The pre-trained models are pretty skilled when it comes to import and export. In a nutshell, declaring deployment consists of importing the trained model and transforming input data into the format that's acceptable by the neural network. There are certain things that you need to keep in mind during deployment:
- The model can be pretty large, for example, hundreds of megabytes, which makes deployment more difficult.
- We need to keep versions of the model and track their performance. If you train the model by yourself, you might need to update the model based on changes in incoming data or if you find better architecture.
- Some models require additional files that translate predicted numbers or ...