28 ms·
Beyond data and model parallelism for deep neural networks
- eslaught 7y agoA bit late to the party, but just FYI, this research uses Legion, a next-generation runtime system we're developing for distributed task execution [1]. Part of the subtext of this paper is that Legion is just much, much more efficient than TensorFlow's execution engine.... [1]: https://legion.stanford.edu/ https://legion.stanford.edu/