SafeStart: Mitigating the Overlooked Safety Cost of Active Expert Reduction in MoE LLM Serving
cs.AI, cs.LG
Submitted: 2026-11-04
Updated: 2026-11-04
Code: https://github.com/ggml-org/llama.cpp
Terminology
Sources
- When Keywords Drop but Classifiers Hold: Soft Refusals under KV Cache Compression
- Training-Free Halving of Activated Experts in Fine-Grained Mixture-of-Experts Models
- You Only Need 2/3 of the Chosen Experts: An Empirical Study of Dynamic Expert Pruning in Fine-Grained MoE LLMs
- LExI: Layer-Adaptive Active Experts for Efficient MoE Model Inference
- Training Verifiers to Solve Math Word Problems
- DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
- Lynx: Enabling Efficient MoE Inference through Dynamic Batch-Aware Expert Selection
- Buffer Overflow in Mixture of Experts
- Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations
- Every FLOP Counts: Scaling a 300B Mixture-of-Experts LING LLM without Premium GPUs
- Muon is Scalable for LLM Training
- AnchorKV: Safety-Aware KV Cache Compression via Soft Penalty with a Refusal Anchor
- gpt-oss-120b & gpt-oss-20b Model Card
- The Joint Effect of Quantization and Sampling Temperature on LLM Safety Alignment: A Factorial Analysis
- Large Language Lobotomy: Jailbreaking Mixture-of-Experts via Expert Silencing
- XShare: Collaborative in-Batch Expert Sharing for Faster MoE Inference
- Alignment Collapse Under KV Cache Quantization: Diagnosis and Mitigation
- RouteHijack: Routing-Aware Attack on Mixture-of-Experts LLMs
- ACE: Adaptive Calibration-Free Expert Skipping for MoE-based LLMs
- Qwen2 Technical Report
Related papers
- Multi-Objective Structured Pruning of LLMs for Latency and Model Size Optimization
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Embeddings for Preferences, Not Semantics
- Can We Trust Item Response Theory for AI Evaluation?
- Diagnosing Task Insensitivity in Language Agents