Structure of a Packflow Project

Packflow provides utilities for ensuring projects are archived in a consistent, reproducible format, ensuring downstream users can process and deploy Packflow projects in a reliable or automated fashion.

The packflow CLI provides a lightweight templating command, packflow create ..., to initialize a Packflow project, consisting of the following files:

File	Required	Description	Default State
packflow.yaml	Yes	high-level metadata and configuration file housing information such as Python version, Inference Backend class, etc.	Template
requirements.txt	Yes	requirements file specifying the list of Python dependencies	Contains only packflow dependency; user must add model-specific dependencies
MODEL_CARD.md	Recommended	markdown file documenting model details, limitations, ownership, etc.	Template
LICENSE.txt	Recommended	the applicable licensing information for the project	Blank
README.md	Recommended	readme file for use by developers and other contributors	Template
inference.py (or custom name)	Required, but can be renamed by inference_backend configuration.	inference file containing the Backend execution object. Required only for loader: local mode. The filename must match the module name in inference_backend (e.g., custom_module:Backend requires custom_module.py)	Template
tests.py	Optional	pytest test file providing a starting point for testing the Backend, including a backend.check_io() format check. Not required for validation or export.	Template

The packflow CLI also contains the packflow export ... command for bundling these components into a single zipped archive, simplifying the process of sharing and deploying inference code across different organizations and environments.

See the CLI Reference for details on these commands.