MLPRouter (training + inference)

MLPRouter is a neural classifier router. It can learn non-linear routing boundaries on top of query embeddings.

Notebooks: - Training: https://github.com/ulab-uiuc/LLMRouter/blob/main/notebooks/mlprouter/01_mlprouter_training.ipynb - Inference: https://github.com/ulab-uiuc/LLMRouter/blob/main/notebooks/mlprouter/02_mlprouter_inference.ipynb

Router docs: https://github.com/ulab-uiuc/LLMRouter/blob/main/llmrouter/models/mlprouter/README.md

Configs

• Train: https://github.com/ulab-uiuc/LLMRouter/blob/main/configs/model_config_train/mlprouter.yaml
• Test: https://github.com/ulab-uiuc/LLMRouter/blob/main/configs/model_config_test/mlprouter.yaml

Note

The inference notebook writes a minimal mlprouter_inference.yaml for convenience. For the CLI, you can also use the repo test config above.

Run (CLI)

Train:

llmrouter train --router mlprouter --config configs/model_config_train/mlprouter.yaml

Route-only inference:

llmrouter infer --router mlprouter --config configs/model_config_test/mlprouter.yaml --query "Explain transformers." --route-only

Full inference:

llmrouter infer --router mlprouter --config configs/model_config_test/mlprouter.yaml --query "Explain transformers."

Tips

• If you overfit quickly, reduce model size or add regularization (see hparam).
• If training is slow, try fewer epochs first.