Bench-CoE evaluates routing and collaboration among specialist language and multimodal experts using benchmark-derived training data.
unassessed
| Category | composite |
|---|---|
| Metric | task performance |
| Direction | higher_is_better |
| Unit | score |
| Publisher | Bench-CoE authors |
Bench-CoE studies whether a router can assign each query to suitable expert models and improve aggregate performance through collaboration. It covers language and multimodal tasks under query-level and subject-level routing settings.
A mixture of language and multimodal benchmark queries routed to specialist experts.
No model card in ModelSpec reports this benchmark yet.