fix: Allow aggregated tasks within benchmarks #1771
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Samoed
reviewed
Jan 11, 2025
- Updated task filtering adding exclusive_language_filter and hf_subset
- fix bug in MTEB where cross-lingual splits were included
- added missing language filtering to MTEB(europe, beta) and MTEB(indic, beta)
The following code outlines the problems:
```py
import mteb
from mteb.benchmarks import MTEB_ENG_CLASSIC
task = [t for t in MTEB_ENG_CLASSIC.tasks if t.metadata.name == "STS22"][0]
# was eq. to:
task = mteb.get_task("STS22", languages=["eng"])
task.hf_subsets
# correct filtering to English datasets:
# ['en', 'de-en', 'es-en', 'pl-en', 'zh-en']
# However it should be:
# ['en']
# with the changes it is:
task = [t for t in MTEB_ENG_CLASSIC.tasks if t.metadata.name == "STS22"][0]
task.hf_subsets
# ['en']
# eq. to
task = mteb.get_task("STS22", hf_subsets=["en"])
# which you can also obtain using the exclusive_language_filter (though not if there was multiple english splits):
task = mteb.get_task("STS22", languages=["eng"], exclusive_language_filter=True)
```
2 tasks
…low-aggregated-tasks-within-benchmarks
Samoed
reviewed
Jan 17, 2025
Samoed
reviewed
Jan 17, 2025
…nto KennethEnevoldsen/issue-Allow-aggregated-tasks-within-benchmarks
10 tasks
…nto KennethEnevoldsen/issue-Allow-aggregated-tasks-within-benchmarks
|
Leaderboard seemingly works - @Samoed is it possible that we did not scrape the CQADupstack scores (if so can we do that?) I will also go over the results repo to and manually create a merged dict for all model which have the scores |
|
No, I think we scrape all of them, e.g. e5-small has CQADupstackWordpressRetrieval in external dir https://github.com/embeddings-benchmark/results/blob/6f32ccc978e2d84b905f10b2848479881f43ab02/results/intfloat__e5-small/external/CQADupstackWordpressRetrieval.json#L4 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes #1231
Adresses #1763 (though we need a fix for the results repo)
Did quite a few refactors here. Not at all settled that this is the right representation, but it is at least much better than where we started.
We will still need to combine the score of CQGA scores on
embedding-benchmark/results.@x-tabdeveloping let me know if this works on the leaderboard end (I believe it should)
Checklist
make test.make lint.