Skip to main content

Pipelines

aiXplain Pipelines are scalable solutions you can build in Design using Models, Metrics and Data assets, logic nodes and custom script nodes for additional flexibility and customisation. Pipelines are easily editable and allow for model hot swapping, letting you prototype quickly and easily swap providers. You can build pipelines through the SDK or through Studio.

Docusaurus themed imageDocusaurus themed image

Scalability and Multi-Threading​

Pipelines are designed to support large and complex processes. You may feed a pipeline with thousands of input instances to be processed, and the Platform will provide the resources to run its nodes and instances in parallel, speeding up the process.

Models vs Pipelines​

Models are intended to process small instances, such as audio up to 5 minutes and texts up to 2000 characters, one at a time. If you aim to process larger instances and many instances in parallel, then pipelines are the way to go.

Pipeline Components​

Decision Nodes​

The decision node allows or prevents data from passing through it based on logical branching criteria.

Input and Output Nodes​

The input and output nodes allow you to pass inputs to your pipelines and receive outputs. Each pipeline should have at least one input, asset, and output node. Asset nodes are used to run models and should have an asset ID.

Metric Nodes​

The Metric Node can be populated with a Metric asset. It accepts input-target pairs and outputs a score, allowing you to test the performance of your models and pipelines.

Router Nodes for Multimodal Inputs​

The Router Node allows pipelines to have multimodal inputs. It does so by detecting the input type, whether it’s text, audio, and so on, and routes it accordingly.

Script Nodes​

You can also add a Script Node to add a Python script to modify the output of the incoming nodes. It is a useful feature when you need to perform an action programatically.

Segmentors and Reconstructors​

Segmentors and Reconstructors are two terms that reflect the behavior of some aiXplain models, and as the names suggest, segmentors segment your data into segments, and reconstructors reconstruct them back together.