ptens documentation¶
ptens is a C++/CUDA library for permutation equivariant message passing and higher order equivariant graph neural networks. For an introduction to the Ptensors formalism, see this preprint. ptens can perform equivariant message between Ptensors on both the CPU and the GPU. Importantly, the library can perform this process in parallel across many pairs of source and destination Ptensors, even when the Ptensors are of different sizes and their reference domains overlap in different ways. This document describes ptens’s PyTorch interface.
ptens is developed by Risi Kondor’s research group and is released for noncommercial use under a custom license that can be found in the file LICENSE.TXT