Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2509.22944 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908797349396480 |
|---|---|
| author | Müller, Lorenz K. Bich, Philippe Zhuang, Jiawei Çelik, Ahmet Benfenati, Luca Cavigelli, Lukas |
| author_facet | Müller, Lorenz K. Bich, Philippe Zhuang, Jiawei Çelik, Ahmet Benfenati, Luca Cavigelli, Lukas |
| contents | Post-training quantization has emerged as the most widely used strategy for deploying large language models at low precision. Still, current methods show perplexity degradation at bit-widths less than or equal to 4, partly because representing outliers causes precision issues in parameters that share the same scales as these outliers. This problem is especially pronounced for calibration-free, uniform quantization methods. We introduce SINQ to augment existing post-training quantizers with an additional second-axis scale factor and a fast Sinkhorn-Knopp-style algorithm that finds scales to normalize per-row and per-column variances. We show that this approximates activation-aware quantization by recovering column scales from the weight matrix structure that are predictive of the typical activation magnitudes the matrix received during training. Our method has no interactions between layers and can be trivially applied to new architectures to quantize any linear layer. We evaluate our method on the Qwen3 model family, among others. SINQ reduces the perplexity gap on WikiText2 and C4 by over 50% against uncalibrated uniform quantization baselines, incurs zero to negligible compute overhead, and can be further enhanced by combining it with calibration and non-uniform quantization levels. Code is available at https://github.com/huawei-csl/SINQ. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_22944 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | SINQ: Sinkhorn-Normalized Quantization for Calibration-Free Low-Precision LLM Weights Müller, Lorenz K. Bich, Philippe Zhuang, Jiawei Çelik, Ahmet Benfenati, Luca Cavigelli, Lukas Machine Learning Post-training quantization has emerged as the most widely used strategy for deploying large language models at low precision. Still, current methods show perplexity degradation at bit-widths less than or equal to 4, partly because representing outliers causes precision issues in parameters that share the same scales as these outliers. This problem is especially pronounced for calibration-free, uniform quantization methods. We introduce SINQ to augment existing post-training quantizers with an additional second-axis scale factor and a fast Sinkhorn-Knopp-style algorithm that finds scales to normalize per-row and per-column variances. We show that this approximates activation-aware quantization by recovering column scales from the weight matrix structure that are predictive of the typical activation magnitudes the matrix received during training. Our method has no interactions between layers and can be trivially applied to new architectures to quantize any linear layer. We evaluate our method on the Qwen3 model family, among others. SINQ reduces the perplexity gap on WikiText2 and C4 by over 50% against uncalibrated uniform quantization baselines, incurs zero to negligible compute overhead, and can be further enhanced by combining it with calibration and non-uniform quantization levels. Code is available at https://github.com/huawei-csl/SINQ. |
| title | SINQ: Sinkhorn-Normalized Quantization for Calibration-Free Low-Precision LLM Weights |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2509.22944 |