[3/N] Refine beginner tutorial by accelerator api

[guangyey]

For #3113

Motivation

• -> [3/N] Refine beginner tutorial by accelerator api #3170
• [2/N] Refine beginner tutorial by accelerator api #3168
• [1/N] Refine beginner tutorial by accelerator api #3167

As titled, we aim to use accelerator API to refine the beginner model tutorial to make it device-agnostic.

Checklist

• The issue that is being fixed is referred in the description (see above "Fixes #ISSUE_NUMBER")
• Only one issue is addressed in this pull request
• Labels from the issue that this PR is fixing are added to this pull request
• No unnecessary issues are included into this pull request.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/tutorials/3170

• 📄 Preview Python docs built from this PR

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

❌ 4 New Failures

As of commit 43a542e with merge base 540bd0c ():

NEW FAILURES - The following jobs have failed:

• Build tutorials / pytorch_tutorial_build_worker (12, 15, linux.4xlarge.nvidia.gpu) (gh)
  AttributeError: module 'torch' has no attribute 'accelerator'
• Build tutorials / pytorch_tutorial_build_worker (13, 15, linux.4xlarge.nvidia.gpu) (gh)
  AttributeError: module 'torch' has no attribute 'accelerator'
• Build tutorials / pytorch_tutorial_build_worker (15, 15, linux.4xlarge.nvidia.gpu) (gh)
  AttributeError: module 'torch' has no attribute 'accelerator'
• Build tutorials / pytorch_tutorial_build_worker (7, 15, linux.4xlarge.nvidia.gpu) (gh)
  AttributeError: module 'torch' has no attribute 'accelerator'

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

[albanD]

Sounds good to me.
Will let @svekars take a final look.