Quick Start Guide

graph_framework: A graph computation framework that supports auto differentiation. It is designed to allow domain scientists to create cross platform GPU accelerated code and embed those codes in existing legacy tools. It is designed for the domain of physics problems where there same physics is applied to a large ensemble of independent systems.

This framework enables:

1. Portability to Nvidia, AMD, and Apple GPUs and CPUs.
2. Abstraction of the physics from the compute.
3. Auto Differentiation.
4. Embedding in C, C++, and Fortran codes.

The compute kernels created have strong scaling to multiple devices

and the best throughput on both GPUs and CPUs compared to other frameworks like MLX and JAX

Documentation

graph_framework-docs Documentation for the graph_framework.

Obtaining the Code

To get started clone this repository using the command.

git clone https://github.com/ORNL-Fusion/graph_framework.git

Compiling the Code

For instructions to build the code consult the build system documentation. This framework uses a cmake based build system and requires the NetCDF-C library.