ml-pipes¶

Build explicit ML pipelines you can validate, run, inspect, trace, and benchmark.

ml-pipes treats the pipeline as the primary artifact: an ordered sequence of small operators with clear input and output boundaries. That makes the whole data path—from loading and preprocessing through inference, postprocessing, and delivery—visible, composable, and testable.

Get Started¶

ml-pipes requires Python 3.10 or later. Install the umbrella package for the core framework, then choose the optional package profiles your pipeline needs:

pip install ml-pipes
pip install 'ml-pipes[onnx,vision]'

See Packages for the package matrix, install profiles, and public import surfaces. The repository’s examples provide runnable starting points.

Explore the Framework¶

• Design explains the composition-first model behind ml-pipes.
• Architecture maps that model to the framework’s runtime and ownership boundaries.
• Build Pipelines covers operators, composition, execution regions, and model scaffolding.
• Tooling covers validation, inspection, tracing, benchmarking, and performance guidance.