zero the gradients after updating weights

[bintang-aswam]

Manually zero the gradients after updating weights by using machine epsilon for standard float (64-bit).

Fixes #ISSUE_NUMBER

Description

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.

cc @albanD @jbschlosser

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/tutorials/3447

• 📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit 5f1fbf5 with merge base b78fc75 ():
💚 Looks good so far! There are no failures yet. 💚

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

[albanD]

Would you be able to share details on why do you want to do this change.
Setting them to None is equivalent to setting them to exactly 0. which is best no?

[bintang-aswam]

Would you be able to share details on why do you want to do this change. Setting them to None is equivalent to setting them to exactly 0. which is best no?

answer:
@albanD, In other words, setting them to weight.grad = None is equivalent to setting them to weight.grad.zero_(). However, there are few conceptual difference in the context of numerically zeroing between zeroing the gradients vs tensor-multiplication of loss by machine-epsilon as follows:

[albanD]

Thanks for sharing these details.
I'm still not sure why this particular tutorial should be updated?
Maybe you want to do momentum-like updates here? If so, we can add a momentum term there?