cuPhoton is organized as a set of reference workflows. Use the README's workflow chooser to match your inputs to a component and see what it produces. The component overview explains the scientific operations.
For a view across components, read how the science components fit together and what an exposure contains. Then run a synthetic quickstart and read the contract for the component you intend to adapt.
The 0.1.3 changelog summarizes the changes and migration requirements.
- Getting started: prerequisites and installation profiles.
- Packaging: local wheel and Conda builds, source builds, and package qualification procedures.
- Quickstarts: data-independent CPU and GPU smoke runs.
- Imaging walkthrough: notebook and script with setup for synthetic FITS loading, alignment, subtraction, dipole fitting, and plotting; requires a CUDA 13-capable NVIDIA GPU and xDR's native extension.
- Command-line index: the umbrella executable and command groups.
- Core: shared CLI and configuration behavior.
- cuPhoton xLens: scientific monitoring, offline dashboards and optional evidence-based investigation through institution-hosted models.
- xDataReader: GPU FITS loading and the shared Astropy/xDR reader interface, with runtime compression controls for CLI, Python and manifests; native GDS depends on the storage setup.
- xFit: batched nonlinear least-squares dipole fitting.
- xImage: multi-exposure reconstruction and PSF calibration.
- xPois: kernel fitting and image subtraction.
- xScan: transient datasets, classification, evaluation, and review.
- xRep (xReproject): FITS/WCS reprojection.
- xRay: X-ray trace and detector analysis.
- Architecture: package boundaries and execution model.
- Distributed architecture: work assignment, GPU ownership, data movement and completion.
- Dragon architecture: process placement, GPU sharing with MPS, queues and worker lifecycle.
- MPI architecture: rank binding, collective coordination and file mode.
- Distributed launch guide: prepare a pipeline manifest and launch MPI or Dragon workers using Slurm or SSH host files.
- Dragon transport and performance: transport selection, queue placement experiments and timing boundaries.
- Pipeline stage benchmark: compare a persistent GPU pipeline with separate stages and inspect transfer costs.
- Adapting the workflows: a practical extension process.
- Data and artifact contracts: shapes, files, metadata, and provenance.
- Uncertainty quantification: the error bars of transient catalogs, their terms, configuration, and checks on your data.
- Troubleshooting: environment, CUDA, data, and CLI failures.
- Glossary: project and astronomy terminology.
- Scientific references: methods represented in the code.
Project-wide policies are in Contributing, Governance, Support, and Security.