AI Paper Atlas

Research map / Architectures and efficiency / Quantization

GlowQ: Group-Shared LOw-Rank Approximation for Quantized LLMs

Selim An, Il hong Suh, Yeseong Kim

ICLR 2026 · Architectures and efficiency › Quantization · approach: Post-training quantization of LLMs · 0 citations (OpenAlex)

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
add covariance-aligned whitening to stacked SVD (ablation)GlowQ (group-shared) · C4, Qwen 3-8Bperplexity: 14.60 -> 13.40Improved
share right factor within input-sharing groups instead of layer-wise (ablation)GlowQ (whitened) · C4, Qwen 3-8Bperplexity: 13.85 -> 13.40Improved
cache shared projection per groupGlowQ vs layer-wise · LLaMA 2 latencyTTFB / throughput: TTFB -5.57%; throughput +9.61%Improved
restore only about half of units (selective restore) (ablation)GlowQ-S · LLaMA 2 latencyTTFB / throughput: TTFB -23.39%; throughput +37.44%Improved
restore only about half of units (selective restore) (ablation)GlowQ-S vs GlowQ · 4 models, 7 tasksavg accuracy / C4 ppl: -0.55% accuracy; +0.15 ppl C4Hurt
add GlowQ on top of GPTQGPTQ · WikiText-2 LLaMA 3.2-3Bperplexity: 9.32 -> 8.19Improved
add GlowQ on top of BnBBnB · WikiText-2 LLaMA 3.2-3Bperplexity: 8.29 -> 8.10Improved
use shared B instead of layer-wise correctionGlowQ · Qwen1.5-MoE-A2.7B WikiText-2perplexity: 7.39 (layerwise) vs 7.41; memory about 63% lowerNo clear effect
use GlowQ instead of L2QER at W4A4GlowQ · WikiText-2 LLaMA 3.2-3Bperplexity: 9.42 -> 9.21Improved
GlowQ vs baselines on perplexity across modelsGlowQ · WikiText-2 W4A16perplexity: best or tied-best on 9 of 11 variantsMixed

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