ISER-MoE: Dynamic In-Switch Replication of Hot Experts for Load Balancing in Expert-Parallel
cs.DC
Submitted: 2026-11-06
Updated: 2026-11-06
Code: https://github.com/MoonshotAI/MoonEP
Project page: https://artificialanalysis.ai/methodology/agentperf
Terminology
Sources
- Training Verifiers to Solve Math Word Problems
- DeepSeek-V3 Technical Report
- DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence
- Demystifying NCCL: An In-depth Analysis of GPU Communication Protocols and Algorithms
- Kimi K3: Open Frontier Intelligence
- FEPLB: Exploiting Copy Engines for Nearly Free MoE Load Balancing in Distributed Training
- SciCode: A Research Coding Benchmark Curated by Scientists
- Blink: Fast and Generic Collectives for Distributed ML
- UltraEP: Unleash MoE Training and Inference on Rack-Scale Nodes with Near-Optimal Load Balancing
- Scalable Training of Mixture-of-Experts Models with Megatron Core
- Efficient MoE Serving in the Memory-Bound Regime: Balance Activated Experts, Not Tokens
- TAOT: Topology-Aware Optimal Transport for Dynamic Expert Replica Placement in MoE Training
- Accelerating MoE with Dynamic In-Switch Computing on Multi-GPUs
- Harnessing Routing Foresight for Micro-step-level MoE load balancing in RL Post-training
- PROBE: Co-Balancing Computation and Communication in MoE Inference via Real-Time Predictive Prefetching
Related papers
- Programmable Probabilistic Computer with 1,000,000 p-bits
- BatchGen: An Architecture for Scalable and Efficient Batch Inference
- Seamless Reconfiguration for DAG-BFT (Technical Report)
- ELDR: Expert-Locality-Aware Decode Routing for PD-Disaggregated MoE Serving
- iScheduler: Reinforcement Learning-Driven Continual Optimization for Large-Scale Resource Investment Problems
- TIRx: A Unified Tile Primitive ML Compiler Across Heterogeneous Backends