While inspecting the sample data under the demo folder in the repository, I noticed an inconsistency in the XLA ops display between V1 and V2.
In V1, the XLA ops section shows three rows, whereas in V2 it only shows two rows. Could you help clarify why this difference exists? Is this expected behavior, or could it indicate a data processing or visualization issue in V2?
Additionally, I noticed that V1 includes parallel ops in the timeline. Could you provide guidance on how to reproduce this scenario? Specifically, I would like to understand what kind of workload or profiling setup is required to generate parallel XLA ops similar to those shown in the V1 demo sample.
Thanks!
While inspecting the sample data under the demo folder in the repository, I noticed an inconsistency in the XLA ops display between V1 and V2.
In V1, the XLA ops section shows three rows, whereas in V2 it only shows two rows. Could you help clarify why this difference exists? Is this expected behavior, or could it indicate a data processing or visualization issue in V2?
Additionally, I noticed that V1 includes parallel ops in the timeline. Could you provide guidance on how to reproduce this scenario? Specifically, I would like to understand what kind of workload or profiling setup is required to generate parallel XLA ops similar to those shown in the V1 demo sample.
Thanks!