Reproduce “PW-Pruning: Phase-Aware Expert Selection for Speech-Centric Omni-Modal Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Junjie Cao, Fengping Tian, Wenjie Zhou, Xuanfan Ni, Chenyang Lyu, Siyou Liu, Longyue Wang, Weihua Luo
PW-Pruning selects a reduced expert bank for speech models processing audio and generating text while preserving separate listening and answering objectives. It builds phase-specific statistics and combines selection gains into one bank without inference-time switching. On Qwen3-Omni with 32 of 128 experts retained, it outperforms pooled selection on MMAU-mini and MMSU after matched fine-tuning, but not every metric.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.