There are many frameworks, libraries, and tools in the machine learning and Python spaces. One that stands out for machine learning is BentoML. BentoML takes a step outward and looks at all the frameworks for modeling, sees the need for presenting a public API for these models, and sees a need for a way to package the models for production. BentoML will take your model, expose a public API, package it in a container suitable for delivery to a variety of targets, and is ideal for Kubernetes.
BentoML supports many of the commonly found ML frameworks: Scikit-learn, PyTorch, Tensorflow, TF Keras, XGBoost, LightGBM, FastText, FastAI, H2O, ONNX, Spacy, Statsmodels, CoreML, Transformers, and Gluon, and is always looking for more.
This lab will introduce BentoML using Scikit-learn. We’ll use the ubiquitous Iris model. BentoML will generate an API endpoint, package the Scikit-learn model in a container, and add a predict REST call.
In this lab, you will learn how to:
☐ Install BentoML
☐ Code a model like Scikit-learn
☐ Use BentoML to generate an API for the model
☐ Use BentoML to package the model and API in a container
☐ Run and access the model on Kubernetes
These instructions have been adapted from the BentoML documentation on Deploying to Kubernetes Cluster .
Because BentoML handles the container build and containers can run anywhere, there is a wide choice of deployment targets for the model in a container.
Beginner
25 minutes
31 Dec, 2021