Improving Multilingual Math Reasoning for African Languages

TL;DR

This paper systematically evaluates various adaptation strategies for extending large language models to African languages, focusing on mathematical reasoning tasks, to identify the most effective methods for low-resource language settings.

Contribution

It provides a comprehensive analysis of data types, training stages, and adaptation configurations for improving LLM performance on African languages.

Findings

Pre-training with translated data improves reasoning accuracy.

Post-training strategies enhance low-resource language adaptation.

Optimal combination of data and training stages yields best results.

Abstract

Researchers working on low-resource languages face persistent challenges due to limited data availability and restricted access to computational resources. Although most large language models (LLMs) are predominantly trained in high-resource languages, adapting them to low-resource contexts, particularly African languages, requires specialized techniques. Several strategies have emerged for adapting models to low-resource languages in todays LLM landscape, defined by multi-stage pre-training and post-training paradigms. However, the most effective approaches remain uncertain. This work systematically investigates which adaptation strategies yield the best performance when extending existing LLMs to African languages. We conduct extensive experiments and ablation studies to evaluate different combinations of data types (translated versus synthetically generated), training stages (pre-training versus post-training), and other model adaptation configurations. Our experiments focuses on mathematical reasoning tasks, using the Llama 3.1 model family as our base model.