Revert extraction setting for IndicesExtractionConfig (#998)

Revert `try_extract_without_anchor` to True in `IndicesExtractionConfig` to avoid issues in `gpqa:diamond` eval

* Change `try_extract_without_anchor` only for GPQA

* Update GPQA test to reflect the extract setting

Author: cmpatino

src/lighteval/metrics/metrics.py

@@ -526,8 +526,12 @@ class Metrics(Enum):
         metric_name="extractive_match",
         sample_level_fn=MultilingualExtractiveMatchMetric(
             language=Language.ENGLISH,
-            gold_extraction_target=[IndicesExtractionConfig(prefix_for_extraction="NativeLetters")],
-            pred_extraction_target=[IndicesExtractionConfig(prefix_for_extraction="NativeLetters")],
+            gold_extraction_target=[
+                IndicesExtractionConfig(prefix_for_extraction="NativeLetters", try_extract_without_anchor=True)
+            ],
+            pred_extraction_target=[
+                IndicesExtractionConfig(prefix_for_extraction="NativeLetters", try_extract_without_anchor=True)
+            ],
             precision=6,
         ),
         category=SamplingMethod.GENERATIVE,
@@ -539,8 +543,12 @@ class Metrics(Enum):
         sample_level_fn=PassAtK(
             sample_scoring_function=MultilingualExtractiveMatchMetric(
                 language=Language.ENGLISH,
-                gold_extraction_target=[IndicesExtractionConfig(prefix_for_extraction="NativeLetters")],
-                pred_extraction_target=[IndicesExtractionConfig(prefix_for_extraction="NativeLetters")],
+                gold_extraction_target=[
+                    IndicesExtractionConfig(prefix_for_extraction="NativeLetters", try_extract_without_anchor=True)
+                ],
+                pred_extraction_target=[
+                    IndicesExtractionConfig(prefix_for_extraction="NativeLetters", try_extract_without_anchor=True)
+                ],
                 precision=6,
             ),
         ),

tests/unit/metrics/test_cases/gpqa_instruct_metric.json

@@ -249,7 +249,7 @@
         ]
       },
       "expected_output": {
-        "extractive_match": 0.0
+        "extractive_match": 1.0
       },
       "tolerance": 0.01,
       "description": "Answer with quotes but still extractable"