UnifiedCrawl: Aggregated Common Crawl for Affordable Adaptation of LLMs on Low-Resource Languages

TL;DR

This work presents a method to efficiently collect text data for low-resource languages from the entire Common Crawl corpus and demonstrates that leveraging this data to fine-tuning multilingual LLMs via efficient adapter methods (QLoRA) significantly boosts performance on the low-resource language, while minimizing VRAM usage.

Abstract

Large language models (LLMs) under-perform on low-resource languages due to limited training data. We present a method to efficiently collect text data for low-resource languages from the entire Common Crawl corpus. Our approach, UnifiedCrawl, filters and extracts common crawl using minimal compute resources, yielding mono-lingual datasets much larger than previously available sources. We demonstrate that leveraging this data to fine-tuning multilingual LLMs via efficient adapter methods (QLoRA) significantly boosts performance on the low-resource language, while minimizing VRAM usage. Our experiments show large improvements in language modeling perplexity and an increase in few-shot prompting scores. Our work and released source code provide an affordable approach to improve LLMs for low-resource languages using consumer hardware. Our source code is available here at https://github.com/bethelmelesse/unifiedcrawl.

Figures (5)

• Figure 1: High-Level Overview of our Method
• Figure 2: Our Dataset is much Larger than all Prior Works
• Figure 3: UnifiedCrawl: Data Extraction Framework
• Figure 4: Distribution of Languages in Common Crawl except English
• Figure 5: Distribution of Languages in Common Crawl after Top 60