Add python bindings #110
Merged
Add python bindings #110
Conversation
This file contains 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
reverendbedford
requested changes
Dec 18, 2023
There was a problem hiding this comment.
- It may be worthwhile to setup the basic python packaging stuff now (i.e. a
setup.py). I don't know if we'll ever want to make this available as a python package directly, but we might. And I'm not sure how compatible your directory layout here will be if we do that. You can look at NEML for asetup.pythat works with cmake. - What is the actual minimum torch and python version required? If we really need python3.11 that could be annoying, as it's not standard on Ubuntu yet.
- The tests error on python3.10 with a legitimate syntax error. Probably should fix that unless 3.11 is really needed.
- We should have the python test action loop over python versions in our valid range. Maybe the same thing with torch, but I'm okay with just forcing people to 2.1 if that's what's needed.
I'll try to add a pyproject.toml
I dropped the 3.11 requirement and fixed the unpacking syntax error.
I agree, if github does not have any limit on the number of checks... |
|
lol I forgot ubuntu 22.04 does not have python 3.6 in the main universe. Should we switch to conda or just drop python 3.6 support? I'm fine with either solution. |
|
This is ready again. |
reverendbedford
requested changes
Mar 7, 2024
There was a problem hiding this comment.
- Add check that the python source has been blacked (
black --check). - I have a fun bug where I installed the package, uninstalled it, and then tried to install it again. It will not build. The problem is with linking to torch. Delete the build folder and trying again fixed the issue. Uhh, maybe try it on Linux and see what happens?
- It would be nice to be able to construct both
BatchTensorand the derived classes ofFixedDimTensorfrom torch tensors and numpy arrays. I think you just need to implement thetorch.tensorvariant and then rely on the torch tensor constructor to handle the rest of the conversions. They must have a "ensure_tensor" or similar method like numpy. ForBatchTensorthe user would supply the tensor and anintindicating the separation between the batch and base dims. ForFixedDimTensorwe should be able to infer all the dimensions (and throw if necessary) without help. - I don't think you test the
torch.tensorview into our classes. A few simple tests would be nice (values are the same, if you change the tensor view you change the underlying neml2 tensor data, devices are the same).
|
I searched online but couldn't find any way of locating the pytorch package without first finding python. I would argue that what I have right now is fine, because if python is not found, it will merrily fall back to plan B (downloading a cpu libtorch). So strictly speaking, we are not introducing python as a dependency (if we are not going to compile python bindings). |
|
Also note that I renamed the github workflow recipe "Formatting Check" to "C++ Formatting Check" (because we now have a "Python Formatting Check"). So you will need to modify the list of expected checks. |
|
A few "new" things to review:
I let the wheels use a special rpath to avoid bundling all the libtorch dependencies into our wheel. We probably want to discuss about this. |
|
I haven't added the "upload to pypi" step yet. We may want to play with the wheels on more machines before uploading to pypi... |
|
At a higher level, we should probably think about how to auto-update python wheels to stay up-to-date with pytorch updates. I am also not entirely satisfied with the hard-coded torch versions scattered across the repository. |
reverendbedford
requested changes
Apr 15, 2024
There was a problem hiding this comment.
- See comment on minimum cmake version.
- We should probably include "build and test" for python bindings in the github actions.
- I suggest have a
load_modelhelper that combinesload_inputandget_model. - Perhaps add a parameter partial type test to
test_model.pyas that's going to be important shortly.
I will make a new PR to actually upload the pypi package after we merge this.
Good work, nearly done!
… need a trampoline class
|
I opened a future issue after merging this PR: #130 |
Python Binding Test Results (macos-latest63 tests 39 ✅ 2s ⏱️ Results for commit c2a467d. |
|
And also #131 |
Python Binding Test Results (ubuntu-latest)63 tests 39 ✅ 2s ⏱️ Results for commit 6a60dd0. |
Python Binding Test Results (macos-latest)63 tests 39 ✅ 1m 20s ⏱️ Results for commit 6a60dd0. |
|
Coverage after merging pybind into main will be
Coverage Report
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
reverendbedford
approved these changes
Apr 17, 2024
Closed
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Add python bindings
CMake configure