aiqbench analysis
This research addresses the critical intersection of model efficiency and ethical deployment by introducing Fair-GPTQ, a quantization technique that explicitly minimizes bias during the compression of large language models. By integrating group-fairness constraints into the rounding objective, the method successfully reduces discriminatory outputs related to gender, race, and religion while maintaining over 90% of baseline accuracy. The work demonstrates that optimizing for fairness is not mutually exclusive with the memory and speed benefits of low-precision quantization, offering a practical path toward more equitable AI systems.
Fair-GPTQ: Bias-Aware Quantization for Large Language Models
arXiv:2509.15206v4 Announce Type: replace Abstract: The high memory demands of generative language models have drawn attention to quantization, which reduces memory usage by mapping model weights to lower-precision integers. However, recent empirical studies show that, while efficient, quantization can increase the likelihood of generating biased outputs and degrade performance on fairness benchmarks. In this work, we draw new links between quantization and model fairness by adding explicit group-fairness constraints to the quantization objective and introduce Fair-GPTQ, the first quantization method explicitly designed to reduce unfairness in large language models. The added constraints guide the learning of the rounding operation toward less-biased text generation for protected groups. Specifically, we focus on stereotype generation involving occupational bias and discriminatory language spanning gender, race, and religion. Fair-GPTQ has minimal impact on performance, preserving at least 90% of baseline accuracy on zero-shot benchmarks, reduces unfairness relative to a half-precision model, and retains the memory and speed benefits of 4-bit quantization.
Abstract:The high memory demands of generative language models have drawn attention to quantization, which reduces memory usage by mapping model weights to lower-precision integers. However, recent empirical studies show that, while efficient, quantization can increase the likelihood of generating biased outputs and degrade performance on fairness benchmarks. In this work, we draw new links between quantization and model fairness by adding explicit group-fairness constraints to the quantization objective and introduce Fair-GPTQ, the first quantization method explicitly designed to reduce unfairness in large language models. The added constraints guide the learning of the rounding operation toward less-biased text generation for protected groups. Specifically, we focus on stereotype generation involving occupational bias and discriminatory language spanning gender, race, and religion. Fair-GPTQ has minimal impact on performance, preserving at least 90% of baseline accuracy on zero-shot benchmarks, reduces unfairness relative to a half-precision model, and retains the memory and speed benefits of 4-bit quantization.
Accepted for publication in TACL. Pre-MIT Press publication version
Computation and Language (cs.CL)
arXiv:2509.15206 [cs.CL]
(or arXiv:2509.15206v4 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2509.15206 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 253 words
Related coverage
Podcast: The Future of AI: from Enterprise Adoption to Open Source Sovereignty
InfoQ · Sep 25, 2026
Quantization and Pruning Methods to Make Your LLM Leaner
KDnuggets · Aug 28, 2026
Jev: System One models for Prod, not God — with Diogo Almeida, CEO, TypeSafe AI
Latent Space · Sep 21, 2026
Debias-SparseGPT: Bias-Aware Pruning for Large Language Models
arXiv · Sep 3, 2026