The paper addresses set-based language model routing, where the goal is to minimize inference calls while ensuring at least one selected model answers correctly.
CopRoute reduces inference calls by exploiting complementary failures among language models
A conformal routing policy that orders models by marginal coverage certifies all-miss risk while using fewer calls than existing risk-controlled methods.
Independent
Siva Rajesh Kasa · Santhosh Kumar Kasa · Karan Gupta
Research Digest··3 min read
The authors propose CopRoute, a dependence-aware conformal routing family that builds model sets sequentially by ordering models on how many new queries each answers beyond those already selected, then calibrates a cutoff to meet a target all-miss risk.
Why this paper
Independent
In one line
CopRoute orders models by marginal coverage and calibrates a cutoff; it uses up to 79 percent fewer calls at the same certified all-miss risk.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (3 noted)
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§