How Quantization Shapes Bias in Large Language Models

Federico Marcuzzi, Xuefei Ning, Roy Schwartz, Iryna Gurevych
8/25/2025
cs.CLcs.LG

No official code found. Best implementation: insait-institute/quantization-affects-social-bias (7 stars, Apache-2.0, last commit 2026-04-14).

Abstract

This work presents a comprehensive evaluation of how quantization affects model bias, with particular attention to its impact on individual demographic subgroups. We focus on weight and activation quantization strategies and examine their effects across a broad range of bias types, including stereotypes, toxicity, sentiment, and fairness. We employ both probabilistic and generated text-based metrics across nine benchmarks and evaluate models varying in architecture family and reasoning ability. Our findings show that quantization has a nuanced impact on bias: while it can reduce model toxicity and does not significantly impact sentiment, it tends to slightly increase stereotypes and unfairness in generative tasks, especially under aggressive compression. These trends are generally consistent across demographic categories and model types, although their magnitude depends on the specific setting. Overall, our results highlight the importance of carefully balancing efficiency and ethical considerations when applying quantization in practice.

View on arXivView PDF

Code Implementations(1)

insait-institute/quantization-affects-social-biasCommunity64%
Apache-2.0

Code for the EACL 2026 paper: How Quantization Shapes Bias in Large Language Models. We examine how quantization influences social bias in LLMs across various demographic subgroups, model architectures, and reasoning capabilities.

73Shell, PythonAug 21, 20256 months agoApache-2.0

Similar Papers

Uncertainty Drives Social Bias Changes in Quantized Large Language Models

91% match2026

No Free Lunch in Language Model Bias Mitigation? Targeted Bias Reduction Can Exacerbate Unmitigated LLM Biases

87% match2025

Fair-GPTQ: Bias-Aware Quantization for Large Language Models

87% match2025

From n-gram to Attention: How Model Architectures Learn and Propagate Bias in Language Modeling

87% match2025

You Never Know: Quantization Induces Inconsistent Biases in Vision-Language Foundation Models

86% match2024

Textual Data Bias Detection and Mitigation -- An Extensible Pipeline with Experimental Evaluation

85% match2025

Cite this paper

@article{marcuzzi2025quantization,
  title  = {How Quantization Shapes Bias in Large Language Models},
  author = {Federico Marcuzzi and Xuefei Ning and Roy Schwartz and Iryna Gurevych},
  year   = {2025},
  eprint = {2508.18088},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2508.18088v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2508.18088.svg)](https://codeofpaper.com/paper/2508.18088)

Discussion