerging ability in LLMs, especially when it comes to logical reasoning. Attempts have been made to induce such ability in small models as well by distilling from the data with CoT generated by Large Language Models (LLMs). However, existing methods often simply generate and incorporate more data from LLMs and fail to note the importance of efficiently utilizing existing CoT data. We here propose a new training paradigm AS-ES (Abstractive Segments - Extractive Segments) learning, which exploits the inherent information in CoT for iterative generation. Experiments show that our methods surpass the direct seq2seq training on CoT-extensive tasks like MWP and PET summarization, without data augmentation or altering the model itself. Furthermore, we explore the reason behind the inefficiency of small models in learning CoT and provide an explanation of why AS-ES learning works, giving insights into the underlying mechanism of CoT.