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

Bethel Melesse Tessema; Akhil Kedia; Tae-Sun Chung

返回文献探索

Paper · arXiv 2411.14343

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

Bethel Melesse Tessema, Akhil Kedia, Tae-Sun Chung

7 upvotesNovember 21, 2024arXiv 预印本

AI 摘要

UnifiedCrawl efficiently collects data from Common Crawl to fine-tune multilingual LLMs using adapter methods, enhancing performance on low-resource languages.

Large language modelsCommon CrawlUnifiedCrawlmono-lingual datasetsmultilingual LLMsefficient adapter methodsQLoRAVRAM usagelanguage modeling perplexityfew-shot prompting scores

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.