Investigating the Impact of Quantization Methods on the Safety and Reliability of Large Language Models
Artyom Kharinaev, Viktor Moskvoretskii, Egor Shvetsov, Kseniia Studenikina, Bykov Mikhail, Evgeny Burnaev
2025-02-25
Summary
This paper talks about how quantization, a method for shrinking large AI models to make them run faster and use less memory, affects their safety and reliability when performing tasks.
What's the problem?
Large language models are powerful but require a lot of computing resources, which makes them expensive and hard to use on smaller devices. Quantization is a way to make these models smaller and more efficient, but it’s unclear how shrinking them impacts their ability to produce safe and trustworthy results.
What's the solution?
The researchers created OpenSafetyMini, a dataset designed to test the safety of quantized models. They evaluated four different quantization techniques on two popular AI models, LLaMA and Mistral, using benchmarks and human evaluations. Their experiments showed that different methods work better depending on the precision level, with vector quantization performing best at very low precision (2-bit).
Why it matters?
This matters because it helps us understand how to make AI models smaller and cheaper without sacrificing their ability to produce safe and reliable results. This could lead to more accessible AI technology that works well even on devices with limited resources, while ensuring that the models remain trustworthy for important applications.
Abstract
Large Language Models (LLMs) have emerged as powerful tools for addressing modern challenges and enabling practical applications. However, their computational expense remains a significant barrier to widespread adoption. Quantization has emerged as a promising technique to democratize access and enable low resource device deployment. Despite these advancements, the safety and trustworthiness of quantized models remain underexplored, as prior studies often overlook contemporary architectures and rely on overly simplistic benchmarks and evaluations. To address this gap, we introduce OpenSafetyMini, a novel open-ended safety dataset designed to better distinguish between models. We evaluate 4 state-of-the-art <PRE_TAG>quantization techniques</POST_TAG> across LLaMA and Mistral models using 4 benchmarks, including human evaluations. Our findings reveal that the optimal quantization method varies for 4-bit precision, while vector quantization techniques deliver the best safety and trustworthiness performance at 2-bit precision, providing foundation for future research.