MMODELYST
People/Jan Eitzinger
PPL

Jan Eitzinger

Papers (3)
Diagnosing Overhead in Dispatch Operations: Cross-architecture ObservatoryThe Illusion of Power Capping in LLM Decode: A Phase-Aware Energy Characterisation Across Attention ArchitecturesIrminsul: MLA-Native Position-Independent Caching for Agentic LLM Serving