CI: Run lit tests on linux

[Pangoraw]

testing this one

[giordano]

They're bound to fail: #369 (comment)

[Pangoraw]

I am filtering to run only the lit tests //lit_tests/*

[giordano]

Tangentially related, I'm testing using Python 3.11 to solve some issues (installing tensorflow with python 3.12 fails because of distutils being removed). That should solve a bunch of issues, but on hydra I still have segfaults in some tests (as I had mentioned in #426 (comment) already)

[giordano]

I am filtering to run only the lit tests //lit_tests/*

Ah, testing only those may work, yes.

[Pangoraw]

It's better than nothing and in the context of GB, python tests are not as important IMHO.

[giordano]

Tangentially related, I'm testing using Python 3.11 to solve some issues

For the record, can confirm that with Python 3.11 packaging issues are solved, but two tests still fail:

• //test:keras_test times out
• //test:neuralgcm_test segfaults, as I also experienced locally
• //test:llama times out on aarch64

If we can exclude these 2-3 tests we can run all the rest (no idea how it's easy to do filter out few tests with Bazel)

[Pangoraw]

That's nice we can replace the filter with something like

bazel test ... $(bazel query //test/... 2>/dev/null | grep -v "keras_test" | grep -v "neuralgcm_test" | grep -v "llama")