Modernize the Ray Tune tutorial

[crypdick]

This is the first of several PRs against the Pytorch Tutorials. This PR unstales the Ray Tune tutorial:

• bumps Ray version from 2.7.2 -> 2.52.1
• update the tutorial with the latest code from the Pytorch CIFAR10 tutorial which it is based off of
• update to use the latest Ray Tune APIs
• fixed the norm values for CIFAR10
• rewrote most of the prose
• added new sections for sampling, observability, conclusion, additional links
• updated tags in the doc indices. Will need approval on the new Ray-Distributed tag.

Opening this as a draft PR so that I can get feedback from the Ray Tune core devs first

cc @subramen @albanD @jbschlosser @pcmoritz @robertnishihara @matthewdeng @richardliaw

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/tutorials/3694

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[crypdick]

cc @xinyuangui2

[xinyuangui2]

Looks good from ray tune side.

[crypdick]

@svekars Ready for review from the PyTorch side :)

[crypdick]

Happy new year @svekars . Gentle bump on the review on this PR. This one is soft blocking other contributions because of the Ray version bump.

[svekars]

Thanks for the reminder! Will review today.

[svekars]

I also see that now it takes 32 minutes to build this tutorial while the current version takes 12 minutes. Can we reduce the time of the build? We could try to put on a different worker and see if it helps.

[crypdick]

@svekars I made the following changes to speed up the tutorial:

• I reverted the number of data loader workers from 2 back to 8
• I made a GPU available to the tutorial by changing the CI worker to linux.g5.4xlarge.nvidia.gpu

[svekars]

LGTM - I added a couple of very minor formatting comments! Thank you for adjusting the execution time and updating the tutorial!