fix: rollback BUCC revision

[Samoed]

Checklist

• Run tests locally to make sure nothing is broken using make test.
• Run the formatter to format the code using make lint.

Closing: #1694

Results for sentence-transformers/LaBSE, since I can't reproduce results for intfloat/multilingual-e5-small. Maybe they used for query: as prefix here.

	Leaderboard (BUCC)	PR (BUCC)	BUCC.v2
de-en	99.35	99.3633	99.4294
fr-en	98.72	98.716	98.9489
ru-en	97.78	97.7755	97.4614
zh-en	99.16	99.1575	99.2277

[KennethEnevoldsen]

Generally intfloat/multilingual-e5-small is a bit hard to reproduce if not run by our implementation and even if it is, the task prompts might change making it harder to reproduce (we could consider logging this).

[KennethEnevoldsen]

shouldn't we change the dataset instead?

[Samoed]

I tried to change it, but it resulted in a different number of sentence1 and sentence2, which prevented me from creating a dataset.Dataset.

[isaac-chung]

I'm not sure if any changes here address this comment in the linked issue. Could you please take a look at that and the paper to confirm what's missing? (i.e. wrong split used/uploaded, should be a cross-lingual retrieval task)

[Samoed]

For now, I've made changes to ensure that BUCC is working. After that, we can investigate how it’s supposed to work and deprecate this

[isaac-chung]

How it's supposed to work

It's ok to do this in a separate PR. As mentioned here, hope my comment in the issue would give you a head start.