Yuan, Man comments

Results 6 comments of


                                            Yuan, Man

Pytorch distributed job failed when master replica start later than worker replica

See https://github.com/kubeflow/pytorch-operator/issues/125 for more details. https://github.com/kubeflow/arena/blob/master/charts/pytorchjob/templates/pytorchjob.yaml#L313 Maybe changing this line to `OnFailure` can help resolving this issue?

S3 addressing style should be able to be configured by environment variable.

Hi @yongtang , can you take a look at this issue? Thanks.

[Feature Request] Add capability for utilizing NVLink across docker containers.

cc @cheyang @WencongXiao

to_sparse failed for Value with ragged_rank > 1 read from parquet file

Thanks for your report, I will look into it.

to_sparse failed for Value with ragged_rank > 1 read from parquet file

@DelightRun Could you try the latest commit ?

to_sparse failed for Value with ragged_rank > 1 read from parquet file

> > @DelightRun Could you try the latest commit ? > > I use your pre-built v0.8.0 wheel package with TensorFlow 1.15.0. It's not very convenient for me to compile...