Closed WardLT closed 3 years ago
We still have some work to do on using the value server to scale inference to >800 nodes.
@gpauloski
Done, we get reasonable efficiency up to 2048 nodes (see MLHPC paper)
We still have some work to do on using the value server to scale inference to >800 nodes.