# stable-pretraining

> PyTorch Lightning framework for self-supervised learning research. Compose
> forward functions, online evaluation, and Manager-managed training. These
> docs follow main; Jet and the packaged quickstart are unreleased additions.

Package: stable-pretraining. Import: stable_pretraining as spt.
Context7 ID: /galilai-group/stable-pretraining.
Check the installed version before using an API from main.

## Task guides (plain Markdown)

- [First run](https://galilai-group.github.io/stable-pretraining/_sources/guides/quickstart.md): installation and a CPU smoke experiment.
- [Custom images](https://galilai-group.github.io/stable-pretraining/_sources/guides/custom_images.md): paired views and ImageFolder input.
- [Online evaluation](https://galilai-group.github.io/stable-pretraining/_sources/guides/online_evaluation.md): detached probe training and labels.
- [Resume](https://galilai-group.github.io/stable-pretraining/_sources/guides/resume.md): full-state checkpoints and configuration invariants.
- [Jet](https://galilai-group.github.io/stable-pretraining/_sources/guides/jet.md): invertible tokens, exact logdet, scale configuration.
- [Agent usage](https://galilai-group.github.io/stable-pretraining/_sources/guides/agents.md): Context7, optional skill, and evaluation protocol.

## Reference

- [API](https://galilai-group.github.io/stable-pretraining/api/backbone/): encoder architectures and utilities.
- [Methods](https://github.com/galilai-group/stable-pretraining/blob/main/METHODS.md): method catalog and forward-function names.
- [Source](https://github.com/galilai-group/stable-pretraining): development code and executable tests.