Preprint
Sep 2026
Analytical Resource Management for Fine-grained MoE Computation-Communication Overlap
A wave-quantized analytical model and launch-time resource manager for dependency-coupled overlap pipelines that selects the communication-CTA count and resource partition before each launch without candidate execution, per-workload profiling, or kernel recompilation is presented.
Hong-Yu Liu, Min-Yu Cui, Miquel Pericàs
· 0 citations