Abstract: State-of-the-art deep learning models rely on large GPU clusters and various parallelism strategies, which in turn depend on collective communication (CC) operators to synchronize data.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results
Feedback