2026-09-27 · America/Los_Angeles · 论文 · #9

JevSoup: System-One Routing for Training-Free LoRA Composition

Building adaptable AI systems requires effective coordination of specialized capabilities across diverse tasks. Low-rank adaptation (LoRA) enables modular expertise, but existing routing approaches may require auxiliary data, additional training, or autoregressive decoding. We propose JevSoup, a training-free framework separating System One expert routing from System Two execution. Using only the input and expert descriptions, Jev selects two experts through structured probabilities. JevSoup retains the leading expert's update, projects the second onto the orthogonal complement of the first…

热度 14.0 / 100；排名与评分保留该期记录。