English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Chain-of-thought (CoT) prompting with large language models has proveneffective in numerous natural language processing tasks, but designing promptsthat generalize well to diverse problem types can be challenging, especially inthe context of math word problem (MWP) solving. Additionally, it is common tohave a large amount of training data that have a better diversity coverage butCoT annotations are not available, which limits the use of supervised learningtechniques. To address these issues, we investigate two approaches to leveragethe training data in a few-shot prompting scenario: dynamic program promptingand program distillation. Our approach is largely inspired by Gao et al.,(2022), where they proposed to replace the CoT with the programs as theintermediate reasoning step. Such a prompting strategy allows us to accuratelyverify the answer correctness through program execution in MWP solving. Ourdynamic program prompting involves annotating the training data by samplingcorrect programs from a large language model, while program distillationinvolves adapting a smaller model to the program-annotated training data. Ourexperiments on three standard MWP datasets demonstrate the effectiveness ofthese approaches, yielding significant improvements over previous baselines forprompting and fine-tuning. Our results suggest that leveraging a large amountof training data can improve the generalization ability of prompts and boostthe performance of fine-tuned small models in MWP solving.

Leveraging Training Data in Few-Shot Prompting for Numerical Reasoning