1 code implementation • 3 Jul 2023 • Xunyi Zhao, Théotime Le Hellard, Lionel Eyraud, Julia Gusak, Olivier Beaumont
We show through experiments on many models that Rockmate is as fast as Rotor and as efficient as Checkmate, and that it allows in many cases to obtain a significantly lower memory consumption for activations (by a factor of 2 to 5) for a rather negligible overhead (of the order of 10% to 20%).
no code implementations • 21 Feb 2022 • Julia Gusak, Daria Cherniuk, Alena Shilova, Alexander Katrutsa, Daniel Bershatsky, Xunyi Zhao, Lionel Eyraud-Dubois, Oleg Shlyazhko, Denis Dimitrov, Ivan Oseledets, Olivier Beaumont
Modern Deep Neural Networks (DNNs) require significant memory to store weight, activations, and other intermediate tensors during training.
no code implementations • NeurIPS 2021 • Olivier Beaumont, Lionel Eyraud-Dubois, Alena Shilova
Rematerialization and offloading are two well known strategies to save memory during the training phase of deep neural networks, allowing data scientists to consider larger models, batch sizes or higher resolution data.
1 code implementation • 9 Nov 2020 • Nathan Grinsztajn, Olivier Beaumont, Emmanuel Jeannot, Philippe Preux
In this paper, we propose a reinforcement learning approach to solve a realistic scheduling problem, and apply it to an algorithm commonly executed in the high performance computing community, the Cholesky factorization.
no code implementations • 27 Nov 2019 • Julien Herrmann, Olivier Beaumont, Lionel Eyraud-Dubois, Julien Hermann, Alexis Joly, Alena Shilova
This paper introduces a new activation checkpointing method which allows to significantly decrease memory usage when training Deep Neural Networks with the back-propagation algorithm.
no code implementations • 13 Feb 2019 • Navjot Kukreja, Alena Shilova, Olivier Beaumont, Jan Huckelheim, Nicola Ferrier, Paul Hovland, Gerard Gorman
Edge computing is the natural progression from Cloud computing, where, instead of collecting all data and processing it centrally, like in a cloud computing environment, we distribute the computing power and try to do as much processing as possible, close to the source of the data.
Distributed, Parallel, and Cluster Computing