I have a question about how the features are transferred between multiple training machines.
For example, if machine A needs to read 10 graph nodes’ features from machine B, will A request 10 times (each time only request one node’s feature) to machine B or request only once to machine B? What’s the unit of request to another machine?
It only makes one request per minibatch per machine.
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.