Cover Image

Stabilizing MoE Reinforcement Learning by Aligning Training and Inference Routers

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Ma, Wenhan, Zhang, Hailin, Zhao, Liang, Song, Yifan, Wang, Yudong, Sui, Zhifang, Luo, Fuli
Format:	Preprint
Published:	2025
Subjects:	Computation and Language Artificial Intelligence Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Turn Waste into Worth: Rectifying Top-$k$ Router of MoE
by: Zeng, Zhiyuan, et al.
Published: (2024)

Enhancing Reliability across Short and Long-Form QA via Reinforcement Learning
by: Wang, Yudong, et al.
Published: (2025)

OmniMoE: An Efficient MoE by Orchestrating Atomic Experts at Scale
by: Shi, Jingze, et al.
Published: (2026)

AquilaMoE: Efficient Training for MoE Models with Scale-Up and Scale-Out Strategies
by: Zhang, Bo-Wen, et al.
Published: (2024)

MoE-Pruner: Pruning Mixture-of-Experts Large Language Model using the Hints from Its Router
by: Xie, Yanyue, et al.
Published: (2024)

GW-MoE: Resolving Uncertainty in MoE Router with Global Workspace Theory
by: Wu, Haoze, et al.
Published: (2024)

Skywork-MoE: A Deep Dive into Training Techniques for Mixture-of-Experts Language Models
by: Wei, Tianwen, et al.
Published: (2024)

Uni-MoE: Scaling Unified Multimodal LLMs with Mixture of Experts
by: Li, Yunxin, et al.
Published: (2024)

LLaDA-MoE: A Sparse MoE Diffusion Language Model
by: Zhu, Fengqi, et al.
Published: (2025)

MoE-SpAc: Efficient MoE Inference Based on Speculative Activation Utility in Heterogeneous Edge Scenarios
by: Li, Shuhuai, et al.
Published: (2026)

Sigma-MoE-Tiny Technical Report
by: Hu, Qingguo, et al.
Published: (2025)

LocMoE: A Low-Overhead MoE for Large Language Model Training
by: Li, Jing, et al.
Published: (2024)

MoRAL: MoE Augmented LoRA for LLMs' Lifelong Learning
by: Yang, Shu, et al.
Published: (2024)

Elastic MoE: Unlocking the Inference-Time Scalability of Mixture-of-Experts
by: Gu, Naibin, et al.
Published: (2025)

Uni-MoE-2.0-Omni: Scaling Language-Centric Omnimodal Large Model with Advanced MoE, Training and Data
by: Li, Yunxin, et al.
Published: (2025)

GRIN: GRadient-INformed MoE
by: Liu, Liyuan, et al.
Published: (2024)

Mix-MoE: Improving Multilingual Machine Translation of Large Language Models through Mixed MoEs
by: Li, Bo, et al.
Published: (2026)

Marco-MoE: Open Multilingual Mixture-of-Expert Language Models with Efficient Upcycling
by: Jiang, Fan, et al.
Published: (2026)

Post-Trained MoE Can Skip Half Experts via Self-Distillation
by: Lv, Xingtai, et al.
Published: (2026)

Mix- and MoE-DPO: A Variational Inference Approach to Direct Preference Optimization
by: Bohne, Jason, et al.
Published: (2025)

MoE-CT: A Novel Approach For Large Language Models Training With Resistance To Catastrophic Forgetting
by: Li, Tianhao, et al.
Published: (2024)

ShardMemo: Masked MoE Routing for Sharded Agentic LLM Memory
by: Zhao, Yang, et al.
Published: (2026)

xRouter: Training Cost-Aware LLMs Orchestration System via Reinforcement Learning
by: Qian, Cheng, et al.
Published: (2025)

Diagnosing Training Inference Mismatch in LLM Reinforcement Learning
by: Zhong, Tianle, et al.
Published: (2026)

Chain-of-Thought Tokens are Computer Program Variables
by: Zhu, Fangwei, et al.
Published: (2025)

Alloc-MoE: Budget-Aware Expert Activation Allocation for Efficient Mixture-of-Experts Inference
by: Liu, Baihui, et al.
Published: (2026)

MoEs Are Stronger than You Think: Hyper-Parallel Inference Scaling with RoE
by: Zibakhsh, Soheil, et al.
Published: (2025)

Prediction Is All MoE Needs: Expert Load Distribution Goes from Fluctuating to Stabilizing
by: Cong, Peizhuang, et al.
Published: (2024)

A Pluggable Multi-Task Learning Framework for Sentiment-Aware Financial Relation Extraction
by: Luo, Jinming, et al.
Published: (2025)

MESA: Improving MoE Safety Alignment via Decentralized Expertise
by: Sun, Yitong, et al.
Published: (2026)

PithTrain: A Compact and Agent-Native MoE Training System
by: Lai, Ruihang, et al.
Published: (2026)

Is Retraining-Free Enough? The Necessity of Router Calibration for Efficient MoE Compression
by: Hyeon, Sieun, et al.
Published: (2026)

Towards an empirical understanding of MoE design choices
by: Fan, Dongyang, et al.
Published: (2024)

OrcaRouter: A Production-Oriented LLM Router with Hybrid Offline-Online Learning
by: Bao, Zhenghua, et al.
Published: (2026)

Hunyuan-Large: An Open-Source MoE Model with 52 Billion Activated Parameters by Tencent
by: Sun, Xingwu, et al.
Published: (2024)

Evaluating Expert Contributions in a MoE LLM for Quiz-Based Tasks
by: Chernov, Andrei
Published: (2025)

Performance Characterization of Expert Router for Scalable LLM Inference
by: Pichlmeier, Josef, et al.
Published: (2024)

HySparse: A Hybrid Sparse Attention Architecture with Oracle Token Selection and KV Cache Sharing
by: Gao, Yizhao, et al.
Published: (2026)

Leave It to the Experts: Detecting Knowledge Distillation via MoE Expert Signatures
by: Li, Pingzhi, et al.
Published: (2025)

JudgeRLVR: Judge First, Generate Second for Efficient Reasoning
by: Duo, Jiangshan, et al.
Published: (2026)