AI Paper Atlas

Research map / Architectures and efficiency / Quantization

SINQ: Sinkhorn-Normalized Quantization for Calibration-Free Low-Precision LLM Weights

Lorenz K Muller, Philippe Bich, Jiawei Zhuang, Ahmet Çelik, Luca Benfenati, Lukas Cavigelli

ICML 2026 · Architectures and efficiency › Quantization · approach: Post-training quantization of LLMs

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace HQQ/RTN single-scale with SINQ dual-scale Sinkhorn normalization (3-bit)Qwen3-1.7B · WikiText2 (3-bit)perplexity: 32.10 -> 22.39Improved
replace RTN with SINQ (4-bit)Qwen3-32B · WikiText2 (4-bit)perplexity: 8.92 -> 7.74Improved
replace Hadamard + RTN with SINQ (3-bit)Qwen3-32B · WikiText2 (3-bit)perplexity: 11.26 -> 8.79Improved
add SINQ normalization to AWQ (A-SINQ) at 3-bit vs GPTQQwen3-14B · Avg flip rate (3-bit)flip rate (%): 8.06 -> 7.56Improved
add SINQ to uncalibrated 3-bit quantizationQwen3-14B · Avg flip rate (3-bit)flip rate (%): 10.07 -> 7.73Improved
use NF4 levels in SINQ instead of uniform INT4SINQ · WikiText2 (4-bit)perplexity: 8.76 -> 8.72Mixed
remove the shift (zero-point) from SINQ (ablation)SINQ · Memory-perplexity Pareto, Qwen3 familyperplexityImproved
vary auxiliary variable precision (int8 vs fp16) (ablation)SINQ · Memory-perplexity Pareto, Qwen3 familyperplexityNo clear effect
use no-overhead SINQ (shared second scale absorbed in prior layer) instead of standard SINQ (ablation)SINQ · WikiText2 (4-bit)perplexity: 17.14 -> 17.63Hurt
apply no-overhead SINQ before GGUF Q3_K_SQwen3-0.6B · WikiText2 (GGUF)perplexity: 35.59 -> 30.20Improved
add SINQ before AWQ scaling (ablation)AWQ · Row-wise kurtosis, Qwen3-1.7Bkurtosis: 1.32x reductionImproved
add the extra input-side scale vector at inferenceSINQ with gemlite kernel · Batch size 1, D=1024latency overhead: 1.8%No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy