AI RESEARCH
DLLG: Dynamic Logit-Level Gating of LLM Experts
arXiv CS.CL
•
ArXi:2606.04378v1 Announce Type: new Leveraging multiple specialized LLMs can combine complementary strengths, but existing approaches trade adaptability for stability: routing commits prematurely, heuristic ensembling depends on fragile proxies, and parameter merging