The Easiest Way to Build and Deploy Machine Learning Microservices
Creating ML apps should be simple and frictionless. UnionML is an open-source Python framework built on top of Flyte™, unifying the complex ecosystem of ML tools into a single interface.
Install UnionML and get started
.png)
Bring your own tools
Combine the tools that you love using a simple, standardized API so you can stop writing so much boilerplate and focus on what matters: the data and the models that learn from them.
Fit the rich ecosystem of tools and frameworks into a common protocol for machine learning.
A unified interface for your ML team
Using industry-standard machine learning methods, implement endpoints for fetching data, training models, serving predictions (and much more) to write a complete ML stack in one place.
Data science, ML engineering, and MLOps practitioners can all gather around UnionML apps as a way of defining a single source of truth about your ML system’s behavior.
This helps you maintain consistent code across your ML stack, from training to prediction logic.
.png)
.png)
Prototype locally
Begin with a proof of concept, starting with small data or simple models to make sure the ML system as a whole works.
Your laptop — or remote Jupyter environment — acts as a regularizer so you can scale your dataset set and reach for more or less complex models depending on where you are in the prototyping phase.
Scale up according to your needs
Once you’re confident your ML system behaves the way you intend it to, leverage the power of Flyte™ and UnionCloud to scale up your training workload in a vendor-agnostic way!
You shouldn’t need to bug an MLOps engineer to accelerate your workloads ... Just request them through the function decorators!
%20(2)%20(2)%20(1).png)
Serve anywhere seamlessly
No need to re-implement this logic in multiple places based on your use case.
Implement feature transform and prediction logic once, deploy to production in batch and online serving contexts.
Streaming coming soon!
Experiment efficiently
coming soon
With built-in data lineage and registry capabilities, never worry about losing experiment metadata again, even when performing ad-hoc experiments!
.png)
