---
id: prefigure
version: "0.0.10"
license: MIT
license_treatment: permissive
maintenance: abandoned
---

# prefigure — Run configuration management utils: combines configparser, argparse, and wandb.API

License: permissive · Maintenance: abandoned · Downloads: 126.3K/mo

## What it is and what it does

Prefigure streamlines experiment configuration for PyTorch Lightning workflows by unifying three sources of settings: defaults from an INI/JSON/Gin file, previous run configs pulled from Weights & Biases, and command-line overrides. It reduces boilerplate to three lines of code (import, get_all_args(), push_wandb_config()) and archives run settings to WandB for reproducibility and easy recovery of past configurations.

The package also includes an optional On-the-Fly Control (OFC) feature that lets you steer hyperparameters during training via a Gradio GUI or by editing a file, with optional logging of changes back to WandB. It supports multiple config formats and can import nested config files as dictionaries. However, the project has been abandoned since mid-2023 and carries maintenance risk.

Use it for:

- Archive and retrieve experiment configurations from past WandB runs to reproduce or iterate on previous training setups.
- Manage hyperparameters across multiple config sources (file, WandB, CLI) with a clear precedence order and minimal code changes.
- Steer learning rate, batch size, or other float/int parameters during training via a Gradio interface without stopping the run.
- Consolidate argparse and configparser logic into a single declarative defaults.ini file for cleaner training scripts.
- Share experiment configs across team members by pulling them from WandB URLs instead of passing files manually.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

Manages run configurations for machine learning experiments by combining command-line arguments, INI/JSON/Gin config files, and Weights & Biases logging into a unified interface.

Yes-with-conditions. The package solves a real problem for PyTorch Lightning + WandB workflows and has low install friction, but it is abandoned (last release June 2023, no commits since). Install only if you can tolerate potential incompatibilities with newer versions of pytorch-lightning, wandb, or gradio, and if you are comfortable maintaining a fork if needed. Not suitable for production systems requiring active maintenance.

## Install

pip install prefigure
uv add prefigure
poetry add prefigure

## Installing prefigure

Before you install: Low install friction with a pure-Python wheel, but the package is abandoned—last release was 2023-06-28 with no commits since. Six runtime dependencies including pytorch-lightning and wandb add moderate weight; maintenance risk is high.

License in practice: MIT license is permissive and imposes no restrictions on use, modification, or distribution in your own projects.

Quickstart:

pip install prefigure

from prefigure import get_all_args, push_wandb_config

args = get_all_args()
push_wandb_config(wandb_logger, args)

Requires pytorch-lightning and wandb to be installed; WandB account and project setup needed for config archiving and retrieval.

Verify before relying:

- Whether the package works with current versions of pytorch-lightning and wandb given its 2023 last release date.
- Whether the Gradio GUI (OFC feature) works with current Gradio versions.
- Compatibility with Python versions beyond 3.10 given the classifier list.

## Package facts

- License: MIT (permissive)
- Python support: unspecified
- Install friction: low
- Maintenance: abandoned
- Downloads: 126.3K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

ml experiment configuration management, wandb config archiving, pytorch lightning config, command line argument management, run settings management, hyperparameter configuration, experiment reproducibility, ml-experiment-tracking, pytorch-lightning, hyperparameter-management

[View on SkillFed](https://skillfed.io/packages/prefigure) · [View on PyPI](https://pypi.org/project/prefigure/)