feat: Add Apple Silicon (MPS) support #559
Merged
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
This change adds support for Apple Silicon, allowing the library to run on MPS-compatible devices. Key changes: - Modified `xfuser/envs.py` to detect and support the MPS backend. - Replaced hardcoded `.cuda()` calls with dynamic device placement (`.to(device)`). - Made CUDA-specific dependencies like `yunchang` and `flash-attn` conditional, preventing import errors on non-CUDA systems. - Updated `setup.py` to be more platform-agnostic. - Added a new test file `tests/core/test_envs.py` to verify the new device-detection logic. Limitation: The `yunchang` library for long context attention is CUDA-specific and is therefore disabled when running on MPS. This means that long context attention features will not be available on Apple Silicon devices.
This commit fixes a `TypeError` that occurred when running the library on an MPS device. The error was caused by `get_device_version()` returning `None` for MPS, which `packaging.version.parse()` cannot handle.
The fix modifies the lambda function for `CUDA_VERSION` in `xfuser/envs.py` to provide a default version string ("0.0") when `get_device_version()` returns `None`.
A new test case has also been added to `tests/core/test_envs.py` to ensure that accessing `envs.CUDA_VERSION` in a mocked MPS environment does not raise an error, preventing future regressions.
This commit pins the `torch` version to `2.4.1` in `setup.py`. This is to resolve dependency conflicts with other libraries like `torchaudio` and `torchvision` that may be present in the user's environment and require a specific version of `torch`. This change will ensure a more stable and predictable installation process for users on all platforms.
feifeibear
reviewed
Sep 9, 2025
| packages=find_packages(), | ||
| install_requires=[ | ||
| "torch>=2.1.0", | ||
| "torch==2.4.1", |
Collaborator
There was a problem hiding this comment.
can you make it as torch>=2.4.1
feifeibear
approved these changes
Sep 14, 2025
This was referenced Sep 15, 2025
Closed
This was referenced Oct 17, 2025
Merged
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
This change adds support for Apple Silicon, allowing the library to run on MPS-compatible devices.
Key changes:
xfuser/envs.pyto detect and support the MPS backend..cuda()calls with dynamic device placement (.to(device)).yunchangandflash-attnconditional, preventing import errors on non-CUDA systems.setup.pyto be more platform-agnostic.tests/core/test_envs.pyto verify the new device-detection logic.Limitation:
The
yunchanglibrary for long context attention is CUDA-specific and is therefore disabled when running on MPS. This means that long context attention features will not be available on Apple Silicon devices.