Published
MLP Router
BymeAI Team
Multi-Layer Perceptron based routing — neural network-powered routing with learnable parameters.
Overview
The MLP Router uses a Multi-Layer Perceptron neural network to learn complex routing patterns from training data, making it one of the most flexible single-round routing strategies.
How It Works
Query embeddings are fed through one or more hidden layers with non-linear activations, allowing the network to learn intricate relationships between query characteristics and optimal LLM selection.
Strategy
Neural network-based routing with learnable parameters.
API Endpoint
autoroute:mlprouter
Use Cases
- Complex routing patterns that simpler models cannot capture
- When you have sufficient training data
- Scenarios with non-linear decision boundaries
Best Practices
Architecture Design
Start with a single hidden layer of 128 units and increase complexity only if needed. Deeper networks require more data and careful regularization to avoid overfitting.
Related Models
- SVM Router — For linear decision boundaries
- KNN Router — Simpler baseline for comparison