On the 30th of July 2020, we have released a new version (v2.1) of the UbiOps platform.
We made major improvements to the underlying architecture over the past months. These result in better performance, faster response times and higher reliability of UbiOps. Besides this, new functionality is available. Check the new Features & Functionality overview below.
Synchronous, low-latency requests
- UbiOps now offers synchronous, low latency, model requests. This means that you can send data to a model and retrieve the result with the same API request. This functionality enables a very low latency response and makes it easy to for instance run models in the back of a web application.
- The asynchronous request functionality still exists as well and is extended with a batch feature, which provides the option of sending many individual items within a single request.
- Both mentioned request types are also available for pipelines.
Command Line Interface available
- Besides our Python client library, we now released a CLI to interact with the UbiOps API from the terminal. It supports configuring models with yaml files and does model packaging and uploading by only pointing to the right directory. This makes it even easier to deploy your models to UbiOps and interact with the platform.
- You can download and install the CLI from our GitHub repository.
More flexible data structures for model input/output
- Data structures for input/output of models are now more flexible. Where previously you had to choose between sending a file to a model, or feed it with key-value pair data in a
json, it is now possible to do both at the same time! We made it more flexible to use files as a data entry for a request by extending our file handling functionality in the backend. This means that you can now run a model which for instance receives both an image as well as metadata. Or a model that receives or outputs multiple images.
- We added a plain text input data type, which allows you to send any plain text string to/from a model as input or output.
- Separate handling of files (blobs) is now supported through the UbiOps API.
- We updated the data pipeline editor in the webapp. This makes it even easier to create pipelines with multiple models in it. Book a demo to learn more about UbiOps!