Published
KNN Router
K-Nearest Neighbors based routing — simple and effective query routing using similarity to training examples.
Overview
The KNN Router is one of the simplest yet most effective routing strategies. It routes queries based on their similarity to labeled training examples using K-nearest neighbors.
How It Works
Given a set of training queries labeled with the best-performing LLM, the KNN Router finds the K most similar examples to a new incoming query and assigns it to the most common LLM among those neighbors.
The similarity metric is computed using embedding-based distance functions, making it effective for semantically similar queries.
Strategy
Routes based on query similarity to training examples using K-nearest neighbors.
API Endpoint
autoroute:knnrouter
Use Cases
- Simple, effective routing for most tasks
- When you have good training data
- Rapid prototyping and baseline comparisons
Best Practices
Choosing K
Start with K=5 and tune based on your dataset. Smaller K values make the router more sensitive to local patterns, while larger K values provide smoother decision boundaries.
Related Models
- SVM Router — For high-dimensional query spaces
- MLP Router — When you need to learn complex routing patterns