lanl/THOR

This repository contains a novel, multi-GPU implementation of tensors in tensor train format. It is based on the modern Fortran MPI/GPU communication library (Thunder), and a CUDA-enabled algebra of distributed arrays (DRay).

The THOR Project (Tensors for High-dimensional Object Representation) aims to advance the state-of-the-art in tensor calculations, manipulation, and research. We strive to provide a high-performance tensor library for various scientific applications, containing ready-to-use utilities and applicaions in Fortran, Matlab, and Python.

BibTeX:

@techreport{alexandrov2024thor,
  title        = {Tensors Optimized for High-level Research (THOR): an efficient and easy-to-use library for tensor networks},
  author       = {Alexandrov, Boian and Boureima, Ismael Djibrilla and Korobkin, Oleg and Danis, Mustafa Engin},
  institution  = {Los Alamos National Laboratory},
  number       = {LA-UR-24-24375},
  year         = {2024},
  month        = {may},
  day          = {6},
  note         = {Approved for public release; distribution is unlimited},
  type         = {Report}
}