Fixes flakey integration test/syncs deps

[aittalam]

What's changing

I updated pyproject.toml to use more recent libs. The main reason was transformers so we could run more recent models but I ran a general update using pip-review to get latest versions. These are the steps I followed:

• used pip-review to update all libs
• check there were no inconsistencies
• test lm-buddy with the new set of libs
• updated pyproject.toml by picking the updated versions from the output of pip freeze
• re-installed everything in a new env and made sure there were no inconsistencies
• tested lm-buddy again

How to test it

The tests I ran were unit/integration tests + manually running lm-buddy for summarization eval (currently our main job).
Interestingly, tests/integration/test_lm_harness.py::test_lm_harness_job passes when I test locally

[aittalam]

Two questions for @binaryaaron:

1. ruff on github returns an error which I do not get at precommit nor when I run it manually. Is there anything else I should change in the repo to make sure the two are aligned? (nothing major, I'd just like to prevent errors happening here if I can)

2. I still wonder what is different between my local tests and github's as they all pass locally (in particular the lm-harness one which breaks online):

[binaryaaron]

I added some minor changes and changed the test task to one that doesn't require HF remote code. there's something odd with how those args are being passed; i opened #115 to revisit this after the demo.

[aittalam]

I added some minor changes and changed the test task to one that doesn't require HF remote code. there's something odd with how those args are being passed; i opened #115 to revisit this after the demo.

That's great, many thanks!