define public API for torch.nn.utils

[ringohoffman]

Adding modules imported here and the following functions to the __all__:

• clip_grad_norm_
• clip_grad_value_
• remove_weight_norm
• parameters_to_vector
• vector_to_parameters
• remove_spectral_norm
• skip_init

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/111026

• 📄 Preview Python docs built from this PR
• 📄 Preview C++ docs built from this PR
• ❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

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

✅ No Failures

As of commit 146d78c with merge base 95ff51d ():
💚 Looks good so far! There are no failures yet. 💚

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

[ringohoffman]

clip_grad_norm (deprecated), fuse_conv_bn_eval, fuse_conv_bn_weights, and convert_conv2d_weight_memory_format are all imported here but do not have docs pages.

[vadimkantorov]

Btw vector_to_params / params_to_vector seems related to flatten_parameters concept ...

[mikaylagawarecki]

You can expose the documentation for the functions you mentioned that are missing from the documentation over here

Those functions you mentioned that are failing the public bindings test should still be added to the __all__

[ringohoffman]

I added the other 2 functions from fusion since it seemed strange that they would be omitted. I also tried to split up nn.rst into more logical sections so that the functions in the same submodule would be kept and documented together. LMK how this looks.

[mikaylagawarecki]

For future reference, you can check out the docs preview from your PR under

📄 Preview Python docs built from this PR

in the pytorch-bot comment above.

It renders nicely to me, it would be awesome if you could add short docstrings for fuse_conv_bn_eval
fuse_conv_bn_weights
fuse_linear_bn_eval
fuse_linear_bn_weights

as well since they come up as blank in the table

[ringohoffman]

@mikaylagawarecki I gave it my best effort! Please LMK what you think of my changes.

[ringohoffman]

fuse_conv_bn_weights & fuse_linear_bn_weights are written such that they expect that bn_rm and bn_rv are not None, but there are a few places where these functions are called without checking that they are not None. I think the easiest thing to do will be to assert that they are not None in all these places these functions are called since they will certainly raise TypeErrors if they are None anyway.

[mikaylagawarecki]

This is some much appreciated cleanup! Thanks a lot for your effort :)

[mikaylagawarecki]

@pytorchbot merge

[pytorchmergebot]

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here