MUTANT: A Recipe for Multilingual Tokenizer Design

TL;DR

MUTANT introduces a comprehensive approach for designing multilingual tokenizers that improve efficiency and performance across diverse languages, with a focus on Indian languages, achieving state-of-the-art results.

Contribution

The paper presents MUTANT, a novel recipe for multilingual tokenizer design incorporating language-aware pre-tokenization and training strategies, and introduces MUTANT-Indic for Indian languages.

Findings

39.5% reduction in fertility score over LLaMA4

44% increase in inference throughput over LLaMA4

State-of-the-art performance on Indian languages and code data

Abstract

Tokenizers play a crucial role in determining the performance, training efficiency, and the inference cost of Large Language Models (LLMs). Designing effective tokenizers for multilingual LLMs is particularly challenging due to diverse scripts and rich morphological variation. While subword methods like Byte Pair Encoding (BPE) are widely adopted, their effectiveness in multilingual settings remains underexplored. We present MUTANT, a recipe for building multilingual tokenizers, with careful vocabulary and training data design, language-aware pre-tokenization, and subword and multiword aware training. We also introduce MUTANT-Indic, a tokenizer for India-specific multilingual LLMs, that produces linguistically coherent tokens and achieves state-of-the-art performance. Evaluated across English, 22 Indian languages and code data, our tokenizer improves the average fertility score by 39.5%$ over LLaMA4 and by 18% over Sutra (the current best). This translates to 44% improvement in inference throughput over LLaMA4 while maintaining comparable performance on English and Indic benchmarks. We present detailed ablations across tokenizer training data size, vocabulary size, merging techniques, and pre-tokenization strategies, demonstrating the robustness of our design choices.