AI Paper Atlas

Research map / Architectures and efficiency / Quantization

S-Quant: Rethinking Weight Quantization with Seed-Based Generation

Mingzi Wang, Lancheng Zou, Shuo Yin, Zhuolun He, Bei Yu

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
use S-Quant instead of SeedLM at about 3 bitsSeedLM · WikiText-2 Llama 3 8Bperplexity: 10.1 -> 9.7Improved
use S-Quant instead of AWQ at 3 bitsAWQ · WikiText-2 Llama 2 7Bperplexity: 15.6 -> 6.5Improved
use S-Quant instead of SeedLM at about 4 bitsSeedLM · WikiText-2 Llama 3 70Bperplexity: 3.8 -> 3.6Improved
use S-Quant instead of SeedLM at about 4 bitsSeedLM · WikiText-2 Llama 2 7Bperplexity: 5.7 -> 5.7No clear effect
use S-Quant instead of SeedLM at 3 bitsSeedLM · zero-shot mean Llama 2 7Bmean accuracy: 64.48 -> 64.56No clear effect
use S-Quant instead of AWQ at 3 bitsAWQ · zero-shot mean Llama 2 7Bmean accuracy: 52.46 -> 64.56Improved
use S-Quant instead of SeedLM at about 4 bitsSeedLM · zero-shot mean Llama 3 70Bmean accuracy: 78.06 -> 78.11No clear effect
use S-Quant at 3.8 bit vs FP16FP16 baseline · zero-shot mean Llama 2 7Bmean accuracy: 68.74 -> 67.65Hurt
use adaptive basis count per block via explained energy ratio (ablation)S-Quant · LLaMA 2-7B blocksexplained energy ratioImproved
vary block size, seed length and explained-energy threshold (ablation)S-Quant · WikiText-2 LLaMA2-7BperplexityMixed
use S-Quant ASIC accelerator instead of uncompressed weightsuncompressed baseline · memory-bound LLM inferencelatency: approximately 4x reduction in latencyImproved

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