AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

The Unseen Frontier: Pushing the Limits of LLM Sparsity with Surrogate-Free ADMM

Kwanhee Lee, Hyeondo Jang, Dongyeop Lee, Dan Alistarh, Namhoon Lee

ICLR 2026 · Large language models › Language, safety and interpretability · approach: LLM pruning and scaling data · 5 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
Replace layer-wise surrogate pruners with surrogate-free ADMM (ELSA)SparseGPT/ALPS · LLaMA-2-7B, 90% sparsityC4 perplexity: 180.9 (ALPS) -> 23.14Improved
Use ELSA instead of Wanda + full retrainingWanda+Full · LLaMA-2-7B, 99% sparsityC4 perplexity: 71.64 -> 40.10Improved
Use ELSA instead of Wanda + LoRAWanda+LoRA · LLaMA-2-7B, 95% sparsityWikiText perplexity: 371.0 -> 38.91Improved
Add objective-aware projection (ablation)ELSA · LLaMA-3.2-3B, 90% sparsityperplexity: 65.41 -> 48.69Improved
Add objective-aware projection (ablation)ELSA · LLaMA-3.2-3B, 70% sparsityperplexity: 29.44 -> 28.24Improved
Use ELSA's uniform allocation instead of EvoPress non-uniform mask (ablation)ELSA (EvoPress mask) · LLaMA-3-8B, 70% sparsityC4 perplexity: 29.33 -> 29.09Improved
Use ELSA instead of non-uniform allocation methodsOWL / EvoPress · LLaMA-3-8B, 70% sparsityC4 perplexity: 33.72 (EvoPress) -> 29.09Improved
Use ELSA under N:M semi-structured sparsity (2:4)ALPS / L-ADMM · LLaMA-2-7B 2:4zero-shot avg accuracy: 53.85 (ALPS) vs 49.59Hurt
Use next-token-prediction objective instead of reconstruction error with more data (ablation)reconstruction-error minimization · OPT-125M, 90% sparsityperplexityMixed
Raise sparsity from 50% to 95% with MACKO kerneldense LLaMA-2-7B · LLaMA-2-7B on RTX 3090throughput: 54.47 -> 216.98 tokens/sImproved

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