Scalable Prompt Routing via Fine-Grained Latent Task Discovery

Open paper
Benchmark Model Rank Results
arithmetic-reasoning-on-gsm8kFineRouter–Accuracy: 91.6
code-generation-on-humanevalFineRouter–Pass@1: 68.4
code-generation-on-mbppFineRouter–Accuracy: 90

Rank counts only results with a code link.