Skip to content

Conversation

@awgu
Copy link
Collaborator

@awgu awgu commented Sep 30, 2024

Stack from ghstack (oldest at bottom):

This PR relaxes the even sharding requirement for the all-gather extensions.

The fsdp_pre_all_gather now expects signature:

def fsdp_pre_all_gather(
    self,
    mesh: DeviceMesh,
+    outer_size: torch.Size,
+    outer_stride: Tuple[int, ...],
    module: nn.Module,
    mp_policy: MixedPrecisionPolicy,
) -> Tuple[Tuple[torch.Tensor, ...], Any]:
  • Since no one is using this new signature yet, we should be safe to change it.
  • Currently, the outer_stride will always be contiguous strides since FSDP2 only supports contiguous strides for now.
  • For the uneven sharding case, the user is responsible to return a padded sharded tensor from fsdp_pre_all_gather. This is risky territory because if the user does not do so, then this may manifest as a NCCL timeout, as only the ranks with padding will error out. However, I am not aware of any way around this.

cc @XilunWu @H-Huang @kwen2501 @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @c-p-i-o

@pytorch-bot
Copy link

pytorch-bot bot commented Sep 30, 2024

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/137005

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

❌ 4 New Failures

As of commit 7767d23 with merge base 8c29a0d (image):

NEW FAILURES - The following jobs have failed:

  • linux-binary-libtorch-pre-cxx11 / libtorch-cpu-shared-with-deps-pre-cxx11-test / test (gh)
    RuntimeError: recursive_directory_iterator in used pre-CXX11 binaries, see; ['std::filesystem::recursive_directory_iterator::recursion_pending() const', 'std::filesystem::recursive_directory_iterator::depth() const', 'std::filesystem::recursive_directory_iterator::options() const', 'std::filesystem::recursive_directory_iterator::operator*() const', 'std::filesystem::recursive_directory_iterator::disable_recursion_pending()', 'std::filesystem::recursive_directory_iterator::pop(std::error_code&)', 'std::filesystem::recursive_directory_iterator::pop()', 'std::filesystem::recursive_directory_iterator::pop() [clone .cold]', 'std::filesystem::recursive_directory_iterator::increment(std::error_code&)', 'std::filesystem::recursive_directory_iterator::increment(std::error_code&) [clone .cold]', 'std::filesystem::recursive_directory_iterator::operator=(std::filesystem::recursive_directory_iterator&&)', 'std::filesystem::recursive_directory_iterator::operator=(std::filesystem::recursive_directory_iterator const&)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*) [clone .cold]', 'std::filesystem::recursive_directory_iterator::~recursive_directory_iterator()', 'std::filesystem::recursive_directory_iterator::~recursive_directory_iterator()', 'std::filesystem::recursive_directory_iterator::operator++()', 'std::filesystem::recursive_directory_iterator::operator++() [clone .cold]']
  • linux-binary-manywheel / manywheel-py3_9-cuda11_8-test / test (gh)
    RuntimeError: recursive_directory_iterator in used pre-CXX11 binaries, see; ['std::filesystem::recursive_directory_iterator::recursion_pending() const', 'std::filesystem::recursive_directory_iterator::depth() const', 'std::filesystem::recursive_directory_iterator::options() const', 'std::filesystem::recursive_directory_iterator::operator*() const', 'std::filesystem::recursive_directory_iterator::disable_recursion_pending()', 'std::filesystem::recursive_directory_iterator::pop(std::error_code&)', 'std::filesystem::recursive_directory_iterator::pop()', 'std::filesystem::recursive_directory_iterator::pop() [clone .cold]', 'std::filesystem::recursive_directory_iterator::increment(std::error_code&)', 'std::filesystem::recursive_directory_iterator::increment(std::error_code&) [clone .cold]', 'std::filesystem::recursive_directory_iterator::operator=(std::filesystem::recursive_directory_iterator&&)', 'std::filesystem::recursive_directory_iterator::operator=(std::filesystem::recursive_directory_iterator const&)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*) [clone .cold]', 'std::filesystem::recursive_directory_iterator::~recursive_directory_iterator()', 'std::filesystem::recursive_directory_iterator::~recursive_directory_iterator()', 'std::filesystem::recursive_directory_iterator::operator++()', 'std::filesystem::recursive_directory_iterator::operator++() [clone .cold]']
  • linux-binary-manywheel / manywheel-py3_9-cuda12_1-test / test (gh)
    RuntimeError: recursive_directory_iterator in used pre-CXX11 binaries, see; ['std::filesystem::recursive_directory_iterator::recursion_pending() const', 'std::filesystem::recursive_directory_iterator::depth() const', 'std::filesystem::recursive_directory_iterator::options() const', 'std::filesystem::recursive_directory_iterator::operator*() const', 'std::filesystem::recursive_directory_iterator::disable_recursion_pending()', 'std::filesystem::recursive_directory_iterator::pop(std::error_code&)', 'std::filesystem::recursive_directory_iterator::pop()', 'std::filesystem::recursive_directory_iterator::pop() [clone .cold]', 'std::filesystem::recursive_directory_iterator::increment(std::error_code&)', 'std::filesystem::recursive_directory_iterator::increment(std::error_code&) [clone .cold]', 'std::filesystem::recursive_directory_iterator::operator=(std::filesystem::recursive_directory_iterator&&)', 'std::filesystem::recursive_directory_iterator::operator=(std::filesystem::recursive_directory_iterator const&)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*) [clone .cold]', 'std::filesystem::recursive_directory_iterator::~recursive_directory_iterator()', 'std::filesystem::recursive_directory_iterator::~recursive_directory_iterator()', 'std::filesystem::recursive_directory_iterator::operator++()', 'std::filesystem::recursive_directory_iterator::operator++() [clone .cold]']
  • linux-binary-manywheel / manywheel-py3_9-cuda12_4-test / test (gh)
    RuntimeError: recursive_directory_iterator in used pre-CXX11 binaries, see; ['std::filesystem::recursive_directory_iterator::recursion_pending() const', 'std::filesystem::recursive_directory_iterator::depth() const', 'std::filesystem::recursive_directory_iterator::options() const', 'std::filesystem::recursive_directory_iterator::operator*() const', 'std::filesystem::recursive_directory_iterator::disable_recursion_pending()', 'std::filesystem::recursive_directory_iterator::pop(std::error_code&)', 'std::filesystem::recursive_directory_iterator::pop()', 'std::filesystem::recursive_directory_iterator::pop() [clone .cold]', 'std::filesystem::recursive_directory_iterator::increment(std::error_code&)', 'std::filesystem::recursive_directory_iterator::increment(std::error_code&) [clone .cold]', 'std::filesystem::recursive_directory_iterator::operator=(std::filesystem::recursive_directory_iterator&&)', 'std::filesystem::recursive_directory_iterator::operator=(std::filesystem::recursive_directory_iterator const&)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*) [clone .cold]', 'std::filesystem::recursive_directory_iterator::~recursive_directory_iterator()', 'std::filesystem::recursive_directory_iterator::~recursive_directory_iterator()', 'std::filesystem::recursive_directory_iterator::operator++()', 'std::filesystem::recursive_directory_iterator::operator++() [clone .cold]']

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

@pytorch-bot pytorch-bot bot added oncall: distributed Add this issue/PR to distributed oncall triage queue release notes: distributed (fsdp) release notes category labels Sep 30, 2024
awgu pushed a commit that referenced this pull request Sep 30, 2024
@awgu awgu added release notes: distributed (fsdp2) release notes category and removed release notes: distributed (fsdp) release notes category labels Sep 30, 2024
…xtensions"

cc XilunWu H-Huang kwen2501 wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o

[ghstack-poisoned]
awgu pushed a commit that referenced this pull request Oct 2, 2024
@awgu awgu requested a review from weifengpy October 4, 2024 16:33
@awgu
Copy link
Collaborator Author

awgu commented Oct 4, 2024

cc: @weifengpy for your thoughts on this

out: Optional[torch.Tensor] = None,
) -> Union[Tuple[torch.Tensor, Tuple[torch.Tensor, ...]], None]:
assert metadata is None, f"{metadata}"
(tensor,) = all_gather_outputs
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

curious where we removed the padding? the unit test works so it should be handled.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, good question. FSDP will trim the padding for you given the original size of the parameter.

@awgu awgu added the ciflow/trunk Trigger trunk jobs on your pull request label Oct 4, 2024
@awgu awgu marked this pull request as ready for review October 4, 2024 17:01
@awgu
Copy link
Collaborator Author

awgu commented Oct 4, 2024

@pytorchbot merge

@pytorchmergebot
Copy link
Collaborator

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

@pytorchmergebot
Copy link
Collaborator

Merge failed

Reason: 1 jobs have failed, first few of them are: linux-binary-libtorch-pre-cxx11 / libtorch-cpu-shared-with-deps-pre-cxx11-test / test

Details for Dev Infra team Raised by workflow job

@awgu
Copy link
Collaborator Author

awgu commented Oct 4, 2024

linux-binary-libtorch-pre-cxx11 / libtorch-cpu-shared-with-deps-pre-cxx11-test / test failure is not related:

Traceback (most recent call last):
  File "/builder/test/check_binary_symbols.py", line 109, in <module>
    main()
  File "/builder/test/check_binary_symbols.py", line 105, in main
    check_lib_symbols_for_abi_correctness(libtorch_cpu_path, pre_cxx11_abi)
  File "/builder/test/check_binary_symbols.py", line 86, in check_lib_symbols_for_abi_correctness
    raise RuntimeError(f"recursive_directory_iterator in used pre-CXX11 binaries, see; {rec_iter_symbols}")
RuntimeError: recursive_directory_iterator in used pre-CXX11 binaries, see; ['std::filesystem::recursive_directory_iterator::recursion_pending() const', 'std::filesystem::recursive_directory_iterator::depth() const', 'std::filesystem::recursive_directory_iterator::options() const', 'std::filesystem::recursive_directory_iterator::operator*() const', 'std::filesystem::recursive_directory_iterator::disable_recursion_pending()', 'std::filesystem::recursive_directory_iterator::pop(std::error_code&)', 'std::filesystem::recursive_directory_iterator::pop()', 'std::filesystem::recursive_directory_iterator::pop() [clone .cold]', 'std::filesystem::recursive_directory_iterator::increment(std::error_code&)', 'std::filesystem::recursive_directory_iterator::increment(std::error_code&) [clone .cold]', 'std::filesystem::recursive_directory_iterator::operator=(std::filesystem::recursive_directory_iterator&&)', 'std::filesystem::recursive_directory_iterator::operator=(std::filesystem::recursive_directory_iterator const&)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*) [clone .cold]', 'std::filesystem::recursive_directory_iterator::~recursive_directory_iterator()', 'std::filesystem::recursive_directory_iterator::~recursive_directory_iterator()', 'std::filesystem::recursive_directory_iterator::operator++()', 'std::filesystem::recursive_directory_iterator::operator++() [clone .cold]']

@awgu
Copy link
Collaborator Author

awgu commented Oct 4, 2024

@pytorchbot merge -i

@pytorchmergebot
Copy link
Collaborator

Merge started

Your change will be merged while ignoring the following 3 checks: linux-binary-libtorch-pre-cxx11 / libtorch-cpu-shared-with-deps-pre-cxx11-test / test, linux-binary-manywheel / manywheel-py3_9-cuda12_4-test / test, linux-binary-manywheel / manywheel-py3_9-cuda12_1-test / test

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

@pytorchmergebot
Copy link
Collaborator

Merge failed

Reason: 1 jobs have failed, first few of them are: linux-binary-manywheel / manywheel-py3_9-cuda11_8-test / test

Details for Dev Infra team Raised by workflow job

@awgu
Copy link
Collaborator Author

awgu commented Oct 4, 2024

Same issue, looks unrelated:

RuntimeError: recursive_directory_iterator in used pre-CXX11 binaries, see; ['std::filesystem::recursive_directory_iterator::recursion_pending() const', 'std::filesystem::recursive_directory_iterator::depth() const', 'std::filesystem::recursive_directory_iterator::options() const', 'std::filesystem::recursive_directory_iterator::operator*() const', 'std::filesystem::recursive_directory_iterator::disable_recursion_pending()', 'std::filesystem::recursive_directory_iterator::pop(std::error_code&)', 'std::filesystem::recursive_directory_iterator::pop()', 'std::filesystem::recursive_directory_iterator::pop() [clone .cold]', 'std::filesystem::recursive_directory_iterator::increment(std::error_code&)', 'std::filesystem::recursive_directory_iterator::increment(std::error_code&) [clone .cold]', 'std::filesystem::recursive_directory_iterator::operator=(std::filesystem::recursive_directory_iterator&&)', 'std::filesystem::recursive_directory_iterator::operator=(std::filesystem::recursive_directory_iterator const&)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*)', 'std::filesystem::recursive_directory_iterator::recursive_directory_iterator(std::filesystem::path const&, std::filesystem::directory_options, std::error_code*) [clone .cold]', 'std::filesystem::recursive_directory_iterator::~recursive_directory_iterator()', 'std::filesystem::recursive_directory_iterator::~recursive_directory_iterator()', 'std::filesystem::recursive_directory_iterator::operator++()', 'std::filesystem::recursive_directory_iterator::operator++() [clone .cold]']

@awgu
Copy link
Collaborator Author

awgu commented Oct 4, 2024

@pytorchbot merge -i

@pytorchmergebot
Copy link
Collaborator

Merge started

Your change will be merged while ignoring the following 4 checks: linux-binary-libtorch-pre-cxx11 / libtorch-cpu-shared-with-deps-pre-cxx11-test / test, linux-binary-manywheel / manywheel-py3_9-cuda12_4-test / test, linux-binary-manywheel / manywheel-py3_9-cuda12_1-test / test, linux-binary-manywheel / manywheel-py3_9-cuda11_8-test / test

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

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ciflow/trunk Trigger trunk jobs on your pull request Merged oncall: distributed Add this issue/PR to distributed oncall triage queue release notes: distributed (fsdp2) release notes category

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants