AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

LSA: Layer-wise Sparsity Allocation for Large Language Model Pruning Based on Minimal Linear Reconstruction Error

Zhiguo Yang, Changjian Deng, Qinke Chen, Zijing Zhou, Jian Cheng

ICLR 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Layer-wise pruning of LLMs

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
use LSA allocation instead of uniform sparsityWanda on LLaMA1-7B · WikiText, 70% sparsityperplexity: 83.45 -> 20.66Improved
use LSA allocation instead of DLPWanda on LLaMA2-13B · WikiText, 70% sparsityperplexity: 16.80 -> 15.36Improved
use LSA allocation instead of DLPSparseGPT on LLaMA2-13B · WikiText, 70% sparsityperplexity: 13.39 -> 12.56Improved
use LSA allocation instead of OWL/DLP on zero-shotWanda on LLaMA1-7B · 7 zero-shot tasks, 70% sparsitymean accuracy: DLP 44.31 vs LSA 44.19No clear effect
use LSA instead of uniform for Magnitude pruningMagnitude on LLaMA2-7B · 7 zero-shot tasks, 70% sparsitymean accuracy: 33.66 -> 39.17Improved
add block-wise granularity to LSALSA Wanda LLaMA3-8B · WikiText, 70% sparsityperplexity: 98.75 -> 80.25Improved
add block-wise granularity to OWLOWL Wanda LLaMA1-7B · WikiText, 70% sparsityperplexity: 24.85 -> 57.91Hurt
add block-wise granularity to DLPDLP Wanda LLaMA1-7B · WikiText, 70% sparsityperplexity: 20.89 -> 40.81Hurt
add block-wise granularity to LSALSA Wanda LLaMA1-7B · WikiText, 70% sparsityperplexity: 20.66 -> 21.60Hurt
add projection-wise granularity to LSALSA Wanda LLaMA1-7B · WikiText, 70% sparsityperplexity: 20.66 -> 24.82Hurt
add LoRA fine-tuning to pruned modelLSA Wanda LLaMA1-7B · WikiText, 70% sparsityperplexity: 20.66 -> 12.27Improved
add LoRA fine-tuning to pruned modelLSA Wanda LLaMA2-7B · WikiText, 70% sparsityperplexity: 22.89 -> 11.75Improved

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