Ata Fatahi

@calestyo would you please give a description of how and where the parities are updated in disks?

@romerojosh we're using pretty much the same patch for our model parallel training, though we call it `PartialDistributedGradientTape` since it's a different use case than the default data parallel use...

Thanks for the insight, @romerojosh. Great, I'll raise a PR for `PartialDistributedGradientTape` once this PR is merged then.

> In `docker/horovod/Dockerfile` it says > > ``` > # TensorFlow version is tightly coupled to CUDA and cuDNN so it should be selected carefully > ``` > > How...

> > @EnricoMi See the GPU table here: https://www.tensorflow.org/install/source#gpu > > Yes, I found that table, but it ends at `tensorflow-2.6.0`. Any idea where I can find those info for...

@romerojosh @maxhgerlach reopening this one again. Do we need support for TF1 in this? currently it's assume TF2 only.

Hi, I opened #1894 it builds the images successfully and resolves the build issues mentioned in #1566.

@googlebot I signed it!

Please checkout the examples directory. One such example for tensorflow custom training loop can be found here: https://github.com/horovod/horovod/blob/master/examples/tensorflow2/tensorflow2_mnist.py

Running into the same issue when building tf addons 0.19 with cuda 11.8. what config should be used in this case? In my case removing cub from WORKSPACE similar to...