🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/155198

• 📄 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.

✅ You can merge normally! (2 Unrelated Failures)

As of commit f2ccb9c with merge base ee5c290 ():

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

❌ 🤖 pytorchbot command failed:

@pytorchbot: error: argument command: invalid choice: 'labels' (choose from 'merge', 'revert', 'rebase', 'label', 'drci', 'cherry-pick', 'close')

usage: @pytorchbot [-h] {merge,revert,rebase,label,drci,cherry-pick,close} ...

Try @pytorchbot --help for more info.

Didn't find following labels among repository labels: module:,docs,topic:,not,user,facing,topic:,docs

@pytorchbot label docathon-h1-2025
@pytorchbot label module: docs
@pytorchbot label topic: not user facing
@pytorchbot label topic: docs

@pytorchbot label topic: not user facing

Didn't find following labels among repository labels: topic:,not,user,facing

@pytorchbot label "topic: not user facing"

@pytorchbot label "topic: docs"

1. Ah I see. I was rebasing on viable/strict but it looks like cuda.rst was touched in the main branch in the meantime.

2. Concerning your changes (i.e. adding python as the code block language), I think I could fork the rst2myst tool and add that functionality fairly easy. For now though, I'll just add the language keyword manually.

3. Also, I think the fact that the "Created On" and "Last Updated On" banner don't show is actually a limitation of MySt, or there is a config option to be set.

4. Finally, it seems that the rst2myst is really bad at handling anchor links.

For example, it converted this in data.rst:

* `single- and multi-process data loading <Single- and Multi-process Data Loading_>`_,

* `automatic memory pinning <Memory Pinning_>`_.

To this:

- [single- and multi-process data loading],
- [automatic memory pinning][memory pinning].

These are definitely not valid anchor links.

I saw that in torch.md, the author is using refs and in my case we would want:

- {ref}`map-style and iterable-style datasets <dataset-types>`,
- {ref}`customizing data loading order <data-loading-order-and-sampler>`,
- {ref}`automatic batching <loading-batched-and-non-batched-data>`,

and then wherever that header is, we would put:

(dataset-types)=
## Dataset Types

...etc

Just want to confirm that this is the style we should adopt. I tried to use explicit headers as specified per here, but those did not output correctly when running make html.

(heading-target)=
### Heading

[reference1](#heading-target)

thanks for providing these details:

1. yes, please rebase to main
2. 👍
3. Created On | Last Updated On is a post-processing script we use - it should show. Let's get a preview for this PR to confirm
4. Using {ref} sounds good.

I think we need to fix these errors by fixing rst -> .md extensions in the include directives in these locations:

• https://github.com/pytorch/pytorch/blob/main/torch/nn/modules/rnn.py#L589
• https://github.com/pytorch/pytorch/blob/main/torch/nn/modules/rnn.py#L946
• https://github.com/pytorch/pytorch/blob/main/torch/nn/modules/rnn.py#L591
• https://github.com/pytorch/pytorch/blob/main/torch/nn/modules/rnn.py#L948
• https://github.com/pytorch/pytorch/blob/main/torch/ao/nn/quantized/dynamic/modules/rnn.py#L797

I think this won't work unfortunately since we probably can't use .md includes in docstrings themselves. Sorry for the confusion. We should undo the changes in the rnn.py files, and keep cudnn_rnn_determinism.rst, cudnn_persistent_rnn.rst as .rst files.

It looks like it doesn't like the reference type of link in the docstring itself. You can change the ref to use a direct link instead:
multiprocessing context <https://docs.python.org/3/library/multiprocessing.html#contexts-and-start-methods>_ in here and delete:

Ahhh now the docstring is too wide:

  Advice (FLAKE8) B950
    line too long (174 > 120 characters)
    See https://github.com/PyCQA/flake8-bugbear#list-of-warnings

         175  |            worker subprocess with the worker id (an int in ``[0, num_workers - 1]``) as
         176  |            input, after seeding and before data loading. (default: ``None``)
         177  |        multiprocessing_context (str or multiprocessing.context.BaseContext, optional): If
    >>>  178  |            ``None``, the default `multiprocessing context <https://docs.python.org/3/library/multiprocessing.html#contexts-and-start-methods>`_ of your operating system will

Any way around this?

One sec - data.md has no git log history. I'll break this into another commit to keep it

you can just undo your changes to data.rst on this branch (save them somewhere), then:

1. From docs dir: git checkout origin/main -- source/optim.rst - delete .md I guess as well.
2. git mv source/data.rst source/data.md
3. Make a commit: git commit -m "Renamed source/data.rst to source/data.md".
4. Can check git log: git log -- follow -- data.md - this should show you a bunch of commits.
5. Push into this PR the above commit.
6. Copy paste changes you've saved in step 1 into data.md. Commit/push into this PR.

Yep should be good now.

LGTM!
@pytorchbot merge -f "Dos change only, doc signal is green."

Merge started

Your change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Please use -f as last resort and instead consider -i/--ignore-current to continue the merge ignoring current failures. This will allow currently pending tests to finish and report signal before the merge.

Learn more about merging in the wiki.

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