AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Learning Semi-Structured Sparsity for LLMs via Shared and Context-Aware Hypernetwork

Lu Sun, Jun Sakuma

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
add layer and component context-aware embeddings to the shared hypernetwork (ablation)HyperPrune · Wikitext-2 (LLaMA-2 7B, 2:4)perplexity: 11.36 -> 10.11Improved
add feature outlier regularization (ablation)HyperPrune · Wikitext-2 (LLaMA-2 7B, 2:4)perplexity: 11.06 -> 10.11Improved
add continual pruning regularization (ablation)HyperPrune · Wikitext-2 (LLaMA-2 7B, 2:4)perplexityImproved
add layer embeddings vs component embeddings (ablation)HyperPrune · Wikitext-2 (LLaMA-2 7B, 2:4)perplexityMixed
replace SparseGPT with HyperPruneLLaMA-2 7B 2:4 pruning · Zero-shot mean of 7 tasksmean accuracy: 51.00 -> 53.76Improved
replace MaskPro with HyperPruneLLaMA-2 7B 2:4 pruning · Zero-shot mean of 7 tasksmean accuracy: 52.81 -> 53.76Improved
replace MaskPro with HyperPruneLLaMA-2 7B 2:4 pruning · Wikitext-2perplexity: 12.29 -> 10.11Improved
replace MaskPro with HyperPruneLLaMA-2 13B 2:4 pruning · Zero-shot mean of 7 tasksmean accuracy: 58.97 -> 59.25Improved
replace Pruner-Zero with HyperPruneLLaMA-2 13B 2:4 pruning · Wikitext-2perplexity: 7.53 -> 7.60Hurt
replace Pruner-Zero with HyperPruneLLaMA-2 70B 2:4 pruning · Wikitext-2perplexity: 4.87 -> 5.13Hurt
replace Pruner-Zero with HyperPruneLLaMA-2 70B 2:4 pruning · Zero-shot mean of 7 tasksmean accuracy: 67.69 -> 68.57Improved
prune LLaMA-2-7B with HyperPrune 2:4 sparsity instead of denseLLaMA-2 7B · End-to-end latency on A100latency (ms): 248 ms -> 174 ms (1.43x)Improved

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