AnalysisAI Models
7 days ago
DLLG: Dynamic Logit-Level Gating of LLM Experts
A new method dynamically combines multiple LLMs at the logit level to improve performance without premature routing or heuristic ensembling. The approach aims to balance adaptability and stability.
·
7 days ago