Unveiling the Impact of Coding Data Instruction Fine-Tuning on Large Language Models Reasoning

TL;DR

This study investigates how coding data used during instruction fine-tuning enhances the reasoning abilities of large language models across various domains, model sizes, and tasks, revealing consistent benefits and domain-dependent effects.

Contribution

It provides a comprehensive analysis of the impact of coding data in instruction fine-tuning on LLM reasoning, covering multiple models, domains, and task-specific outcomes.

Findings

Coding data tuning improves overall reasoning capabilities.

Impact varies by domain but shows consistent trends within each domain.

Optimal coding data proportions are task-dependent.

Abstract

Instruction Fine-Tuning (IFT) significantly enhances the zero-shot capabilities of pretrained Large Language Models (LLMs). While coding data is known to boost LLM reasoning abilities during pretraining, its role in activating internal reasoning capacities during IFT remains understudied. This paper investigates a key question: How does coding data impact LLMs' reasoning capacities during IFT stage? To explore this, we thoroughly examine the impact of coding data across different coding data proportions, model families, sizes, and reasoning domains, from various perspectives. Specifically, we create three IFT datasets with increasing coding data proportions, fine-tune six LLM backbones across different families and scales on these datasets, evaluate the tuned models' performance across twelve tasks in three reasoning domains, and analyze the outcomes from three broad-to-granular perspectives: overall, domain-level, and task-specific. Our holistic analysis provides valuable insights into each perspective. First, coding data tuning enhances the overall reasoning capabilities of LLMs across different model families and scales. Moreover, while the impact of coding data varies by domain, it shows consistent trends within each domain across different model families and scales. Additionally, coding data generally provides comparable task-specific benefits across model families, with optimal proportions in IFT datasets being task-dependent.