← Deep Learning learning pathDeep Learning · Module 10 of 10
Deploying deep learning models
Take trained models into real applications.
📖 Key ideas
- Save models (PyTorch state_dict, Keras .keras, ONNX).
- Serve via APIs (FastAPI, TorchServe, TF Serving).
- Optimise: quantisation, pruning, smaller models.
- Edge deployment: TFLite, ONNX Runtime, Core ML on phones.
- Monitor accuracy and data drift after launch.
💡 Real-life example: Deploying a model is opening a restaurant after perfecting the recipe in a test kitchen.
ONNXOpen format for model exchange
QuantisationLower-precision weights for speed
Model servingExposing a model via an API
Data driftInput data changing over time