srinivas212
srinivas212
Does affinitizing MPI rank to GPU expected to help?
This is a useful fix. ET feeder is being used by other downstream use cases and we need (a) unit test (b) before and after improvements. Thanks!
@changhai0109 plz resolve conflicts. will review. we landed few changes recently.
@Anchorrrr it would be great if you can update this PR given the feedback / pointers from @TaekyungHeo. Please let us know if you have additional questions. Thanks!
> Add github actions for enforcing clang format rules > > Address comment - > [#18 (comment)](https://github.com/openucx/torch-ucc/pull/18#issuecomment-718854392) WIP - investigating why clang format is failing
> > Add github actions for enforcing clang format rules > > Address comment - > > [#18 (comment)](https://github.com/openucx/torch-ucc/pull/18#issuecomment-718854392) > > WIP - investigating why clang format is failing Fixed....
Thanks for reporting this issue. @TaekyungHeo - we probably need to handle this as COMM_SEND_NODE right? Wdyt? This cannot be a collective operation.
@qyysjtu this issue should be fixed now - [#PR112](https://github.com/mlcommons/chakra/pull/112)
Thanks for reporting this issue. We will consider this in the next revision of schema. One issue with adopting nanosecond would be to convert the default of every op we...
These text files are an artifact of ASTRA-sim 1.0 and not Chakra. The best way to get these traces is collect it by running PyTorch model and enabling the profiler....