Search for a command to run...
When to Ensemble: Identifying Token-Level Points for Stable and Fast LLM Ensembling