LLM Routing on Big Math In-Domain
0.6618AUROCProbeDirichlet
Evaluation results
| Method | Links | |
|---|---|---|
| ProbeDirichletsignal_modality=Hidden-state2026.02 | 0.6618 | |
| ConfidenceMarginsignal_modality=Logit-based2026.02 | 0.5618 | |
| EmbeddingMLPsignal_modality=Embedding-based2026.02 | 0.5618 | |
| SemanticEntropysignal_modality=Verbose-based2026.02 | 0.5581 | |
| Entropysignal_modality=Logit-based2026.02 | 0.5141 | |
| MaxLogitssignal_modality=Logit-based2026.02 | 0.4739 | |
| SelfAsksignal_modality=Verbose-based2026.02 | 0.472 |