Published June 2, 2026

RouterDC

BymeAI TeamUpdated June 2, 2026

Dual Contrastive learning based routing — learns query-LLM compatibility with limited labeled data.

Overview

RouterDC uses dual contrastive learning to learn query-LLM compatibility representations without requiring large labeled datasets.

How It Works

The router learns embedding spaces where compatible query-LLM pairs are pulled together and incompatible pairs are pushed apart. This contrastive objective produces robust routing decisions even with limited labeled examples.

Strategy

Uses dual contrastive learning to learn query-LLM compatibility.

API Endpoint

autoroute:routerdc

Use Cases

When you need robust routing with limited labeled data
Semi-supervised routing scenarios
Building generalizable routing representations

Best Practices

Data Efficiency

RouterDC excels in low-data regimes. Start with as few as 50-100 labeled examples and evaluate whether additional labeling improves performance before scaling up.

KNN Router — Also works with limited data but less robust
MLP Router — Requires more data but can capture more complex patterns