Pass is_reasoning_model through QAEvaluator

sdk/evaluation/azure-ai-evaluation/azure/ai/evaluation/_evaluators/_qa/_qa.py

@@ -101,12 +101,35 @@ def __init__(
             if not isinstance(value, (int, float)):
                 raise TypeError(f"{name} must be an int or float, got {type(value)}")
 
+        # Extract is_reasoning_model from kwargs to pass to LLM-based evaluators
+        is_reasoning_model = kwargs.get("is_reasoning_model", False)
+
         evaluators = [
-            GroundednessEvaluator(model_config, threshold=groundedness_threshold),
-            RelevanceEvaluator(model_config, threshold=relevance_threshold),
-            CoherenceEvaluator(model_config, threshold=coherence_threshold),
-            FluencyEvaluator(model_config, threshold=fluency_threshold),
-            SimilarityEvaluator(model_config, threshold=similarity_threshold),
+            GroundednessEvaluator(
+                model_config,
+                threshold=groundedness_threshold,
+                is_reasoning_model=is_reasoning_model,
+            ),
+            RelevanceEvaluator(
+                model_config,
+                threshold=relevance_threshold,
+                is_reasoning_model=is_reasoning_model,
+            ),
+            CoherenceEvaluator(
+                model_config,
+                threshold=coherence_threshold,
+                is_reasoning_model=is_reasoning_model,
+            ),
+            FluencyEvaluator(
+                model_config,
+                threshold=fluency_threshold,
+                is_reasoning_model=is_reasoning_model,
+            ),
+            SimilarityEvaluator(
+                model_config,
+                threshold=similarity_threshold,
+                is_reasoning_model=is_reasoning_model,
+            ),
             F1ScoreEvaluator(threshold=f1_score_threshold),
         ]
         super().__init__(evaluators=evaluators, **kwargs)

sdk/evaluation/azure-ai-evaluation/tests/unittests/test_qa_evaluator.py

@@ -0,0 +1,29 @@
+import pytest
+from azure.ai.evaluation import QAEvaluator
+
+
+@pytest.mark.usefixtures("mock_model_config")
+@pytest.mark.unittest
+class TestQAEvaluator:
+    def test_is_reasoning_model_passed_to_sub_evaluators(self, mock_model_config):
+        """Test that is_reasoning_model is passed to all LLM-based sub-evaluators"""
+        qa_evaluator = QAEvaluator(model_config=mock_model_config, is_reasoning_model=True)
+
+        # Verify that all LLM-based sub-evaluators have is_reasoning_model=True
+        for evaluator in qa_evaluator._evaluators:
+            # F1ScoreEvaluator doesn't use LLM, so it doesn't have _is_reasoning_model
+            if hasattr(evaluator, "_is_reasoning_model"):
+                assert (
+                    evaluator._is_reasoning_model is True
+                ), f"{type(evaluator).__name__} did not receive is_reasoning_model=True"
+
+    def test_is_reasoning_model_defaults_to_false(self, mock_model_config):
+        """Test that is_reasoning_model defaults to False for sub-evaluators"""
+        qa_evaluator = QAEvaluator(model_config=mock_model_config)
+
+        # Verify that all LLM-based sub-evaluators have is_reasoning_model=False
+        for evaluator in qa_evaluator._evaluators:
+            if hasattr(evaluator, "_is_reasoning_model"):
+                assert (
+                    evaluator._is_reasoning_model is False
+                ), f"{type(evaluator).__name__} did not default to is_reasoning_model=False"