SINAPSE SDK

The SINAPSE SDK is a curated collection of software components for AI-coupled HPC workflows (Scalable INfrastructure for AI-coupled Predictive Simulation Enhancement, NSF award 2514139).

The SDK curates existing components and provides common packaging, testing, documentation, and deployment across them, with well-defined integration points between the components.

Aims

Modern scientific computing increasingly couples AI models with physics-based simulation — surrogate models steering ensembles, simulations generating training data, inference services running alongside HPC tasks. The tools that make this possible exist, but they are developed independently, packaged differently, and integrated ad hoc.

The SINAPSE SDK addresses this by:

  • Curating a set of proven components for AI-coupled HPC workflows rather than building a monolithic framework.

  • Standardizing packaging, testing, documentation, and release processes across the components.

  • Defining integration points so components interoperate through well-specified APIs.

Components mature through three levels:

  • L0 — technologies individually collected: uniform documentation, packaging, testing, and processes

  • L1 — components interoperate via pointwise, tool-specific integrations

  • L2 — sustainable integration: well-defined APIs and integration points, deeply configurable

Installation

The sinapse-sdk meta-package installs a mutually compatible set of the SDK components. The conda package is the complete set:

$ conda install -c conda-forge sinapse-sdk

It pulls every component plus flux-core for RHAPSODY’s Flux backend. It installs on Linux (x86_64 and aarch64) with Python 3.11 or newer, the floor set by its conda-only dependencies. The Dragon backend for RHAPSODY is PyPI-only: pip install dragonhpc into that conda environment.

The PyPI package covers the components that have PyPI releases:

$ pip install sinapse-sdk

Note

Seekrflow (depends on SEEKR and OpenMM) and Flux are conda-only, so the pip install lacks Seekrflow and RHAPSODY’s Flux backend. The pip package requires Python 3.9 or newer.

Each component can also be installed on its own; see the package links on the Components page. Spack packaging is planned.

Component documentation

Repositories and documentation for the SDK and its released components:

The full component list, including package links, is on the Components page.

Acknowledgment

This work is supported by the National Science Foundation under award 2514139 (collaborative award; Rutgers University, University of Chicago, Princeton University, UC San Diego).