AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Rethinking Residual Errors in Compensation-based LLM Quantization

Shuaiting Li, Juncan Deng, Kedong Xu, Rongtao Deng, Hong Gu, Minghan Jiang, Haibin Shen, Kejie Huang

ICLR 2026 · Architectures and efficiency › Quantization · approach: Post-training quantization of LLMs · 4 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 compensation-aware error to GPTQGPTQ · Llama-2-7B 3-bit C4perplexity: 13.60 -> 8.34Improved
add compensation-aware error to GPTAQGPTAQ · Llama-3.1-8B-Instruct 3-bitaverage accuracy: 70.3 -> 72.3Improved
add compensation-aware error to GPTAQGPTAQ · Llama-3-70B 3-bitaverage accuracy: 74.7 -> 74.4No clear effect
add compensation-aware error to GPTAQ (W2A16)QuaRot+GPTAQ · Llama-2-13Baverage accuracy: 55.8 -> 58.2Improved
add compensation-aware error to GPTQ (W2A16) (ablation)GPTQ · Llama-2-13Baverage accuracy: 50.5 -> 54.3Improved
add compensation-aware error to GPTAQ (W2A16) (ablation)GPTAQ · Llama-2-7Baverage accuracy: 51.5 -> 54.0Improved
add compensation-aware error to SpinQuant+GPTAQSpinQuant+GPTAQ · Llama-2-13B W2A4KV4WikiText-2 perplexity: 9.55 -> 8.60Improved
add compensation-aware error to SpinQuant+GPTAQSpinQuant+GPTAQ · Llama-3-70B W2A4KV4WikiText-2 perplexity: 5004 -> 3282No clear effect
add compensation-aware error to GPTAQGPTAQ · DeiT-Base W2A4 ImageNetaccuracy: 61.3 -> 62.1Improved
add compensation-aware error to GPTAQGPTAQ · DeiT-Base W4A4 ImageNetaccuracy: 78.1 -> 78.0No clear effect
add compensation-aware errorGPTAQ · Quantization time Llama-2-7Bseconds: 952 -> 1001Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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