![[2602.07954] Bielik Guard: Efficient Polish Language Safety Classifiers for LLM Content Moderation](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.07954] Bielik Guard: Efficient Polish Language Safety Classifiers for LLM Content Moderation
Summary
The Bielik Guard presents efficient Polish language classifiers for moderating content in large language models, achieving high precision and low false positive rates across various safety categories.
Why It Matters
As LLMs are increasingly utilized in Polish applications, the development of effective content moderation tools is crucial to ensure user safety. Bielik Guard addresses this need by providing classifiers that not only block harmful content but also offer appropriate responses, enhancing user experience and safety.
Key Takeaways
- Bielik Guard includes two model variants optimized for Polish language content moderation.
- The 0.5B model variant achieves high F1 scores, indicating strong performance in safety classification.
- The 0.1B variant excels in efficiency, with superior precision and low false positive rates.
- Both models are publicly available and designed to provide nuanced responses rather than simple content blocking.
- The classifiers cover critical safety categories including Hate/Aggression, Vulgarities, and Self-Harm.
Computer Science > Computation and Language arXiv:2602.07954 (cs) [Submitted on 8 Feb 2026 (v1), last revised 13 Feb 2026 (this version, v3)] Title:Bielik Guard: Efficient Polish Language Safety Classifiers for LLM Content Moderation Authors:Krzysztof Wróbel, Jan Maria Kowalski, Jerzy Surma, Igor Ciuciura, Maciej Szymański View a PDF of the paper titled Bielik Guard: Efficient Polish Language Safety Classifiers for LLM Content Moderation, by Krzysztof Wr\'obel and 4 other authors View PDF Abstract:As Large Language Models (LLMs) become increasingly deployed in Polish language applications, the need for efficient and accurate content safety classifiers has become paramount. We present Bielik Guard, a family of compact Polish language safety classifiers comprising two model variants: a 0.1B parameter model based on MMLW-RoBERTa-base and a 0.5B parameter model based on PKOBP/polish-roberta-8k. Fine-tuned on a community-annotated dataset of 6,885 Polish texts, these models classify content across five safety categories: Hate/Aggression, Vulgarities, Sexual Content, Crime, and Self-Harm. Our evaluation demonstrates that both models achieve strong performance on multiple benchmarks. The 0.5B variant offers the best overall discrimination capability with F1 scores of 0.791 (micro) and 0.785 (macro) on the test set, while the 0.1B variant demonstrates exceptional efficiency. Notably, Bielik Guard 0.1B v1.1 achieves superior precision (77.65%) and very low false positive rate (0.63...
