Deploy Python code with Model Serving

This article describes how to deploy Python code with Model Serving.

MLflow’s Python function, pyfunc, provides flexibility to deploy any piece of Python code or any Python model. The following are example scenarios where you might want to use the guide.

  • Your model requires preprocessing before inputs can be passed to the model’s predict function.

  • Your model framework is not natively supported by MLflow.

  • Your application requires the model’s raw outputs to be post-processed for consumption.

  • The model itself has per-request branching logic.

  • You are looking to deploy fully custom code as a model.

Construct a custom MLflow Python function model

MLflow offers the ability to log Python code with the custom Python models format.

There are two required functions when packaging arbitrary python code with MLflow:

  • load_context - anything that needs to be loaded just one time for the model to operate should be defined in this function. This is critical so that the system minimize the number of artifacts loaded during the predict function, which speeds up inference.

  • predict - this function houses all the logic that is run every time an input request is made.

Log your Python function model

Even though you are writing your model with custom code, it is possible to use shared modules of code from your organization. With the code_path parameter, authors of models can log full code references that load into the path and are usable from other custom pyfunc models.

For example, if a model is logged with:

mlflow.pyfunc.log_model(CustomModel(), "model", code_path = ["preprocessing_utils/"])

Code from the preprocessing_utils is available in the loaded context of the model. The following is an example model that uses this code.

class CustomModel(mlflow.pyfunc.PythonModel):
    def load_context(self, context):
        self.model = torch.load(context.artifacts["model-weights"])
        from preprocessing_utils.my_custom_tokenizer import CustomTokenizer
        self.tokenizer = CustomTokenizer(context.artifacts["tokenizer_cache"])

    def format_inputs(self, model_input):
        # insert some code that formats your inputs

    def format_outputs(self, outputs):
        predictions = (torch.sigmoid(outputs)).data.numpy()
        return predictions

    def predict(self, context, model_input):
        model_input = self.format_inputs(model_input)
        outputs = self.model.predict(model_input)
        return self.format_outputs(outputs)

Serve your model

After you log your custom pyfunc model, you can register it to the Unity Catalog or Workspace Registry and serve your model to a Model Serving endpoint.

Notebook example

The following notebook example demonstrates how to customize model output when the raw output of the queried model needs to be post-processed for consumption.

Customize model serving output with MLflow PyFunc notebook

Open notebook in new tab