mixture-of-expertsLLM servingpublished Nov 5, 2026 · arXiv:2611.06147
SafeStart: Mitigating the Overlooked Safety Cost of Active Expert Reduction in MoE LLM Serving
Mixture-of-experts modelsmodel safetypublished Oct 5, 2026 · arXiv:2610.02910
Frequency Is Not Sensitivity Identifying Safety-Sensitive Experts in Sparse MoE LLM
Mixture-of-Expertsattention-aware routingpublished Sep 21, 2026 · arXiv:2609.20974
Attention-Aware Routing: Coupling Routing and Attention in MoEs
LLM pruningcompression statisticspublished Aug 5, 2026 · arXiv:2608.02940
When Compression Scores Cannot Decide: Information Boundaries for Group-Robust LLM Pruning