About

Author: Alessandro Masat.

The RAPTOR family exists to make GPU numerics — kernels, graph capture, automatic differentiation, zero-copy array exchange — accessible on modest and edge hardware, for research and education, without forcing a choice between writing a kernel once and running it on CPU threads when a GPU isn’t there. Four libraries split that problem cleanly: aether lays out the arrays, eagle launches and captures what runs on them, raptor is the dependency-free spine that lets a kernel producer and an executor agree without depending on each other, and hawk authors and compiles the kernels. Each works alone; adding a companion buys speed or a deployment option, never a capability class.

Contact: via GitHub — github.com/amasat01. If you want to build on this, or apply these ideas to a new problem, get in touch: collaboration is the point, and a citation is the currency.