Edward Kim
Edward Kim
> Just to double-check, does the current implementation of this method returns branches that contain at least one feature with the filter tag or only columns that contains the Tag....
Update: I changed the approach so that `ParallelBlock` calls the `select_by_tag` method of the branches: `EmbeddingTable` and `TabularBlock`. Now, `ParallelBlock` simply calls `EmbeddingTable.select_by_tag()` and `TabularBlock.select_by_tag()` (and also `Filter.select_by_tag`. With the...
Note: I'm not sure why `tensorflow/examples` test is failing and can't reproduce it locally. On my local machine, all tests under `examples` passed.
rerun tests
> 1. We require to initialize a separate dataloader - we do not require it in any other example. Is it required we will do it in this case? @bschifferer...
One test on Jenkins which was added very recently in https://github.com/NVIDIA-Merlin/models/pull/841 is failing. It seems like introducing horovod in the dependency or something else caused a dependency conflict with dask,...
@jperez999 Is there a way to produce equal number of batches so that the workload is balanced across workers? Although nvtabular seems to produce equal-sized batches in [tf_trainer.py](https://github.com/NVIDIA-Merlin/NVTabular/blob/main/examples/multi-gpu-movielens/tf_trainer.py), the number...
I've been doing some experimentation with horovod integration with Models API, based on @bschifferer's example https://github.com/NVIDIA-Merlin/models/pull/778, and I fully agree with his point on all the success criteria he listed...
@WonderingWJ Is this PR still relevant? If so, we will need to rebase this PR.