AI Paper Atlas

Research map / Architectures and efficiency / Quantization

WinQ: Accelerating Quantization-Aware Training of Language Models Around Saddle Points

Dongyue Li, Zechun Liu, Kai Yi, Zhenshuo Zhang, Changsheng Zhao, Raghuraman Krishnamoorthi, Harshit Khaitan, Hongyang R. Zhang, Steven Li

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 noise injection (sigma=0.001) to WinQ (ablation)WinQ · LLaMA-1B W1A16 WikiText2perplexity: about 4% improvement over sigma=0Improved
add weight re-initialization (alpha>0) to WinQ (ablation)WinQ · LLaMA-1B W1A16 WikiText2perplexity: about 6.7% improvement over alpha=0Improved
add WinQ on top of ParetoQParetoQ · LLaMA-1B W1A16 WikiText2perplexity: 16.9 -> 15.3Improved
add WinQ on top of ParetoQParetoQ · LLaMA-1B W4A16 WikiText2perplexity: 10.3 -> 10.2No clear effect
add WinQ on top of ParetoQParetoQ · LLaMA-3B W1.58A8 accuracyaccuracy: 55.9 -> 58.6Improved
add WinQ on top of ParetoQ with 2x training budget (40B tokens)ParetoQ · LLaMA-1B W1A16 WikiText2perplexity: 16.2 -> 14.7Improved
add WinQ on top of ParetoQ for Llama-8BParetoQ · Llama-8B W2A16 zero-shot avgaccuracy: 64.7 -> 65.3Improved
increase noise sigma from 0.001 to 0.004 (ablation)WinQ · Llama-1B W1A16 batch 64 WikiText2perplexity: 15.3 -> 18.5Hurt
reduce batch size from 64 to 16 at sigma=0.001 (ablation)WinQ · Llama-1B W1A16 WikiText2perplexity: 15.3 -> 16.7Hurt
interpolate weights at alpha=0.4 before Hessian estimate (ablation)LLaMA-1B 2-bit QAT · max absolute Hessian eigenvaluemax abs eigenvalue: 1.64 -> 3.09Improved
add noise injection sigma=0.001 to 2-bit QAT (ablation)LLaMA-1B 2-bit QAT · max absolute Hessian eigenvalue at alpha=0.4max abs eigenvalue: 3.09 -> 3.95Improved
add WinQ to QuEST in 1-bit weight 1-bit activationQuEST · 30M LLaMA SlimPajamatest perplexity: 5% reduction in test perplexityImproved

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