Hi all,
I am looking for guidance regarding a PR I opened that adds pivoted QR decomposition to ATen and torch.linalg:
The PR includes tests and documentation and has been open for some time.
It is currently marked with the stale label and at risk of auto-closure due to missing reviewer bandwidth.
I realize that new features are typically discussed prior to merge requests.
I did not open a design discussion beforehand, but the feature was previously requested in:
- [feature request] Rank-Revealing QR - Adding dgeqp3 support to torch.qr · Issue #10454 · pytorch/pytorch · GitHub
- Add pivoted QR from MAGMA (PR wanted?) · Issue #82092 · pytorch/pytorch · GitHub
Both are referenced in the PR.
At this point, I would appreciate guidance on:
- Whether this functionality aligns with current PyTorch direction
- Whether the implementation approach seems acceptable at a high level
- Who would be the appropriate maintainer to review or sponsor this change
If the preferred path is to restructure this into a formal RFC or adjust the scope, I’m happy to do so.
Thanks for your time and guidance.