Upgrade Python and Pytorch versions for some examples

[tenzen-y]

What this PR does / why we need it:
I upgraded Python and Pytorch versions for darts-cnn-cifar10 and pytorch-mnist examples.

Which issue(s) this PR fixes (optional, in fixes #<issue number>(, fixes #<issue_number>, ...) format, will close the issue(s) when PR gets merged):
Fixes #1883

Checklist:

• Docs included if any changes are user facing

[coveralls]

Coverage decreased (-0.1%) to 72.857% when pulling 224a6ef on tenzen-y:upgrade-python-pytorch-version into 9ee8fda on kubeflow:master.

[tenzen-y]

@johnugeorge
This improvement would be better to include in the next Katib release.

[anencore94]

/lgtm
It built well in amd64, cpu, linux

[tenzen-y]

I will do force-push to re-run Go Test.

[anencore94]

/lgtm

[johnugeorge]

/approve

[google-oss-prow]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: anencore94, johnugeorge, tenzen-y

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [johnugeorge]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[djmcgreal-cc]

Since the base image is no longer built with GPU support, CUDA acceleration is not available. Since we were using this image in our tests, they broke. It's misleading because the mnist.py still tries to use cuda but it will never be available.

[tenzen-y]

Thanks for your comment @djmcgreal-cc.
We have stopped supporting CUDA for the pytorch-mnist image in order to reduce image size due to test-infra issues in #1882.

@johnugeorge
Would you like to provide another pytorch-mnist image with CUDA support?

[johnugeorge]

@tenzen-y cuda image resources are not available with Github actions?

[tenzen-y]

@tenzen-y cuda image resources are not available with Github actions?

@johnugeorge Since the image with CUDA support size is very heavy as shown below, we will face no space left on device: unknown error.

Not CUDA support image size: 639.63 MB
CUDA support image size: 1.2 GB

[johnugeorge]

@tenzen-y We can then defer this building step till we find a solution. eg: CI in AWS

Other option, is to manually build and push image as changes are not frequent to this build

[tenzen-y]

@johnugeorge
We can build the image in this workflow, although we can not run an operation check of the image in this workflow.

So, we cannot confirm that the image works, but we can provide the image with support CUDA built in gh-actions.

[johnugeorge]

@tenzen-y I feel, that should be ok for now

[tenzen-y]

@johnugeorge
I see.
I will create a PR to add the pytorch-mnist-gpu image by the end of this week.