fix: subsets to run

mteb/abstasks/AbsTask.py

@@ -118,7 +118,7 @@ def evaluate(
             hf_subsets = copy(self.hf_subsets)
 
         if subsets_to_run is not None:  # allow overwrites of pre-filtering
-            hf_subsets = subsets_to_run
+            hf_subsets = [s for s in hf_subsets if s in subsets_to_run]
 
         for hf_subset in hf_subsets:
             logger.info(

tests/test_benchmark/mock_tasks.py

@@ -927,12 +927,10 @@ def load_data(self, **kwargs):
             ),
         }
 
-        self.dataset = DatasetDict(
-            {
-                "eng": data,
-                "fra": data,
-            }
-        )
+        self.dataset = {}
+        for lang in self.hf_subsets:
+            self.dataset[lang] = data
+
         self.data_loaded = True
 
     @property

tests/test_evaluation/test_split_evaluation.py

@@ -8,6 +8,7 @@
 )
 from tests.test_benchmark.mock_tasks import (
     MockMultilingualRetrievalTask,
+    MockMultilingualSTSTask,
     MockRetrievalTask,
 )
 
@@ -362,3 +363,12 @@ def test_all_splits_subsets_evaluated_with_overwrite(
     for split in ["test"]:
         assert len(results2[0].scores[split]) == 2
         assert sorted(results2[0].languages) == ["eng", "fra"]
+
+
+def test_splits_evaluated_with_prefiltering():
+    """Test that the evaluation only runs on the specified languages. Issue https://github.com/embeddings-benchmark/mteb/pull/1787#issuecomment-2598205049"""
+    task = MockMultilingualSTSTask().filter_languages(languages=["fra"])
+
+    evaluation = MTEB(tasks=[task])
+
+    evaluation.run(MockSentenceTransformer(), overwrite_results=True)