Advancing Polish Language Modeling through Tokenizer Optimization in the Bielik v3 7B and 11B Series

TL;DR

This paper presents a Polish language-specific tokenizer and training strategies for Bielik v3 models, improving efficiency and language understanding over universal tokenizers.

Contribution

It introduces a dedicated Polish tokenizer and optimized training procedures, enhancing model performance and efficiency for Polish language modeling.

Findings

Reduced fertility ratios with language-specific tokenizer

Improved inference costs and context window utilization

Enhanced language understanding in Polish models

Abstract

The development of the Bielik v3 PL series, encompassing both the 7B and 11B parameter variants, represents a significant milestone in the field of language-specific large language model (LLM) optimization. While general-purpose models often demonstrate impressive multilingual capabilities, they frequently suffer from a fundamental architectural inefficiency: the use of universal tokenizers. These tokenizers, typically designed to cover a broad spectrum of languages, often fail to capture the morphological nuances of specific languages like Polish, leading to higher fertility ratios, increased inference costs, and restricted effective context windows. This report details the transition from the universal Mistral-based tokenization to a dedicated Polish-optimized vocabulary for the Bielik v3 models, exploring the FOCUS-based embedding initialization, the multi-stage pretraining curriculum, and the subsequent post-training alignment involving Supervised Fine-Tuning, Direct Preference Optimization, and Reinforcement Learning through Group Relative Policy Optimization with verifiable rewards.