feat: Add HallucinationsV1 evaluation metric

PiperOrigin-RevId: 813456369

src/google/adk/evaluation/app_details.py

@@ -39,7 +39,7 @@ class AgentDetails(EvalBaseModel):
 class AppDetails(EvalBaseModel):
   """Contains details about the App (the agentic system).
 
-  This structure is only a projection of the acutal app. Only details
+  This structure is only a projection of the actual app. Only details
   that are relevant to the Eval System are captured here.
   """
 

src/google/adk/evaluation/eval_metrics.py

@@ -52,6 +52,8 @@ class PrebuiltMetrics(Enum):
       "rubric_based_final_response_quality_v1"
   )
 
+  HALLUCINATIONS_V1 = "hallucinations_v1"
+
   RUBRIC_BASED_TOOL_USE_QUALITY_V1 = "rubric_based_tool_use_quality_v1"
 
 
@@ -130,6 +132,24 @@ class RubricsBasedCriterion(BaseCriterion):
   )
 
 
+class HallucinationsCriterion(BaseCriterion):
+  """Criterion to use when evaluating agents response for hallucinations."""
+
+  judge_model_options: JudgeModelOptions = Field(
+      default_factory=JudgeModelOptions,
+      description="Options for the judge model.",
+  )
+
+  evaluate_intermediate_nl_responses: bool = Field(
+      default=False,
+      description=(
+          "Whether any intermediate NL responses should be evaluated"
+          " for hallucinations or not. By default, the metric only evaluates"
+          " final response from the Agent for hallucinations."
+      ),
+  )
+
+
 class EvalMetric(EvalBaseModel):
   """A metric used to evaluate a particular aspect of an eval case."""
 

src/google/adk/evaluation/metric_evaluator_registry.py

@@ -23,6 +23,7 @@ from .eval_metrics import MetricInfo
 from .eval_metrics import PrebuiltMetrics
 from .evaluator import Evaluator
 from .final_response_match_v2 import FinalResponseMatchV2Evaluator
+from .hallucinations_v1 import HallucinationsV1Evaluator
 from .response_evaluator import ResponseEvaluator
 from .rubric_based_final_response_quality_v1 import RubricBasedFinalResponseQualityV1Evaluator
 from .rubric_based_tool_use_quality_v1 import RubricBasedToolUseV1Evaluator
@@ -117,6 +118,10 @@ def _get_default_metric_evaluator_registry() -> MetricEvaluatorRegistry:
       metric_info=RubricBasedFinalResponseQualityV1Evaluator.get_metric_info(),
       evaluator=RubricBasedFinalResponseQualityV1Evaluator,
   )
+  metric_evaluator_registry.register_evaluator(
+      metric_info=HallucinationsV1Evaluator.get_metric_info(),
+      evaluator=HallucinationsV1Evaluator,
+  )
   metric_evaluator_registry.register_evaluator(
       metric_info=RubricBasedToolUseV1Evaluator.get_metric_info(),
       evaluator=RubricBasedToolUseV1Evaluator,