analiza aiqbench
PRQuant introduces a training-free quantization framework that improves low-bit inference accuracy by reorganizing weight channels into contiguous blocks, thereby eliminating expensive online gathering operations during execution. This approach effectively mitigates outlier-induced errors in linear layers while maintaining hardware-friendly layouts that reduce latency compared to standard baselines like MXFP4. The method demonstrates consistent gains across multiple downstream benchmarks without introducing new accuracy bottlenecks associated with previous smoothing or rotation techniques.
PRQuant: Permutation Residual Quantization for Low-Overhead Inference
arXiv:2609.22106v1 Announce Type: new Abstract: Accuracy of Low-bit quantization of linear layers is often dominated by a small number of outliers. Although existing methods, such as smoothing, rotation, or residual-based approaches, may mitigate this problem, they often introduce new accuracy bottlenecks to weights. Besides, most of these techniques are implemented as online approaches, which can result in heavy execution overheads. To address the afore-mentioned issues, We propose PRQuant (Permutation Residual Quantization), a training-free and low-overhead framework that combines channel reorganization with static weight-side residual compensation. After AWQ-style scaling, PRQuant identifies the input channels that contribute most to weight quantization error, permutes them into contiguous tail blocks, and constructs their residual weight sub-tensors offline. During inference, this contiguous structure enables the activation side to use tail blocks seamlessly without the expensive online gathering operation, and turns scattered residual compensation into a regular tail-augmented GEMM, substantially reducing latency. Experiments demonstrate that PRQuant effectively reduces down-projection reconstruction error. Ablation studies confirm that smoothing and residual compensation are the primary drivers of numerical improvement, while permutation provides a consistent marginal numerical benefit and, more importantly, enables a hardware-friendly contiguous layout that eliminates dynamic gathering overhead. Overall, PRQuant outperforms default MXFP4 and the evaluated PTQ baselines in average accuracy across five downstream benchmarks, improving over MXFP4 by 1.24 and 0.55 on Qwen3-4B-Instruct-2507 and Qwen3-30B-A3B-Instruct-2507, respectively.
Abstract:Accuracy of Low-bit quantization of linear layers is often dominated by a small number of outliers. Although existing methods, such as smoothing, rotation, or residual-based approaches, may mitigate this problem, they often introduce new accuracy bottlenecks to weights. Besides, most of these techniques are implemented as online approaches, which can result in heavy execution overheads. To address the afore-mentioned issues, We propose PRQuant (Permutation Residual Quantization), a training-free and low-overhead framework that combines channel reorganization with static weight-side residual compensation. After AWQ-style scaling, PRQuant identifies the input channels that contribute most to weight quantization error, permutes them into contiguous tail blocks, and constructs their residual weight sub-tensors offline. During inference, this contiguous structure enables the activation side to use tail blocks seamlessly without the expensive online gathering operation, and turns scattered residual compensation into a regular tail-augmented GEMM, substantially reducing latency. Experiments demonstrate that PRQuant effectively reduces down-projection reconstruction error. Ablation studies confirm that smoothing and residual compensation are the primary drivers of numerical improvement, while permutation provides a consistent marginal numerical benefit and, more importantly, enables a hardware-friendly contiguous layout that eliminates dynamic gathering overhead. Overall, PRQuant outperforms default MXFP4 and the evaluated PTQ baselines in average accuracy across five downstream benchmarks, improving over MXFP4 by 1.24 and 0.55 on Qwen3-4B-Instruct-2507 and Qwen3-30B-A3B-Instruct-2507, respectively.
Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2609.22106 [cs.LG]
(or arXiv:2609.22106v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2609.22106 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.
Text integral sub licența deschisă a publicației · 313 cuvinte
Subiecte conexe
Optimizing ATOM and vLLM-ATOM for High-Interactivity Inference
AMD · 1 sept. 2026
Productionizing TurboQuant on AMD GPUs for KV-Cache-Bound LLM Inference
AMD · 11 iun. 2026
RDQ: Residual Distribution Quantization for Large Language Models
arXiv · 29 iul. 2026
PhaseQuant: Mitigating Activation Quantization Collapse via Residual-Stream Dynamics
arXiv · 14 oct. 2026