Remove MXNet example

[tenzen-y]

Remove MXNet example

[tenzen-y]

As this is blocking all Katib PR merges, we should work on this ASAP.
@andreyvelich @johnugeorge WDYT?

[andreyvelich]

Yes, we should remove MXNet example from Katib Trials.
Maybe we could update this Katib PyTorch FashionMNIST example with the same PyTorch example as here: https://github.com/kubeflow/training-operator/blob/master/examples/sdk/create-pytorchjob-from-func.ipynb
But we need to see how fast this training will be executed on Katib E2Es with amount of resources that we have with GitHub actions.

[tenzen-y]

Yes, we should remove MXNet example from Katib Trials. Maybe we could update this Katib PyTorch FashionMNIST example with the same PyTorch example as here: https://github.com/kubeflow/training-operator/blob/master/examples/sdk/create-pytorchjob-from-func.ipynb But we need to see how fast this training will be executed on Katib E2Es with amount of resources that we have with GitHub actions.

@andreyvelich Why do we need to update this with the training-operator one? Could we just use https://github.com/kubeflow/katib/tree/master/examples/v1beta1/trial-images/pytorch-mnist?

[andreyvelich]

@tenzen-y Do we want to remove training evaluation to increase training time ?
Also, maybe we should update the PyTorch version to improve performance ?
https://github.com/kubeflow/katib/blob/master/examples/v1beta1/trial-images/pytorch-mnist/requirements.txt#L2

[tenzen-y]

Do we want to remove training evaluation to increase training time ?

@andreyvelich I'm ok with either way. I'm just wondering if reusing the existing example would be better.

Also, maybe we should update the PyTorch version to improve performance ?
https://github.com/kubeflow/katib/blob/master/examples/v1beta1/trial-images/pytorch-mnist/requirements.txt#L2

I agree with updating PyTorch version, but I think that we can work on version updates in a separate issue.

[andreyvelich]

Sure, let's see how much time it takes for E2Es if we just re-use this example rather than MXNet training.

[tenzen-y]

Sure, let's see how much time it takes for E2Es if we just re-use this example rather than MXNet training.

It makes sense.

[tenzen-y]

/assign