First-Step Advantage: Importance of Starting Right in Multi-Step Math Reasoning
Abstract: Language models can solve complex reasoning tasks better by learning to generate rationales for their predictions. Often these models know how to solve a task but their auto-regressive decoding nature leads to incorrect results if they start incorrectly.
We observe that smaller models in particular, when corrected, can solve a task that they would have otherwise struggled with. We demonstrate this phenomenon by using a larger model to guide smaller models, which leads to significantly improved performance (up to $\texttt{+24}$ points on the GSM8K dataset by 7B models). Furthermore, to assist smaller models in initiating the first step correctly, we propose QuestCoT, where the smaller model first _asks itself how to start_ , before proceeding with a chain of reasoning.
On various multistep mathematical reasoning datasets over multiple smaller models, we show that getting the right start can lead to significant performance gains across all models (gains of up to $\texttt{+6}$ points on GSM8K, $\texttt{+9}$ on SVAMP, $\texttt{+5}$ on ASDiv, and $\texttt{+7}$ on MultiArith).
Paper Type: Long
Research Area: Language Modeling
Research Area Keywords: Reasoning, Smaller Language Models
Contribution Types: Model analysis & interpretability, NLP engineering experiment
Languages Studied: English
Submission Number: 175
Loading