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

• URL: http://arxiv.org/abs/2609.05228v1
• Date: Fri, 04 Sep 2026 14:56:00 GMT
• Title: ACE: Adaptive Calibration-Free Expert Skipping for MoE-based LLMs
• Abstract summary: Mixture-of-Experts (MoE) architectures provide an efficient paradigm for scaling large language models (LLMs)<n>We propose ACE, a training-free, calibration-free, and checkpoint-preserving framework for token-adaptive expert skipping.
• Score: 15.24786135265898
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 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.

Related papers

• Expert-Space Exploration in MoE Reinforcement Learning [41.1632714379556]
  Expert-Space Exploration Reinforcement Learning (ESRL) is an architecture-aware framework that explicitly explores the expert-routing space of MoE models.<n>ESRL achieves the best performance across MoE backbones with top-K, top-1, and shared-expert routing, as well as across mathematics, science, and code tasks without additional sampling or computational cost.
  arXiv  Detail & Related papers  (2026-09-11T16:58:41Z)
• Share First, Route What Remains: A Unified Framework for Token-Adaptive MoE Computation [49.8459545291965]
  Mixture-of-experts (MoE) models have recently moved beyond routing a fixed number of complete experts.<n>We study this dependency by decomposing sparsely upcycled feed-forward experts into key-value channels.<n>We instantiate it in UniF-MoE, a unified framework for token-adaptive MoE computation.
  arXiv  Detail & Related papers  (2026-08-11T02:40:58Z)
• Tying the Loop -- Tied Expert Layers in Mixture-of-Experts Language Models [50.34585122347149]
  We introduce Expert Tying, an architectural modification that shares expert parameters across consecutive transformer layers.<n>We evaluate this approach across common, state-of-the-art architectures, including OLMoE, Qwen3, and DeepSeek-style MoEs.
  arXiv  Detail & Related papers  (2026-06-15T15:08:09Z)
• Adaptive Inverted-Index Routing for Granular Mixtures-of-Experts [49.09151538536423]
  Mixture-of-experts (MoE) models enable scalable transformer architectures by activating only a subset of experts per token.<n>Recent evidence suggests that performance improves with increasingly granular experts, i.e., many small experts instead of a few large ones.<n>We introduce adaptive inverted-index routing for MoE (AIR-MoE), an inverted-index-inspired routing architecture based on vector quantization (VQ)
  arXiv  Detail & Related papers  (2026-05-06T14:15:10Z)
• MoE-Sieve: Routing-Guided LoRA for Efficient MoE Fine-Tuning [0.5076419064097734]
  MoE-Sieve is a simple routing-guided framework for LoRA fine-tuning.<n>We show that tuning only the top 25% routed experts per layer remains competitive with full LoRA.<n>We also observe a non-monotonic relationship between expert count and seed-to-seed variance.
  arXiv  Detail & Related papers  (2026-03-25T07:53:09Z)
• REAL: Regression-Aware Reinforcement Learning for LLM-as-a-Judge [83.2858110368572]
  We propose textbfREAL (underlineREgression-underlineAware Reinforcement underlineLThought), a principled RL framework designed to optimize regression rewards.<n>We show that REAL consistently outperforms both regression-aware SFT baselines and standard RL methods.
  arXiv  Detail & Related papers  (2026-03-17T21:19:08Z)
• A Replicate-and-Quantize Strategy for Plug-and-Play Load Balancing of Sparse Mixture-of-Experts LLMs [64.8510381475827]
  Sparse Mixture-of-Experts (SMoE) architectures are increasingly used to scale large language models efficiently.<n>SMoE models often suffer from severe load imbalance across experts, where a small subset of experts receives most tokens while others are underutilized.<n>We present a systematic analysis of expert routing during inference and identify three findings: (i) load imbalance persists and worsens with larger batch sizes, (ii) selection frequency does not reliably reflect expert importance, and (iii) overall expert workload and importance can be estimated using a small calibration set.
  arXiv  Detail & Related papers  (2026-02-23T15:11:16Z)
• Too Helpful, Too Harmless, Too Honest or Just Right? [19.134202394422285]
  Large Language Models (LLMs) exhibit strong performance across a wide range of NLP tasks.<n> aligning their outputs with the principles of Helpfulness, Harmlessness, and Honesty (HHH) remains a persistent challenge.<n>We propose TrinityX, a modular alignment framework that incorporates a Mixture of Calibrated Experts (MoCaE) within the Transformer architecture.
  arXiv  Detail & Related papers  (2025-09-10T10:51:47Z)
• HOBBIT: A Mixed Precision Expert Offloading System for Fast MoE Inference [54.40808356999408]
  We present HOBBIT, a mixed precision expert offloading system to enable flexible and efficient MoE inference. Our key insight is that dynamically replacing less critical cache-miss experts with low precision versions can substantially reduce expert-loading latency. HOBBIT achieves up to a 9.93x speedup in decoding compared to state-of-the-art MoE offloading systems.
  arXiv  Detail & Related papers  (2024-11-03T04:25:46Z)
• Expert-Token Resonance MoE: Bidirectional Routing with Efficiency Affinity-Driven Active Selection [19.365009652356793]
  Expert-Token Resonance (ETR) is a theoretically-grounded bidirectional routing mechanism that reimagines expert-token interactions.<n>ETR achieves 5.4%-46.6% improvements in end-to-end training efficiency compared to baseline MoE implementations.
  arXiv  Detail & Related papers  (2024-05-24T02:50:44Z)
• T-REX: Mixture-of-Rank-One-Experts with Semantic-aware Intuition for Multi-task Large Language Model Finetuning [31.276142111455847]
  Large language models (LLMs) encounter significant adaptation challenges in diverse multitask finetuning.<n>We design a novel framework, mixunderlinetextbfTureunderlinetextbf-of-underlinetextbfRank-onunderlinetextbfE-eunderlinetextbfXper ts (textttT-REX)<n>Rank-1 experts enable a mix-and-match mechanism to quadratically expand the vector subspace of experts with linear parameter overheads, achieving approximate error reduction with optimal
  arXiv  Detail & Related papers  (2024-04-13T12:14:58Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.