Real-World ML #016: Serverless ML
Join a community of builders 👨👨👦👦🛠️
Read time: 3 minutes
A Machine Learning service is a sequence of computation and storage steps, that takes in raw data and outputs model predictions, that help a business make better/smarter decisions.
To run such a system, you need an underlying infrastructure. That is a collection of services that let you compute, store artifacts (features/models), and orchestrate pipeline executions.
However, setting up and managing infrastructure is costly and time-consuming.
And this is when Serverless ML enters into the picture.
What is Serverless ML?
Serverless means you do not need to worry about setting up, running, and maintaining the infrastructure for your MLOps platform. Someone else (the service provider) will do it for you, better and probably cheaper.
Your job is to write business logic, and integrate the different services at the Python code level, which means you spend 0-time:
Setting up Docker registries
Creating IAM roles.
Managing EC2 instances or Kubernetes clusters, among others.
For example, the following is a Serverless ML stack that uses best-of-breed tools and has 0 maintenance costs for you.
Feature store by Hopsworks 📊
Experiment tracker and Model registry by Weights&Biases 🧪
Compute engine by Modal Labs 🏋️♂️
Model serving by HuggingFace Inference 💁
When you use a Serverless stack, you spend more time developing the ML pipelines that you wanna run on top of it, which means you ship ML products and services faster.
Join the Community of ML builders 🤗
Serverless ML is a growing Discord community for building ML systems
using Python
following MLOps best practices, and
managing 0 infrastructure.
Join today, connect with other builders, and accelerate your learning.
It is time to BUILD.
See you there!
Pau