Hacker News new | past | comments | ask | show | jobs | submit login
Beyond data and model parallelism for deep neural networks (acolyer.org)
50 points by feross 14 days ago | hide | past | web | favorite | 1 comment



A 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/




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: