AI Paper Atlas

Research map / Architectures and efficiency / Quantization

WaterSIC: Information-Theoretically (Near) Optimal \\Linear Layer Quantization

Egor Lifar, Semyon Savkin, Or Ordentlich, Yury Polyanskiy

ICML 2026 · Architectures and efficiency › Quantization · approach: Post-training weight quantization · 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 post-quantization finetuning of rescalers (ablation)WaterSIC (Llama-3.2-1B, 1.00 bits) · WikiText-2WikiText-2 perplexity: 82.44 -> 30.47Improved
add post-quantization finetuning of rescalers (ablation)WaterSIC (Llama-3.2-1B, 1.50 bits) · WikiText-2WikiText-2 perplexity: 30.73 -> 18.50Improved
add post-quantization finetuning of rescalers (ablation)WaterSIC (Llama-3.2-1B, 2.00 bits) · WikiText-2WikiText-2 perplexity: 16.19 -> 13.59Improved
add post-quantization finetuning of rescalers (ablation)WaterSIC (Llama-3.2-1B, 4.00 bits) · WikiText-2WikiText-2 perplexity: 9.92 -> 9.91Improved
add post-quantization finetuning of rescalers (ablation)WaterSIC (Qwen3-8B, 2.125 bits) · WikiText-2WikiText-2 perplexity: 11.37 -> 10.18Improved
add post-quantization finetuning of rescalers (ablation)WaterSIC (Qwen3-8B, 4.125 bits) · WikiText-2WikiText-2 perplexity: 9.79 -> 9.67Improved
use WaterSIC instead of Huffman-GPTQLlama-3.2-1B at about 2 bits · WikiText-2WikiText-2 perplexity: 86.80 -> 16.19Improved
use WaterSIC instead of Huffman-GPTQQwen3-8B at 2.125 bits · WikiText-2WikiText-2 perplexity: 13.97 -> 11.37Improved
use WaterSIC instead of GPTQQwen3-8B at 3.125 bits · WikiText-2WikiText-2 perplexity: 12.77 -> 10.03Improved
use WaterSIC instead of QTIPLlama-3.2-1B at about 3 bits · WikiText-2WikiText-2 perplexity: 11.17 -> 10.57Improved
use WaterSIC instead of AWQ (g128)Llama-3.2-1B at about 3.2 bits · WikiText-2WikiText-2 perplexity: 16.74 -> 10.35Improved

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