azure-ai-evaluation 1.1.0__py3-none-any.whl → 1.3.0__py3-none-any.whl

azure/ai/evaluation/_evaluators/_common/_base_eval.py

@@ -4,15 +4,18 @@
 
 import inspect
 from abc import ABC, abstractmethod
-from typing import Any, Callable, Dict, Generic, List, TypedDict, TypeVar, Union, cast, final
+from typing import Any, Callable, Dict, Generic, List, TypedDict, TypeVar, Union, cast, final, Optional
 
 from promptflow._utils.async_utils import async_run_allowing_running_loop
 from typing_extensions import ParamSpec, TypeAlias, get_overloads
 
-from azure.ai.evaluation._common.math import list_mean
 from azure.ai.evaluation._exceptions import ErrorBlame, ErrorCategory, ErrorTarget, EvaluationException
 from azure.ai.evaluation._common.utils import remove_optional_singletons
+from azure.ai.evaluation._constants import _AggregationType
 from azure.ai.evaluation._model_configurations import Conversation
+from azure.ai.evaluation._common._experimental import experimental
+
+from ._conversation_aggregators import GetAggregator, GetAggregatorType
 
 P = ParamSpec("P")
 T = TypeVar("T")
@@ -25,6 +28,7 @@ class DerivedEvalInput(TypedDict, total=False):
     query: Dict[str, Any]
     response: Dict[str, Any]
     context: str
+    ground_truth: str
 
 
 AggregateResult: TypeAlias = Dict[str, Union[float, Dict[str, List[T]]]]
@@ -69,6 +73,13 @@ class EvaluatorBase(ABC, Generic[T_EvalValue]):
     :type not_singleton_inputs: List[str]
     :param eval_last_turn: If True, only the last turn of the conversation will be evaluated. Default is False.
     :type eval_last_turn: bool
+    :param conversation_aggregation_type: The type of aggregation to perform on the per-turn results of a conversation
+        to produce a single result.
+        Default is ~azure.ai.evaluation._AggregationType.MEAN.
+    :type conversation_aggregation_type: ~azure.ai.evaluation._AggregationType
+    :param conversation_aggregator_override: A function that will be used to aggregate per-turn results. If provided,
+        overrides the standard aggregator implied by conversation_aggregation_type. None by default.
+    :type conversation_aggregator_override: Optional[Callable[[List[float]], float]]
     """
 
     # ~~~ METHODS THAT ALMOST ALWAYS NEED TO BE OVERRIDDEN BY CHILDREN~~~
@@ -80,11 +91,17 @@ class EvaluatorBase(ABC, Generic[T_EvalValue]):
         *,
         not_singleton_inputs: List[str] = ["conversation", "kwargs"],
         eval_last_turn: bool = False,
+        conversation_aggregation_type: _AggregationType = _AggregationType.MEAN,
+        conversation_aggregator_override: Optional[Callable[[List[float]], float]] = None,
     ):
         self._not_singleton_inputs = not_singleton_inputs
         self._eval_last_turn = eval_last_turn
         self._singleton_inputs = self._derive_singleton_inputs()
         self._async_evaluator = AsyncEvaluatorBase(self._real_call)
+        self._conversation_aggregation_function = GetAggregator(conversation_aggregation_type)
+        if conversation_aggregator_override is not None:
+            # Type ignore since we already checked for None, but mypy doesn't know that.
+            self._conversation_aggregation_function = conversation_aggregator_override  # type: ignore[assignment]
 
     # This needs to be overridden just to change the function header into something more informative,
     # and to be able to add a more specific docstring. The actual function contents should just be
@@ -158,6 +175,7 @@ class EvaluatorBase(ABC, Generic[T_EvalValue]):
         include_context = "context" in self._singleton_inputs
         include_query = "query" in self._singleton_inputs
         include_response = "response" in self._singleton_inputs
+        include_ground_truth = "ground_truth" in self._singleton_inputs
 
         def converter(conversation: Dict) -> List[DerivedEvalInput]:
             messages = cast(List[Dict[str, Any]], conversation["messages"])
@@ -198,6 +216,8 @@ class EvaluatorBase(ABC, Generic[T_EvalValue]):
                     eval_input["response"] = response.get("content", "")
                 if include_context:
                     eval_input["context"] = str(context)
+                if include_ground_truth:
+                    eval_input["ground_truth"] = response.get("ground_truth", "")
                 eval_inputs.append(eval_input)
             return eval_inputs
 
@@ -355,7 +375,7 @@ class EvaluatorBase(ABC, Generic[T_EvalValue]):
         # Find and average all numeric values
         for metric, values in evaluation_per_turn.items():
             if all(isinstance(value, (int, float)) for value in values):
-                aggregated[metric] = list_mean(cast(List[Union[int, float]], values))
+                aggregated[metric] = self._conversation_aggregation_function(cast(List[Union[int, float]], values))
         # Slap the per-turn results back in.
         aggregated["evaluation_per_turn"] = evaluation_per_turn
         return aggregated
@@ -383,10 +403,51 @@ class EvaluatorBase(ABC, Generic[T_EvalValue]):
         # Otherwise, aggregate results.
         return self._aggregate_results(per_turn_results=per_turn_results)
 
+    # ~~~ METHODS THAT SHOULD NOT BE OVERRIDDEN BY CHILDREN~~~``
+
     @final
     def _to_async(self) -> "AsyncEvaluatorBase":
         return self._async_evaluator
 
+    @experimental
+    @final
+    def _set_conversation_aggregation_type(self, conversation_aggregation_type: _AggregationType) -> None:
+        """Input a conversation aggregation type to re-assign the aggregator function used by this evaluator for
+        multi-turn conversations. This aggregator is used to combine numeric outputs from each evaluation of a
+        multi-turn conversation into a single top-level result.
+
+        :param conversation_aggregation_type: The type of aggregation to perform on the per-turn
+            results of a conversation to produce a single result.
+        :type conversation_aggregation_type: ~azure.ai.evaluation._AggregationType
+        """
+        self._conversation_aggregation_function = GetAggregator(conversation_aggregation_type)
+
+    @experimental
+    @final
+    def _set_conversation_aggregator(self, aggregator: Callable[[List[float]], float]) -> None:
+        """Set the conversation aggregator function directly. This function will be applied to all numeric outputs
+        of an evaluator when it evaluates a conversation with multiple-turns thus ends up with multiple results per
+        evaluation that is needs to coalesce into a single result. Use when built-in aggregators do not
+        suit your needs, but use with caution.
+
+        :param aggregator: The function to use to aggregate per-turn results.
+        :type aggregator: Callable[[List[float]], float]
+        """
+        self._conversation_aggregation_function = aggregator
+
+    @experimental
+    @final
+    def _get_conversation_aggregator_type(self) -> _AggregationType:
+        """Get the current conversation aggregation type used by this evaluator. This refers to the
+        method used when a single input produces multiple evaluation results (ex: when a multi-turn conversation
+        is inputted into an evaluator that evaluates each turn individually). The individual inputs
+        are combined by the function implied here to produce a single overall result.
+
+        :return: The conversation aggregation type.
+        :rtype: ~azure.ai.evaluation._AggregationType
+        """
+        return GetAggregatorType(self._conversation_aggregation_function)
+
 
 class AsyncEvaluatorBase:
     """The asynchronous evaluator hidden underneath all evaluators. This makes generous use passing functions
@@ -402,7 +463,9 @@ class AsyncEvaluatorBase:
     # are just not passed into this function instead of ending up in kwargs.
     # Since we want this to be relatively call-agnostic, we just account for every input that any children
     # are known to throw at this, mash them into kwargs, and then pass them into the real call.
-    async def __call__(self, *, query=None, response=None, context=None, conversation=None, **kwargs):
+    async def __call__(
+        self, *, query=None, response=None, context=None, conversation=None, ground_truth=None, **kwargs
+    ):
         if conversation is not None:
             kwargs["conversation"] = conversation
         if query is not None:
@@ -411,4 +474,6 @@ class AsyncEvaluatorBase:
             kwargs["response"] = response
         if context is not None:
             kwargs["context"] = context
+        if ground_truth is not None:
+            kwargs["ground_truth"] = ground_truth
         return await self._real_call(**kwargs)

azure/ai/evaluation/_evaluators/_common/_base_multi_eval.py

@@ -0,0 +1,61 @@
+# ---------------------------------------------------------
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# ---------------------------------------------------------
+from concurrent.futures import as_completed
+from typing import TypeVar, Dict, List
+
+from promptflow.tracing import ThreadPoolExecutorWithContext as ThreadPoolExecutor
+from typing_extensions import override
+
+from azure.ai.evaluation._evaluators._common import EvaluatorBase
+
+T = TypeVar("T")
+
+
+class MultiEvaluatorBase(EvaluatorBase[T]):
+    """
+    Base class for evaluators that contain and run multiple other evaluators to produce a
+    suite of metrics.
+
+    Child classes still need to implement the __call__ methods, but they shouldn't need a _do_eval.
+
+    :param evaluators: The list of evaluators to run when this evaluator is called.
+    :type evaluators: List[~azure.ai.evaluation._evaluators._common.EvaluatorBase]
+    :param kwargs: Additional arguments to pass to the evaluator.
+    :type kwargs: Any
+    :return: An evaluator that runs multiple other evaluators and combines their results.
+    """
+
+    def __init__(self, evaluators: List[EvaluatorBase[T]], **kwargs):
+        super().__init__()
+        self._parallel = kwargs.pop("_parallel", True)
+        self._evaluators = evaluators
+
+    @override
+    async def _do_eval(self, eval_input: Dict) -> Dict[str, T]:
+        """Run each evaluator, possibly in parallel, and combine the results into
+        a single large dictionary containing each evaluation. Inputs are passed
+        directly to each evaluator without additional processing.
+
+
+        :param eval_input: The input to the evaluation function.
+        :type eval_input: Dict
+        :return: The evaluation result.
+        :rtype: Dict
+        """
+        results: Dict[str, T] = {}
+        if self._parallel:
+            with ThreadPoolExecutor() as executor:
+                # pylint: disable=no-value-for-parameter
+                futures = {executor.submit(evaluator, **eval_input): evaluator for evaluator in self._evaluators}
+
+                for future in as_completed(futures):
+                    results.update(future.result())
+        else:
+            for evaluator in self._evaluators:
+                result = evaluator(**eval_input)
+                # Ignore is to avoid mypy getting upset over the amount of duck-typing
+                # that's going on to shove evaluators around like this.
+                results.update(result)  # type: ignore[arg-type]
+
+        return results

azure/ai/evaluation/_evaluators/_common/_base_rai_svc_eval.py

@@ -15,6 +15,7 @@ from azure.ai.evaluation._common.rai_service import evaluate_with_rai_service, e
 from azure.ai.evaluation._common.utils import validate_azure_ai_project
 from azure.ai.evaluation._exceptions import EvaluationException
 from azure.ai.evaluation._common.utils import validate_conversation
+from azure.ai.evaluation._constants import _AggregationType
 from azure.core.credentials import TokenCredential
 
 from . import EvaluatorBase
@@ -35,6 +36,10 @@ class RaiServiceEvaluatorBase(EvaluatorBase[T]):
         aggregated. Per-turn results are still be available in the output via the "evaluation_per_turn" key
         when this occurs. Default is False, resulting full conversation evaluation and aggregation.
     :type eval_last_turn: bool
+    :param conversation_aggregation_type: The type of aggregation to perform on the per-turn results of a conversation
+        to produce a single result.
+        Default is ~azure.ai.evaluation._AggregationType.MEAN.
+    :type conversation_aggregation_type: ~azure.ai.evaluation._AggregationType
     """
 
     @override
@@ -44,8 +49,9 @@ class RaiServiceEvaluatorBase(EvaluatorBase[T]):
         azure_ai_project: dict,
         credential: TokenCredential,
         eval_last_turn: bool = False,
+        conversation_aggregation_type: _AggregationType = _AggregationType.MEAN,
     ):
-        super().__init__(eval_last_turn=eval_last_turn)
+        super().__init__(eval_last_turn=eval_last_turn, conversation_aggregation_type=conversation_aggregation_type)
         self._eval_metric = eval_metric
         self._azure_ai_project = validate_azure_ai_project(azure_ai_project)
         self._credential = credential

azure/ai/evaluation/_evaluators/_common/_conversation_aggregators.py

@@ -0,0 +1,49 @@
+# ---------------------------------------------------------
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# ---------------------------------------------------------
+
+from typing import Callable, List
+from azure.ai.evaluation._common.math import list_mean
+from azure.ai.evaluation._exceptions import ErrorBlame, ErrorCategory, ErrorTarget, EvaluationException
+from azure.ai.evaluation._constants import _AggregationType
+
+
+def GetAggregator(aggregation_type: _AggregationType) -> Callable[[List[float]], float]:
+    if aggregation_type == _AggregationType.SUM:
+        return sum
+    if aggregation_type == _AggregationType.MEAN:
+        return list_mean
+    if aggregation_type == _AggregationType.MAX:
+        return max
+    if aggregation_type == _AggregationType.MIN:
+        return min
+    if aggregation_type == _AggregationType.CUSTOM:
+        msg = (
+            "Cannot 'get' aggregator function associated with custom aggregation enum."
+            + " This enum value should only be outputted as an indicator of an injected"
+            + " aggregation function, not inputted directly"
+        )
+        raise EvaluationException(
+            message=msg,
+            blame=ErrorBlame.UNKNOWN,
+            category=ErrorCategory.INVALID_VALUE,
+            target=ErrorTarget.EVALUATE,
+        )
+    raise EvaluationException(
+        message=f"Unaccounted for aggregation type: {aggregation_type}",
+        blame=ErrorBlame.UNKNOWN,
+        category=ErrorCategory.INVALID_VALUE,
+        target=ErrorTarget.EVALUATE,
+    )
+
+
+def GetAggregatorType(aggregation_function: Callable) -> _AggregationType:
+    if aggregation_function == sum:  # pylint: disable=comparison-with-callable
+        return _AggregationType.SUM
+    if aggregation_function == list_mean:  # pylint: disable=comparison-with-callable
+        return _AggregationType.MEAN
+    if aggregation_function == max:  # pylint: disable=comparison-with-callable
+        return _AggregationType.MAX
+    if aggregation_function == min:  # pylint: disable=comparison-with-callable
+        return _AggregationType.MIN
+    return _AggregationType.CUSTOM

azure/ai/evaluation/_evaluators/_content_safety/_content_safety.py

@@ -1,13 +1,11 @@
 # ---------------------------------------------------------
 # Copyright (c) Microsoft Corporation. All rights reserved.
 # ---------------------------------------------------------
-from concurrent.futures import as_completed
-from typing import Callable, Dict, List, Union
+from typing import Dict, List, Union
 
-from promptflow.tracing import ThreadPoolExecutorWithContext as ThreadPoolExecutor
 from typing_extensions import overload, override
 
-from azure.ai.evaluation._evaluators._common import EvaluatorBase
+from azure.ai.evaluation._evaluators._common import MultiEvaluatorBase
 from azure.ai.evaluation._model_configurations import Conversation
 from azure.ai.evaluation._common._experimental import experimental
 
@@ -18,7 +16,7 @@ from ._violence import ViolenceEvaluator
 
 
 @experimental
-class ContentSafetyEvaluator(EvaluatorBase[Union[str, float]]):
+class ContentSafetyEvaluator(MultiEvaluatorBase[Union[str, float]]):
     """
     Initialize a content safety evaluator configured to evaluate content safety metrics for QA scenario.
 
@@ -44,16 +42,14 @@ class ContentSafetyEvaluator(EvaluatorBase[Union[str, float]]):
     id = "content_safety"
     """Evaluator identifier, experimental and to be used only with evaluation in cloud."""
 
-    # TODO address 3579092 to re-enabled parallel evals.
     def __init__(self, credential, azure_ai_project, **kwargs):
-        super().__init__()
-        self._parallel = kwargs.pop("_parallel", True)
-        self._evaluators: List[Callable[..., Dict[str, Union[str, float]]]] = [
+        evaluators = [
             ViolenceEvaluator(credential, azure_ai_project),
             SexualEvaluator(credential, azure_ai_project),
             SelfHarmEvaluator(credential, azure_ai_project),
             HateUnfairnessEvaluator(credential, azure_ai_project),
         ]
+        super().__init__(evaluators=evaluators, **kwargs)
 
     @overload
     def __call__(
@@ -109,36 +105,3 @@ class ContentSafetyEvaluator(EvaluatorBase[Union[str, float]]):
         :rtype: Union[Dict[str, Union[str, float]], Dict[str, Union[float, Dict[str, List[Union[str, float]]]]]]
         """
         return super().__call__(*args, **kwargs)
-
-    @override
-    async def _do_eval(self, eval_input: Dict) -> Dict[str, Union[str, float]]:
-        """Perform the evaluation using the Azure AI RAI service.
-        The exact evaluation performed is determined by the evaluation metric supplied
-        by the child class initializer.
-
-        :param eval_input: The input to the evaluation function.
-        :type eval_input: Dict
-        :return: The evaluation result.
-        :rtype: Dict
-        """
-        query = eval_input.get("query", None)
-        response = eval_input.get("response", None)
-        conversation = eval_input.get("conversation", None)
-        results: Dict[str, Union[str, float]] = {}
-        # TODO fix this to not explode on empty optional inputs (PF SKD error)
-        if self._parallel:
-            with ThreadPoolExecutor() as executor:
-                # pylint: disable=no-value-for-parameter
-                futures = {
-                    executor.submit(evaluator, query=query, response=response, conversation=conversation): evaluator
-                    for evaluator in self._evaluators
-                }
-
-                for future in as_completed(futures):
-                    results.update(future.result())
-        else:
-            for evaluator in self._evaluators:
-                result = evaluator(query=query, response=response, conversation=conversation)
-                results.update(result)
-
-        return results

azure/ai/evaluation/_evaluators/_content_safety/_hate_unfairness.py

@@ -9,6 +9,7 @@ from azure.ai.evaluation._common._experimental import experimental
 from azure.ai.evaluation._common.constants import EvaluationMetrics
 from azure.ai.evaluation._evaluators._common import RaiServiceEvaluatorBase
 from azure.ai.evaluation._model_configurations import Conversation
+from azure.ai.evaluation._constants import _AggregationType
 
 
 @experimental
@@ -71,6 +72,7 @@ class HateUnfairnessEvaluator(RaiServiceEvaluatorBase[Union[str, float]]):
             eval_metric=EvaluationMetrics.HATE_FAIRNESS,
             azure_ai_project=azure_ai_project,
             credential=credential,
+            conversation_aggregation_type=_AggregationType.MAX,
         )
 
     @overload

azure/ai/evaluation/_evaluators/_content_safety/_self_harm.py

@@ -9,6 +9,7 @@ from azure.ai.evaluation._common._experimental import experimental
 from azure.ai.evaluation._common.constants import EvaluationMetrics
 from azure.ai.evaluation._evaluators._common import RaiServiceEvaluatorBase
 from azure.ai.evaluation._model_configurations import Conversation
+from azure.ai.evaluation._constants import _AggregationType
 
 
 @experimental
@@ -65,6 +66,7 @@ class SelfHarmEvaluator(RaiServiceEvaluatorBase[Union[str, float]]):
             eval_metric=EvaluationMetrics.SELF_HARM,
             azure_ai_project=azure_ai_project,
             credential=credential,
+            conversation_aggregation_type=_AggregationType.MAX,
         )
 
     @overload

azure/ai/evaluation/_evaluators/_content_safety/_sexual.py

@@ -9,6 +9,7 @@ from azure.ai.evaluation._common._experimental import experimental
 from azure.ai.evaluation._common.constants import EvaluationMetrics
 from azure.ai.evaluation._evaluators._common import RaiServiceEvaluatorBase
 from azure.ai.evaluation._model_configurations import Conversation
+from azure.ai.evaluation._constants import _AggregationType
 
 
 @experimental
@@ -67,6 +68,7 @@ class SexualEvaluator(RaiServiceEvaluatorBase[Union[str, float]]):
             eval_metric=EvaluationMetrics.SEXUAL,
             azure_ai_project=azure_ai_project,
             credential=credential,
+            conversation_aggregation_type=_AggregationType.MAX,
         )
 
     @overload

azure/ai/evaluation/_evaluators/_content_safety/_violence.py

@@ -9,6 +9,7 @@ from azure.ai.evaluation._common._experimental import experimental
 from azure.ai.evaluation._common.constants import EvaluationMetrics
 from azure.ai.evaluation._evaluators._common import RaiServiceEvaluatorBase
 from azure.ai.evaluation._model_configurations import Conversation
+from azure.ai.evaluation._constants import _AggregationType
 
 
 @experimental
@@ -67,6 +68,7 @@ class ViolenceEvaluator(RaiServiceEvaluatorBase[Union[str, float]]):
             eval_metric=EvaluationMetrics.VIOLENCE,
             azure_ai_project=azure_ai_project,
             credential=credential,
+            conversation_aggregation_type=_AggregationType.MAX,
         )
 
     @overload

azure/ai/evaluation/_evaluators/_f1_score/_f1_score.py

@@ -3,45 +3,44 @@
 # ---------------------------------------------------------
 
 from collections import Counter
-from typing import List
+from typing import List, Dict
+from typing_extensions import overload, override
 
-from promptflow._utils.async_utils import async_run_allowing_running_loop
+from azure.ai.evaluation._evaluators._common import EvaluatorBase
 
-from azure.ai.evaluation._exceptions import ErrorBlame, ErrorCategory, ErrorTarget, EvaluationException
 
+class F1ScoreEvaluator(EvaluatorBase):
+    """
+    Calculates the F1 score for a given response and ground truth or a multi-turn conversation.
 
-class _AsyncF1ScoreEvaluator:
-    def __init__(self):
-        pass
+    F1 Scores range from 0 to 1, with 1 being the best possible score.
 
-    async def __call__(self, *, response: str, ground_truth: str, **kwargs):
-        """
-        Evaluate F1 score.
+    The F1-score computes the ratio of the number of shared words between the model generation and
+    the ground truth. Ratio is computed over the individual words in the generated response against those in the ground
+    truth answer. The number of shared words between the generation and the truth is the basis of the F1 score:
+    precision is the ratio of the number of shared words to the total number of words in the generation, and recall
+    is the ratio of the number of shared words to the total number of words in the ground truth.
 
-        :keyword response: The response to be evaluated.
-        :paramtype response: str
-        :keyword ground_truth: The ground truth to be evaluated.
-        :paramtype ground_truth: str
-        :return: The F1 score.
-        :rtype: Dict[str, float]
-        """
-        # Validate inputs
-        if not (response and response.strip() and response != "None") or not (
-            ground_truth and ground_truth.strip() and ground_truth != "None"
-        ):
-            msg = "Both 'response' and 'ground_truth' must be non-empty strings."
-            raise EvaluationException(
-                message=msg,
-                internal_message=msg,
-                error_category=ErrorCategory.MISSING_FIELD,
-                error_blame=ErrorBlame.USER_ERROR,
-                error_target=ErrorTarget.F1_EVALUATOR,
-            )
+    Use the F1 score when you want a single comprehensive metric that combines both recall and precision in your
+    model's responses. It provides a balanced evaluation of your model's performance in terms of capturing accurate
+    information in the response.
 
-        # Run f1 score computation.
-        f1_result = self._compute_f1_score(response=response, ground_truth=ground_truth)
 
-        return {"f1_score": f1_result}
+    .. admonition:: Example:
+
+        .. literalinclude:: ../samples/evaluation_samples_evaluate.py
+            :start-after: [START f1_score_evaluator]
+            :end-before: [END f1_score_evaluator]
+            :language: python
+            :dedent: 8
+            :caption: Initialize and call an F1ScoreEvaluator.
+    """
+
+    id = "azureml://registries/azureml/models/F1Score-Evaluator/versions/3"
+    """Evaluator identifier, experimental and to be used only with evaluation in cloud."""
+
+    def __init__(self):
+        super().__init__()
 
     @classmethod
     def _compute_f1_score(cls, response: str, ground_truth: str) -> float:
@@ -103,41 +102,24 @@ class _AsyncF1ScoreEvaluator:
 
         return f1
 
+    @override
+    async def _do_eval(self, eval_input: Dict) -> Dict[str, float]:
+        """Produce an f1 score evaluation result.
 
-class F1ScoreEvaluator:
-    """
-    Calculates the F1 score for a given response and ground truth or a multi-turn conversation.
-
-    F1 Scores range from 0 to 1, with 1 being the best possible score.
-
-    The F1-score computes the ratio of the number of shared words between the model generation and
-    the ground truth. Ratio is computed over the individual words in the generated response against those in the ground
-    truth answer. The number of shared words between the generation and the truth is the basis of the F1 score:
-    precision is the ratio of the number of shared words to the total number of words in the generation, and recall
-    is the ratio of the number of shared words to the total number of words in the ground truth.
-
-    Use the F1 score when you want a single comprehensive metric that combines both recall and precision in your
-    model's responses. It provides a balanced evaluation of your model's performance in terms of capturing accurate
-    information in the response.
-
-
-    .. admonition:: Example:
-
-        .. literalinclude:: ../samples/evaluation_samples_evaluate.py
-            :start-after: [START f1_score_evaluator]
-            :end-before: [END f1_score_evaluator]
-            :language: python
-            :dedent: 8
-            :caption: Initialize and call an F1ScoreEvaluator.
-    """
-
-    id = "azureml://registries/azureml/models/F1Score-Evaluator/versions/3"
-    """Evaluator identifier, experimental and to be used only with evaluation in cloud."""
+        :param eval_input: The input to the evaluation function.
+        :type eval_input: Dict
+        :return: The evaluation result.
+        :rtype: Dict
+        """
+        ground_truth = eval_input["ground_truth"]
+        response = eval_input["response"]
+        # Run f1 score computation.
+        f1_result = self._compute_f1_score(response=response, ground_truth=ground_truth)
 
-    def __init__(self):
-        self._async_evaluator = _AsyncF1ScoreEvaluator()
+        return {"f1_score": f1_result}
 
-    def __call__(self, *, response: str, ground_truth: str, **kwargs):
+    @overload  # type: ignore
+    def __call__(self, *, response: str, ground_truth: str) -> Dict[str, float]:
         """
         Evaluate F1 score.
 
@@ -149,9 +131,20 @@ class F1ScoreEvaluator:
         :rtype: Dict[str, float]
         """
 
-        return async_run_allowing_running_loop(
-            self._async_evaluator, response=response, ground_truth=ground_truth, **kwargs
-        )
+    @override
+    def __call__(  # pylint: disable=docstring-missing-param
+        self,
+        *args,
+        **kwargs,
+    ):
+        """
+        Evaluate F1 score.
 
-    def _to_async(self):
-        return self._async_evaluator
+        :keyword response: The response to be evaluated.
+        :paramtype response: str
+        :keyword ground_truth: The ground truth to be evaluated.
+        :paramtype ground_truth: str
+        :return: The F1 score.
+        :rtype: Dict[str, float]
+        """
+        return super().__call__(*args, **kwargs)