Torch>=2.0 Unittest support

[ktsitsi]

This PR:

• Extends the CI test suite by introducing tests for torch>=2.0 for all python versions.

This can unblock any pending work that requires the aforementioned dependency.

[bkmartinjr]

While we are touching this file, should update the action versions as they are generating warnings.

The following actions use a deprecated Node.js version and will be forced to run on node20: actions/checkout@v3, actions/setup-python@v4, actions/cache@v3. For more info: https://github.blog/changelog/2024-03-07-github-actions-all-actions-will-run-on-node20-instead-of-node16-by-default/

The latest are:

• actions/checkout@v4
• actions/setup-python@v5
• actions/cache@v4

[bkmartinjr]

Could we also add a version that runs on a more modern Python and PyTorch?

Suggest Python 3.12, Torch ~2.4 and torchdata 0.8 (latest)

[bkmartinjr]

Digging through the logs for the 2.0 run there are a error messages, but I'm not sure any of them matter.

2024-08-27T10:13:20.6013473Z   DEPRECATION: Legacy editable install of tiledb-ml[cloud]==0.1.dev1 from file:///home/runner/work/TileDB-ML/TileDB-ML (setup.py develop) is deprecated. pip 25.0 will enforce this behaviour change. A possible replacement is to add a pyproject.toml or enable --use-pep517, and use setuptools >= 64. If the resulting installation is not behaving as expected, try using --config-settings editable_mode=compat. Please consult the setuptools documentation for more information. Discussion can be found at https://github.com/pypa/pip/issues/11457
2024-08-27T10:13:20.6017632Z   Running setup.py develop for tiledb-ml
2024-08-27T10:13:21.7079650Z ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
2024-08-27T10:13:21.7081668Z tensorflow-cpu 2.13.0 requires numpy<=1.24.3,>=1.22, but you have numpy 2.0.2 which is incompatible.

[ktsitsi]

Digging through the logs for the 2.0 run there are a error messages, but I'm not sure any of them matter.

2024-08-27T10:13:20.6013473Z   DEPRECATION: Legacy editable install of tiledb-ml[cloud]==0.1.dev1 from file:///home/runner/work/TileDB-ML/TileDB-ML (setup.py develop) is deprecated. pip 25.0 will enforce this behaviour change. A possible replacement is to add a pyproject.toml or enable --use-pep517, and use setuptools >= 64. If the resulting installation is not behaving as expected, try using --config-settings editable_mode=compat. Please consult the setuptools documentation for more information. Discussion can be found at https://github.com/pypa/pip/issues/11457
2024-08-27T10:13:20.6017632Z   Running setup.py develop for tiledb-ml
2024-08-27T10:13:21.7079650Z ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
2024-08-27T10:13:21.7081668Z tensorflow-cpu 2.13.0 requires numpy<=1.24.3,>=1.22, but you have numpy 2.0.2 which is incompatible.

For the pytorch implementation doesn't matter but I guess since we fully install the tiledb-ml package inside the UDF images this will create some issues with the tensorflow distribution. The error makes sense since pytorch>=2.0 and especially torchvision seems to require numpy>=2.0 where the latest version we support for TF is 2.13 is conflicting with that since it has an upper bound. We tried to increase the supported versions for TF but it will require some time/capacity. Interestingly the issue you pointed out appears only for PY-3.9. I will take a further look in case we wanna make sure that nothing breaks.

[bkmartinjr]

The error makes sense since pytorch>=2.0 and especially torchvision seems to require numpy>=2.0

If you use one release back, I think it works:

pip install -f https://download.pytorch.org/whl/torch_stable.html protobuf==3.* torch==2.3.1+cpu torchvision==0.18.1+cpu torchdata torchaudio==2.3.1 tensorflow-cpu 'numpy<=2'

This gets me numpy 1.26.4 without complaint

[bkmartinjr]

Looking at the latest fail, the tensorflow-cpu pin is too low. In my succesfull install, I believe I used an unpinned tensorflow-cpu, which ended up installing 2.17.0