Difference between `Custom handler with module level entry point` and `Custom handler with class level entry point` about serve HOT 2 CLOSED

Hi @IonBoleac,

the difference is that for module level entry point you provide a single function that is called for model initialization and prediction (handle). It is supposed to take two arguments: data and ctx. If data is None the function should initialize the model. When data is not None do inference on it. In the class level entry point case this corresponds to the two methods that need to be provided by the class: initialize and handle. The recommended way (and the one most if not all our examples use) is the class level entry point where you start by deriving from BaseHandler and implement initialize. The handle method is already implemented by BaseHandler and you only need to overwrite pre-/postprocessing and inference depending on your use case.

Hope that helps. Let me know if you still have questions. Happy to dive deeper into this.

Matthias

from serve.

Closing this issue for now. Feel free to re-open if you have further questions.

from serve.