arXiv cs.CL· Junjie Cao, Fengping Tian, Wenjie Zhou, Xuanfan Ni, Chenyang Lyu, Siyou Liu, Longyue Wang, Weihua Luo·· 12 小时前AI 评分22
PW-Pruning：面向语音中心全模态模型的阶段感知专家选择方法
PW-Pruning: Phase-Aware Expert Selection for Speech-Centric Omni-Modal Models
AI 导读
PW-Pruning 提出一种阶段感知的专家剪枝方法，为声学预填充与教师强制答案 token 分别构建重要性加权核，再结合阶段加权对数行列式边际增益选出单一静态专家库。在 Qwen3-Omni 上每层保留 128 个专家中的 32 个，经监督微调后 MMAU-mini 与 MMSU 准确率较池化选择分别提升 3.20 和 4.38 个百分点（三次运行平均），且推理时无需切换 mask。
正文 · 原文
Abstract:Sparse routing reduces per-token computation but leaves the full expert bank resident in memory. We study expert pruning for speech-centric omni-modal models, where pooling calibration statistics can obscure the different expert demands of acoustic-input processing and text-answer generation. We propose PW-Pruning, a phase-aware extension of DO-ACP that constructs separate importance-weighted kernels for acoustic-prefill and teacher-forced answer tokens, then combines phase-weighted log-determinant marginal gains to select one static expert bank. On Qwen3-Omni, retaining 32 of 128 experts per layer improves MMAU-mini and MMSU accuracy over matched pooled selection by 3.20 and 4.38 percentage points, respectively, after supervised fine-tuning, averaged over three runs. Single-phase and shuffled-label controls on this public backbone suggest that semantic phase information contributes to these gains. Experiments on a non-public Qwen3.5-Omni backbone and English-only zero-shot evaluation of Nemotron provide complementary evidence. Across the tested Qwen budgets, AU gains are larger under stronger pruning, without inference-time mask switching.
| Comments: | 5 pages, 3 figures, 4 tables |
| Subjects: | Audio and Speech Processing (eess.AS); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Sound (cs.SD) |
| Cite as: | arXiv:2610.26891 [eess.AS] |
| (or arXiv:2610.26891v1 [eess.AS] for this version) | |
| https://doi.org/10.48550/arXiv.2610.26891 arXiv-issued DOI via DataCite |
Submission history
From: Junjie Cao [view email]
[v1]
Thu, 15 Oct 2026 12:36:40 UTC (25 KB)
来源：arXiv cs.CL · arxiv.org