feat: experiment with different dimensions for OpenAI models on MTEB(Medical)

[dbuades]

As discussed in #71, this PR evaluates both text-embeddings-3-small (256, 512, 768, 1024) and text-embeddings-3-large (256, 512, 768, 1024, 1536) on the MTEB(Medical) benchmark.

This is a total of 9 combinations, currently stored in 9 separate folders. Code used to run the models can be found in this commit, following the approach described here.

I'll leave the PR in draft for now and we can use it to explore different ways to display experiments in the leaderboard without adding new model duplicates.

Thanks again @Muennighoff for providing the API key!

Checklist

• Run tests locally to make sure nothing is broken using make test.
• Run the results files checker make pre-push.

[KennethEnevoldsen]

Thanks for you work on this @dbuades. I believe we are currently working on transitioning to v2 where I don't think this feature is planned. So will close this for now, but I would love to add it in the future.

Again thanks PR, I simply think there is no one who has time to finalize this feature (@sam-hey might be interested - if you are let me know)

[sam-hey]

@KennethEnevoldsen A bit short on time this month, but I'll take a look next month

[Muennighoff]

I think we can still merge this given the folders for those models already exist? I think they were also displayed on the previous leaderboard so maybe just displaying them in the same way on the new leaderboard 🤔

[Samoed]

CC @x-tabdeveloping

[dbuades]

No worries, I understand! If merging them as-is is not appropriate and there’s any way I can help with the implementation, please let me know. I think these results are interesting since they highlight the impact of the dimensions. If you have any other preliminary designs, I have bandwidth next week to work on them.