AI Paper Atlas

Research map / Architectures and efficiency / Quantization

OSAQ: Outlier Self-Absorption for Accurate Low-bit LLM Quantization

Zhikai Li, Zhen Dong, Xuewen Liu, Jing Zhang, Qingyi Gu

ICML 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 OSAQ additive transformation before GPTQ at 2-bitGPTQ · WikiText2 W2A16 g128, LLaMA2-7Bperplexity: 36.8 -> 21.2Improved
add coordinate descent iterations on top of OSAQ+GPTQ at 2-bitOSAQ+GPTQ · WikiText2 W2A16 g128, LLaMA2-7Bperplexity: 21.2 -> 10.6Improved
add OSAQ before GPTQ at 3-bit per-channelGPTQ · WikiText2 W3A16, LLaMA2-13Bperplexity: 6.44 -> 5.72Improved
add OSAQ before GPTQ at 4-bitGPTQ · WikiText2 W4A16, LLaMA2-7Bperplexity: 5.83 -> 5.73Improved
add OSAQ before AWQAWQ · WikiText2 W3A16 g128, LLaMA2-7Bperplexity: 6.24 -> 6.08Improved
add OSAQ before QuIPQuIP · WikiText2 W3A16, LLaMA2-7Bperplexity: 6.50 -> 6.37Improved
add OSAQ before GPTQ at 3-bitGPTQ · Commonsense QA average, LLaMA3-8B, W3A16 g128zero-shot accuracy: 63.6 -> 65.2Improved
add OSAQ before GPTQ at 4-bitGPTQ · MMLU, LLaMA2-7B and LLaMA3-8B, W4A16average accuracy: 38.7 -> 39.6; 57.6 -> 58.2Improved
add OSAQ before GPTQ at 3-bit on MMLUGPTQ · MMLU, LLaMA3-8B, W3A16average accuracy: 30.0 -> 30.9No clear effect
add OSAQ before GPTQ on very large instruction-tuned modelsGPTQ · Mistral-Large-123B-Instruct W4A16, MMLU averageaccuracy: 82.0 -> 82.3Improved
apply additive transformation to the FP16 model without quantization (ablation)LLaMA2-7B FP16 · WikiText2perplexity: 5.47 -> 5.52No clear effect
use Softmax-inf approximation instead of plain l2 norm objective (ablation)OSAQ+GPTQ · WikiText2 W3A16, LLaMA2-7Bperplexity: 7.82 -> 6.75Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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