Recommend cpython agnosticism in cpp custom op tutorial #3250
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/tutorials/3250
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 7b1ab2c with merge base 15ef015 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
janeyx99
changed the title
advanced_source/cpp_custom_ops.rst
Outdated
| This ``setup`` would build one wheel that could be installed across multiple CPython | ||
| versions ``>=3.9``. |
There was a problem hiding this comment.
Uh, does the wheel have the right platform tags to reject Python 3.8? Or is the advice to people to always use the smallest version that is not deprecated?
There was a problem hiding this comment.
the latter hehe
i haven't checked against 3.8, but it was cuz i chose 3.9 as the min version out there. will include a sentence for that
There was a problem hiding this comment.
Yeah we need to tell people that the numbers they choose should be the min version pytorch supports
janeyx99
force-pushed
the
python-agnostic-1st
branch
from
9e388d2 to
1372553
Compare
Description
In 2.6, PyTorch will better support python agnosticism in custom operators. This PR updates the tutorial to recommend the torch.ops.load_library path so that future custom op writers can follow this better pattern instead. These changes are paired with updates to our example extension-cpp repo that have now landed: pytorch/extension-cpp#105
Checklist