Remove deprecated torchpippy dependency (Issue #3846) #3865
+50
−48
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Remove torchpippy>=0.2.0 from the test_dev extras dependencies list in setup.py. This dependency is no longer required for the test development environment.
…lelism Update the PiPPy examples README to document the migration from the separate torchpippy package to PyTorch's built-in pipeline parallelism (torch.distributed.pipelining) available in PyTorch 2.4.0+. Changes include: - Update title from 'PiPPy' to 'Pipeline Parallelism' - Replace references to PiPPy library with PyTorch's built-in torch.distributed.pipelining - Update installation instructions to require PyTorch 2.4.0+ instead of torchpippy>=0.2.0 - Add note explaining that torchpippy has been merged into PyTorch 2.4.0+ and no additional package installation is needed
Update the RuntimeError message in t5.py to: - Use the correct module name 'torch.distributed.pipelining' instead of 'torch.pipelining' - Update the message to reflect that torchpippy has been merged into PyTorch 2.4.0+ as torch.distributed.pipelining, removing the outdated reference to using torchpippy separately
…pport Update the docstring for the require_pippy decorator to reflect that it now requires PyTorch 2.4.0+ (which includes torch.distributed.pipelining) instead of requiring a separate pippy package installation. The docstring now clearly states that tests are skipped when PyTorch < 2.4.0, aligning with the migration from torchpippy to PyTorch's built-in pipeline parallelism support.
Apply automatic code formatting changes from 'make style' command using ruff check --fix and ruff format. This includes: - Formatting fixes in example files (complete_cv_example.py, cv_example.py) - Code style improvements in core accelerate modules (accelerator.py, optimizer.py, scheduler.py, state.py, tracking.py) - Formatting updates in utility modules (dataclasses.py, deepspeed.py, megatron_lm.py, modeling.py, operations.py) - Style fixes in test utilities (testing.py) and tests (test_tracking.py) - Code formatting in command modules (estimate.py) All changes are formatting-only and do not affect functionality.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Remove deprecated torchpippy dependency
Fixes #3846
What does this PR do?
This PR addresses issue #3846 by removing the deprecated
torchpippydependency from the codebase. Thetorchpippypackage has been merged into PyTorch 2.4.0+ astorch.distributed.pipelining, making the separate package dependency obsolete.Changes Made:
Removed dependency from setup.py: Removed
torchpippy>=0.2.0fromextras["test_dev"]insetup.pysince pipeline parallelism is now available natively in PyTorch 2.4.0+Updated documentation:
examples/inference/pippy/README.mdto reflect that PyTorch 2.4.0+ is required (which includestorch.distributed.pipelining) instead of the separatetorchpippypackagetorchpippyhas been merged into PyTorchUpdated error messages:
examples/inference/pippy/t5.pyto correctly referencetorch.distributed.pipeliningand mention the PyTorch mergeUpdated docstrings:
require_pippydecorator docstring insrc/accelerate/test_utils/testing.pyto clarify that it checks for PyTorch 2.4.0+ rather than a separate packageNote: Any other files changed in this PR are related to running
make style, which automatically formats code according to the project's style guidelines (ruff formatting).Motivation:
The
torchpippyrepository has been archived and its functionality merged into PyTorch core. Maintaining a dependency on the deprecated package is unnecessary and can cause confusion for users. The codebase already usestorch.distributed.pipeliningfrom PyTorch 2.4.0+, so this change aligns the dependencies and documentation with the actual implementation.Testing:
is_pippy_available()correctly checks for PyTorch >= 2.4.0torch.distributed.pipeliningfrom PyTorch core@require_pippywill continue to work as they check for PyTorch version, not the package