Request: to open-source the torch_directml native extension now that DirectML is in maintenance mode
Nobody has claimed this yet.
Assessment
- Difficulty
- 5/5
- Estimated time
- Over a week
- Newbie friendliness
- 15/100
- Issue type
- Feature
- Clarity
- Mostly clear
- Activity status
- Stale
- Domain
- build-system, machine-learning
Research direction
The request concerns the closed C++/pybind11 layer that produces torch_directml_native.*.pyd, but it names no source files or tests. Start by reviewing the repository and package contents to identify the native extension boundary and its build inputs. Done would mean the binding source and a supported way to rebuild it are publicly available.
Written by the indexing model from the issue text.
Description
Hi Microsoft team,
Thank you for your work on DirectML and for making it available to the community. As noted in the repository, "DirectML is in maintenance mode", which implies that the project is no longer under active feature development but will continue to be supported.
Given this status, I’d like to respectfully request that the native Python extension source code (i.e., the C++/pybind11 layer that builds torch_directml_native.*.pyd) be made publicly available.
Currently, only the precompiled .pyd binaries are distributed via PyPI, while the binding code that connects PyTorch tensors to DirectML operators remains closed. This creates several practical limitations:
- Users cannot rebuild the package for custom environments (e.g., specific Python versions, embedded systems, or air-gapped deployments).
- Debugging, auditing, or extending the backend (e.g., adding experimental operators or modifying tensor layout handling) is impossible.
- The project cannot be considered truly open source if a core component—the Python-native bridge—is not available for inspection or contribution.
Since Microsoft has already open-sourced the DMLX operator library, headers, and samples, releasing the pybind11 binding layer would align with the spirit of open collaboration—especially for a project now in maintenance mode, where community contributions could help sustain usability and portability.
Open-sourcing this final piece would greatly empower researchers, enterprise users, and developers who rely on DirectML as a cross-vendor GPU inference solution on Windows.
Thank you for considering this request!
- Dominant language
- C++
- Stars
- 2.6k
- Forks
- 338
- PR merge metrics
- No merged PRs in 30d
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from microsoft/DirectML
-
Difficulty 1/5 Under an hour Newbie friendliness 68/100
-
Difficulty 4/5 3-5 days Newbie friendliness 55/100
-
Difficulty 4/5 3-5 days Newbie friendliness 62/100
-
Difficulty 4/5 3-5 days Newbie friendliness 25/100
-
Difficulty 4/5 3-5 days Newbie friendliness 45/100
All issues in microsoft/DirectML
Similar issues
-
Difficulty 1/5 Under an hour Newbie friendliness 90/100
AXERA-TECH/ax-llm#77 ·
-
Difficulty 1/5 Under an hour Newbie friendliness 90/100
games-on-whales/wolf#509 ·
-
Difficulty 2/5 1-3 hours Newbie friendliness 74/100
-
bug-unconfirmed
Difficulty 2/5 1-3 hours Newbie friendliness 76/100
-
Difficulty 2/5 1-3 hours Newbie friendliness 74/100
NVIDIA/cuda-samples#453 ·