TY - COMP
AU - Palazoglu, Berkant
AU - Hoppe, Fabian
AU - Scheib, Lukas
AU - Tarnawa, Michael
AU - Akdag, Hakan
AU - Comito, Claudia
AU - Gutiérrez Hermosillo Muriedas, Juan Pedro
AU - Hees, Jörn
AU - Jindra, Marc
AU - Korten, Till
AU - Krajsek, Kai
AU - Lemmen, Jonas
AU - Coquelin, Daniel
AU - Debus, Charlotte
AU - Hagemeier, Björn
AU - Knechtges, Philipp
AU - Rüttgers, Alexander
AU - Götz, Markus
TI - Heat (v1.7.0); 1.7.0
M1 - FZJ-2026-00696
PY - 2025
AB - <b>Overview</b><br>Version 1.7.0 introduces significant improvements in distributed data handling and numerical linear algebra, including a PyTorch-compatible DistributedSampler, randomized symmetric eigenvalue decomposition, and incremental SVD from HDF5. The release expands interoperability with partial Array API support and PyTorch 2.9.1, adds new distributed operations such as peak-to-peak, and delivers multiple robustness and device-placement bug fixes across MPI, GPU, and I/O workflows.<br>We are grateful to our community of users, students, open-source contributors, the European Space Agency, and the Helmholtz Association for their support and feedback.<br><br><b>Highlights</b><ul><li>DistributedSampler for efficient data loading and shuffling across multiple nodes with PyTorch</li><li>Randomized Symmetric eigenvalue decomposition (eigh)</li><li>Incremental SVD directly from an HDF5 file</li><li>Partial support of the Array API Standard (version: 2020.10), and API namespace under x.__array_namespace__(api_version='2020.10')</li><li>Distributed PTP (peak to peak) function</li></ul>
LB - PUB:(DE-HGF)33
DO - DOI:10.5281/ZENODO.18019822
UR - https://juser.fz-juelich.de/record/1052023
ER -