Dali and uber horovod incompatibility issues about dali HOT 5 CLOSED

Tracked as DALI-181

from dali.

Hi,
Could you provide us info about models you are using and rest of the script (dataloader) so we could reproduce this on our side?

from dali.

Hi,It is the fault of my side. Our server updated the software a few days ago, which caused some of my horovod libraries to be incompatible. I have reinstalled it.
I am coming back to adjust my program, can I provide a tensorflow overall example?
I do data loading through dali and horovod to do multi-machine communication to accelerate tensorflow and achieve good results.
Can I provide the entire example? Dali does solve the problem that tensorflow originally loaded data slowly, which is really great.

from dali.

Hi,
As I understand after reinstalling horovod libraries everything works fine.
Regarding reproduction, next time please provide the full, stand-alone script we could run to reproduce the problem you have. Without it, the issue has a lower priority for us as we need to spend additional time trying to recreate env you have with bigger chance doing something different, which can prevent us from detecting the problem.
If the problem still exists please reopen this task.

from dali.

It's great that DALI solves the input pipeline speed for you.
If you want to contribute an example of using Tensorflow with DALI, that's great too. Contributions are very welcome!

from dali.