Beyond Retraining-Free MoE Compression: A Cost-Normalized Study of Post-Compression Adjustment

arXiv:2609.06076 · cs.LG · Submitted 2026-09-05 · Read on arXiv

cs.LG

Submitted: 2026-09-05

Updated: 2026-09-29

Comments: Accepted to EMNLP 2026 (Main Conference)

Code: https://github.com/AIDASLab/Post-Compression-Adjustment

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Fine-Tuning — 2× in this paper · explained in 5 episodes
• Pruning — 2× in this paper · explained in 4 episodes
• Benchmark — 1× in this paper · explained in 2 episodes

Sources

• Program Synthesis with Large Language Models
• MoBE: Mixture-of-Basis-Experts for Compressing MoE-based LLMs
• Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge
• Training Verifiers to Solve Math Word Problems
• Domain-Specific Pruning of Large Mixture-of-Experts Models with Few-shot Demonstrations
• ToxiGen: A Large-Scale Machine-Generated Dataset for Adversarial and Implicit Hate Speech Detection
• Towards Efficient Mixture of Experts: A Holistic Study of Compression Techniques
• Distilling the Knowledge in a Neural Network
• Is Retraining-Free Enough? The Necessity of Router Calibration for Efficient MoE Compression
• Finding Fantastic Experts in MoEs: A Unified Study for Expert Dropping Strategies and Observations
• Scaling Laws for Neural Language Models
• Dynin-Omni: Omnimodal Unified Large Diffusion Language Model
• REAP the Experts: Why Pruning Prevails for One-Shot MoE compression
• GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding
• Sub-MoE: Efficient Mixture-of-Expert LLMs Compression via Subspace Expert Merging
• The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning
• Efficient Expert Pruning for Sparse Mixture-of-Experts Language Models: Enhancing Performance and Reducing Inference Costs
• MoLAE: Mixture of Latent Experts for Parameter-Efficient Language Models
• AIMER: Calibration-Free Task-Agnostic MoE Expert Pruning
• Merging Models with Fisher-Weighted Averaging

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks