cs.CVNov 16, 2026

ProSE: Probing Reconstruction Error for Semantic-guided Expert Skipping in MoE-MLLMs

Authors: Jiameng Li, Minye Wu, Matthew B. Blaschko

Organizations: KU Leuven · University of Derby

Abstract

Mixture-of-Experts Multimodal Large Language Models (MoE-MLLMs) offer scalable inference through sparse expert activation, yet substantial computational redundancy remains among the activated experts. Existing expert skipping methods typically calibrate thresholds based on target skipping ratios or final-output divergence, overlooking the accumulation of skipping-induced error across layers and the varying semantic importance of individual tokens. To address that, we propose ProSE, a training-free, semantic-guided expert skipping framework that probes layer-wise reconstruction error to promote representation consistency compared with default activation. Specifically, we construct query-aware semantic masks to identify salient visual tokens and progressively calibrate modality-specific skipping thresholds for each MoE layer. By minimizing masked reconstruction error under a given budget, ProSE maintains semantic consistency after skipping and allows adaptive expert allocation under limited activation budgets. Extensive experiments on 9 benchmarks demonstrate consistent improvements on average over existing expert skipping methods, particularly under aggressive skipping settings. ProSE maintains over 96% performance on KimiVL-A3B-Inst with 83% expert skipping and accelerates prefilling by 1.7×1.7\times.

Figures & tables

Explore similar work

Sep 4, 2026

ACE: Adaptive Calibration-Free Expert Skipping for MoE-based LLMs

Mixture-of-Experts (MoE) architectures provide an efficient paradigm for scaling large language models (LLMs), yet fixed top-k routing activates the same number of expert slots for every token, causing substantial redundant computation. Existing expert-skipping methods often rely on router confidence, calibration data, or additional training, and therefore cannot reliably estimate the actual contribution of routed experts. To this end, we propose ACE, a training-free, calibration-free, and checkpoint-preserving framework for token-adaptive expert skipping in MoE-based LLMs. ACE contains two complementary components: 1) Global Spectral Proxy (GSP), which estimates global transformation capacity from the coupled gate, up, and down projections together with RMSNorm scaling; and 2) Router-Conditioned Refinement (RCR), which constructs expert-specific direction prototypes from centered router weights and evaluates expert responses along routing-preferred directions. During inference, ACE combines both estimates with runtime router gates and skips an expert slot only when both views identify it as low-contribution, while always retaining the top-1 expert. All expert statistics are computed offline, leaving only table lookups and lightweight scalar operations online. Extensive experiments across three MoE-based LLMs and eight benchmarks demonstrate that ACE consistently outperforms existing static and dynamic baselines, with increasingly pronounced advantages under aggressive expert skipping. For instance, at a 50% skipping ratio on Qwen3.6-35B-A3B, ACE reduces WikiText-2 perplexity by 7.96% and improves average downstream accuracy by 4.15 percentage points over the strongest competing method.
Nov 4, 2026

Modality-Aware Expert Pruning for MoE-Based Multimodal Large Language Models

Mixture-of-Experts (MoE) enables multimodal large language models (MLLMs) to scale capacity efficiently, but expert parameters dominate memory even though only a few experts are activated per token: they account for over 90% of model size in modern MoE-MLLMs (e.g., 54 GB out of about 58 GB in BF16 for InternVL3.5-30B-A3B and Qwen3-VL-30B-A3B). Expert pruning has emerged as an effective compression strategy, but existing methods aggregate importance uniformly across tokens, ignoring that the same expert may serve different roles for vision versus language processing. To address this, we propose MAEP (Modality-Aware Expert Pruning), a training-free framework that decomposes expert importance by modality in a single forward pass. MAEP introduces (1) Delta-H Output (ΔHΔH), a redistribution-aware metric measuring the MoE output change when an expert is removed, and (2) low-attention vision tokens as a vision-side pruning signal that is dataset-stable: its expert-importance ranking remains stable across different datasets. Text-side importance serves as a preservation signal while vision-side importance serves as a pruning signal, and experts are ranked globally across layers. Experiments on three MoE-based MLLMs across multimodal and text benchmarks show that MAEP outperforms modality-agnostic baselines on the majority of multimodal benchmarks while remaining competitive on text-only tasks, reducing the BF16 expert weight footprint by about 27 GB (about 47% of total memory) so that the full weight footprint of an approximately 58 GB model fits within a single 40 GB GPU. Code is available at https://github.com/nota-github/MAEP.
Oct 20, 2026

Don't Skip the Baseline: A Controlled Comparison of Dynamic Expert Skipping in MoEs

A growing body of work proposes dynamic expert-skipping methods for Mixture-of-Experts (MoE) models that adjust the number of experts activated per token during inference, aiming to preserve model quality while further reducing the number of active parameters. These methods have added calibrated importance scores, inter-expert similarity metrics, or learned per-token allocators in order to identify experts to skip. However, different calibration regimes, different reductions in the effective number of experts, and evaluation suites make comparisons difficult. We classify four representative prior methods into a unified taxonomy and benchmark them against a natural, simple baseline: statically reducing the number of experts kk (i.e., same reduced kk across all layers and tokens). We provide a fair comparison of these methods at matched expert-compute budget, on two modern MoEs (Gemma4-26B-A4B and Qwen3-30B-A3B) in a text-only setting and a contemporary vision-language model (Qwen3-VL) in a multimodal setting. Surprisingly, we find that the simple baseline is hard to outperform consistently across models, modalities, benchmarks, and calibration settings. The methods we benchmark often underperform the baseline and achieved gains, when present, are minor. We further find that the effective kk achieved (or optimized for) in calibration can diverge from the effective kk in testing, resulting in over- or under-estimating the compute at inference time. Based on these findings, we recommend a calibration-free per-modality static kk reduction and validate this on a held-out model and modality. We further argue that a static control at matched measured compute is the baseline any dynamic skipping method must beat.