Krutrim LLM: A Novel Tokenization Strategy for Multilingual Indic Languages with Petabyte-Scale Data Processing

TL;DR

This paper introduces Krutrim LLM, a multilingual Indic language model with a novel tokenization strategy, developed through meticulous data collection, deduplication, and custom tokenizer training, resulting in improved performance over existing tokenizers.

Contribution

The paper presents a new tokenization method and data processing pipeline specifically designed for multilingual Indic languages, enhancing model performance.

Findings

Custom Indic tokenizer outperforms OpenAI Tiktoken in token-to-word ratio.

Deduplication reduces redundancy in web crawl data by 70%.

High-quality, diverse data improves multilingual Indic language modeling.

Abstract

We present a novel approach to data preparation for developing multilingual Indic large language model. Our meticulous data acquisition spans open-source and proprietary sources, including Common Crawl, Indic books, news articles, and Wikipedia, ensuring a diverse and rich linguistic representation. For each Indic language, we design a custom preprocessing pipeline to effectively eliminate redundant and low-quality text content. Additionally, we perform deduplication on Common Crawl data to address the redundancy present in 70% of the crawled web pages. This study focuses on developing high-quality data, optimizing tokenization for our multilingual dataset for Indic large language models with 3B and 7B parameters, engineered for superior performance in Indic languages. We introduce a novel multilingual tokenizer training strategy, demonstrating our custom-trained Indic tokenizer outperforms the state-of-the-art OpenAI Tiktoken tokenizer, achieving a superior token-to-word ratio for Indic languages.