AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

SEAP: Sparse Expert Activation Pruning Unlocks the Brainpower of Large Language Models

Xun Liang, Hanyu Wang, Huayi Lai, Simin Niu, Shichao Song, Jiawei Yang, Jihao Zhao, Feiyu Xiong, Bo Tang, Zhiyu Li

AAAI 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Layer-wise pruning of LLMs · 1 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
use SEAP (sF) instead of FLAP at 20% pruningLLaMA2-7B · 9 zero-shot tasksaccuracy avg (%): 52.81 -> 58.52Improved
use SEAP (sF) instead of FLAP at 50% pruningLLaMA2-7B · 9 zero-shot tasksaccuracy avg (%): 35.26 -> 46.98Improved
use SEAP (sF) instead of FLAP at 50% pruningLLaMA2-13B · 9 zero-shot tasksaccuracy avg (%): 39.68 -> 50.34Improved
use SEAP (sF) instead of FLAP at 20% pruningLLaMA2-7B · GSM8Kaccuracy (%): 1.81 -> 8.87Improved
use SEAP (sF) instead of FLAP at 20% pruningLLaMA2-7B · MBPPpass@1 (%): 1.20 -> 16.60Improved
add retention-based sparsity instead of uniform layerwise (ablation)SEAP (sF) on LLaMA2-7B · 9 zero-shot tasks, 20% pruningaccuracy avg (%): 56.54 -> 58.52Improved
add retention-based sparsity instead of uniform layerwise (ablation)SEAP (sF) on LLaMA2-7B · 9 zero-shot tasks, 50% pruningaccuracy avg (%): 43.17 -> 46.98Improved
add retention-based sparsity instead of adaptive layerwise (ablation)FLAP on LLaMA2-7B · 9 zero-shot tasks, 50% pruningaccuracy avg (%): 35.26 -> 38.88Improved
add retention-based sparsity instead of uniform layerwise (ablation)WandA-sp on LLaMA2-7B · 9 zero-shot tasks, 50% pruningaccuracy avg (%): 32.87 -> 34.50Improved
use WandA score (sW) instead of FLAP score (sF)SEAP on LLaMA2-7B · 9 zero-shot tasks, 20% pruningaccuracy avg (%): 58.52 -> 57.88No clear effect
use SEAP instead of FLAP for inference speed at 20% pruningLLaMA2-7B · inference throughputtokens/s: 38.90 -> 37.32Hurt
use SEAP instead of WandA for inference speed at 50% pruningLLaMA2-7B · inference throughputtokens/s: 31.24 -> 47.10Improved

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