Research map / Architectures and efficiency / Mixture of experts
RaGEP: Rank-aware Geometric Expert Pruning for Mixture-of-Experts Language Models
Wentao Hu, Zeyu Zhu, Mingkuan Zhao, Zhenhua An, Yanbo Zhai, Shanhong yu, Huilin Zhou, Xin Lai, Xiaoyan Zhu, Jiayin Wang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Add rank-aware budget allocation (ablation) | EEP · Qwen3-30B-A3B, retain 50% | average accuracy: 61.60 -> 61.96 | Improved |
| Add spectral-salience pruning (ablation) | EEP · Qwen3-30B-A3B, retain 50% | average accuracy: 61.60 -> 62.18 | Improved |
| Combine rank-aware budget allocation and SSP (ablation) | EEP · Qwen3-30B-A3B, retain 50% | average accuracy: 61.60 -> 63.12 | Improved |
| Add rank-aware budget allocation (ablation) | EEP · Qwen3-30B-A3B, retain 75% | average accuracy: 66.96 -> 67.53 | Improved |
| Add spectral-salience pruning (ablation) | EEP · Qwen3-30B-A3B, retain 75% | average accuracy: 66.96 -> 67.30 | Improved |
| Combine both components (ablation) | EEP · Qwen3-30B-A3B, retain 75% | average accuracy: 66.96 -> 67.72 | Improved |
| Add rank-aware budget allocation (MMLU only) (ablation) | EEP · Qwen3-30B-A3B MMLU, retain 50% | accuracy: 55.67 -> 53.12 | Hurt |
| Add spectral-salience pruning (MMLU only) (ablation) | EEP · Qwen3-30B-A3B MMLU, retain 50% | accuracy: 55.67 -> 50.80 | Hurt |
| Raise balancing factor beta from 0.0 to 0.8 (ablation) | RaGEP · DeepSeek-V2-Lite | average accuracy | Improved |
| Raise beta from 0.8 to 1.0 (pure salience) (ablation) | RaGEP · DeepSeek-V2-Lite | average accuracy | Hurt |
| Replace EEP/MoP selection with RaGEP | EEP · DeepSeek-V2-Lite, 50% pruning | average accuracy: 45.70 -> 47.33 | Improved |
| Replace EEP with RaGEP | EEP · Mixtral 8x7B, 50% pruning | average accuracy: 62.86 -> 63.83 | Improved |
Closest papers by meaning
- DiEP: Adaptive Mixture-of-Experts Compression through Differentiable Expert Pruning (NeurIPS 2025 · similarity 0.91)
- Domain-Specific Pruning of Large Mixture-of-Experts Models with Few-shot Demonstrations (NeurIPS 2025 · similarity 0.90)
- Pruning and Distilling Mixture-of-Experts into Dense Language Models (NeurIPS 2026 · similarity 0.90)
- A Provably Effective Method for Pruning Experts in Fine-tuned Sparse Mixture-of-Experts (ICML 2024 · similarity 0.89)
- FlexMoE: One-for-All Nested Intra-Expert Pruning for MoE Language Models (NeurIPS 2026 · similarity 0.89)
- Advancing Expert Specialization for Better MoE (NeurIPS 2025 · similarity 0.88)
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.
