ibm-watsonx-gov 1.3.3__cp313-cp313-macosx_11_0_arm64.whl

ibm_watsonx_gov/metrics/harm/harm_decorator.py

@@ -0,0 +1,60 @@
+# ----------------------------------------------------------------------------------------------------
+# IBM Confidential
+# Licensed Materials - Property of IBM
+# 5737-H76, 5900-A3Q
+# © Copyright IBM Corp. 2025  All Rights Reserved.
+# US Government Users Restricted Rights - Use, duplication or disclosure restricted by
+# GSA ADPSchedule Contract with IBM Corp.
+# ----------------------------------------------------------------------------------------------------
+
+from functools import partial
+from typing import Callable, Optional
+
+from wrapt import decorator
+
+from ibm_watsonx_gov.config.agentic_ai_configuration import \
+    AgenticAIConfiguration
+from ibm_watsonx_gov.entities.enums import EvaluatorFields
+from ibm_watsonx_gov.entities.metric import GenAIMetric
+from ibm_watsonx_gov.metrics.base_metric_decorator import BaseMetricDecorator
+from ibm_watsonx_gov.metrics.harm.harm_metric import HarmMetric
+
+
+class HarmDecorator(BaseMetricDecorator):
+
+    def evaluate_harm(self,
+                      func: Optional[Callable] = None,
+                      *,
+                      configuration: Optional[AgenticAIConfiguration] = None,
+                      metrics: list[GenAIMetric] = []
+                      ) -> dict:
+        """
+        An evaluation decorator for computing harm risk on an agentic node via granite guardian.
+        """
+        if func is None:
+            return partial(self.evaluate_harm, configuration=configuration, metrics=metrics)
+
+        if not metrics:
+            metrics = [HarmMetric()]
+
+        @decorator
+        def wrapper(func, instance, args, kwargs):
+
+            try:
+                self.validate(func=func, metrics=metrics,
+                              valid_metric_types=(HarmMetric,))
+
+                metric_inputs = [EvaluatorFields.INPUT_FIELDS]
+
+                original_result = self.compute_helper(func=func, args=args, kwargs=kwargs,
+                                                      configuration=configuration,
+                                                      metrics=metrics,
+                                                      metric_inputs=metric_inputs,
+                                                      metric_outputs=[])
+
+                return original_result
+            except Exception as ex:
+                raise Exception(
+                    f"There was an error while evaluating harm risk on {func.__name__},") from ex
+
+        return wrapper(func)

ibm_watsonx_gov/metrics/harm/harm_metric.py

@@ -0,0 +1,103 @@
+# ----------------------------------------------------------------------------------------------------
+# IBM Confidential
+# Licensed Materials - Property of IBM
+# 5737-H76, 5900-A3Q
+# © Copyright IBM Corp. 2025  All Rights Reserved.
+# US Government Users Restricted Rights - Use, duplication or disclosure restricted by
+# GSA ADPSchedule Contract with IBM Corp.
+# ----------------------------------------------------------------------------------------------------
+
+from typing import Annotated, Literal
+
+import pandas as pd
+from pydantic import Field
+
+from ibm_watsonx_gov.config.gen_ai_configuration import GenAIConfiguration
+from ibm_watsonx_gov.entities.enums import MetricGroup, TaskType
+from ibm_watsonx_gov.entities.evaluation_result import AggregateMetricResult
+from ibm_watsonx_gov.entities.metric import GenAIMetric
+from ibm_watsonx_gov.entities.metric_threshold import MetricThreshold
+from ibm_watsonx_gov.providers.detectors_provider import DetectorsProvider
+from ibm_watsonx_gov.utils.async_util import run_in_event_loop
+from ibm_watsonx_gov.utils.validation_util import validate_input
+
+HARM = "harm"
+
+
+class HarmMetric(GenAIMetric):
+    """
+    Defines the Harm metric class.
+
+    The Harm metric measures the risk of content considered universally harmful. This is general category, which should encompass a variety of risks.
+    It is computed using the granite guardian model.
+
+    Examples:
+        1. Create Harm metric with default parameters and compute using metrics evaluator.
+            .. code-block:: python
+
+                metric = HarmMetric()
+                result = MetricsEvaluator().evaluate(data={"input_text": "...", metrics=[metric])
+
+        2. Create Harm metric with a custom threshold.
+            .. code-block:: python
+
+                threshold  = MetricThreshold(type="lower_limit", value=0.5)
+                metric = HarmMetric(threshold=threshold)
+    """
+    name: Annotated[Literal["harm"],
+                    Field(title="Name",
+                          description="The harm metric name.",
+                          default=HARM, frozen=True)]
+    display_name: Annotated[Literal["Harm"],
+                            Field(title="Display Name",
+                                  description="The harm metric display name.",
+                                  default="Harm", frozen=True)]
+    method: Annotated[Literal["granite_guardian"],
+                      Field(title="Method",
+                            description="The method used to compute harm metric.",
+                            default="granite_guardian")]
+    tasks: Annotated[list[TaskType],
+                     Field(title="Tasks",
+                           description="The list of supported tasks.",
+                           default=TaskType.values(), frozen=True)]
+    thresholds: Annotated[list[MetricThreshold],
+                          Field(title="Thresholds",
+                                description="The metric thresholds.",
+                                default=[MetricThreshold(type="upper_limit", value=0.5)])]
+    group: Annotated[MetricGroup,
+                     Field(title="Group",
+                           description="The metric group.",
+                           default=MetricGroup.CONTENT_SAFETY, frozen=True)]
+
+    async def evaluate_async(
+        self,
+        data: pd.DataFrame,
+        configuration: GenAIConfiguration,
+        **kwargs
+    ) -> list[AggregateMetricResult]:
+
+        validate_input(data.columns.to_list(), configuration)
+        kwargs["detector_params"] = {"risk_name": HARM}
+        provider = DetectorsProvider(configuration=configuration,
+                                     metric_name=self.name,
+                                     metric_display_name=self.display_name,
+                                     metric_method=self.method,
+                                     metric_group=self.group,
+                                     thresholds=self.thresholds,
+                                     **kwargs)
+        aggregated_metric_result = await provider.evaluate_async(data=data)
+        return aggregated_metric_result
+
+    def evaluate(
+        self,
+        data: pd.DataFrame | dict,
+        configuration: GenAIConfiguration,
+        **kwargs,
+    ):
+        # If ran in sync mode, block until it is done
+        return run_in_event_loop(
+            self.evaluate_async,
+            data=data,
+            configuration=configuration,
+            **kwargs,
+        )

ibm_watsonx_gov/metrics/harm_engagement/__init__.py

@@ -0,0 +1,8 @@
+# ----------------------------------------------------------------------------------------------------
+# IBM Confidential
+# Licensed Materials - Property of IBM
+# 5737-H76, 5900-A3Q
+# © Copyright IBM Corp. 2025  All Rights Reserved.
+# US Government Users Restricted Rights - Use, duplication or disclosure restricted by
+# GSA ADPSchedule Contract with IBM Corp.
+# ----------------------------------------------------------------------------------------------------

ibm_watsonx_gov/metrics/harm_engagement/harm_engagement_decorator.py

@@ -0,0 +1,61 @@
+# ----------------------------------------------------------------------------------------------------
+# IBM Confidential
+# Licensed Materials - Property of IBM
+# 5737-H76, 5900-A3Q
+# © Copyright IBM Corp. 2025  All Rights Reserved.
+# US Government Users Restricted Rights - Use, duplication or disclosure restricted by
+# GSA ADPSchedule Contract with IBM Corp.
+# ----------------------------------------------------------------------------------------------------
+
+from functools import partial
+from typing import Callable, Optional
+
+from wrapt import decorator
+
+from ibm_watsonx_gov.config.agentic_ai_configuration import \
+    AgenticAIConfiguration
+from ibm_watsonx_gov.entities.enums import EvaluatorFields
+from ibm_watsonx_gov.entities.metric import GenAIMetric
+from ibm_watsonx_gov.metrics.base_metric_decorator import BaseMetricDecorator
+from ibm_watsonx_gov.metrics.harm_engagement.harm_engagement_metric import \
+    HarmEngagementMetric
+
+
+class HarmEngagementDecorator(BaseMetricDecorator):
+
+    def evaluate_harm_engagement(self,
+                                 func: Optional[Callable] = None,
+                                 *,
+                                 configuration: Optional[AgenticAIConfiguration] = None,
+                                 metrics: list[GenAIMetric] = []
+                                 ) -> dict:
+        """
+        An evaluation decorator for computing harm engagement on an agentic node via granite guardian.
+        """
+        if func is None:
+            return partial(self.evaluate_harm_engagement, configuration=configuration, metrics=metrics)
+
+        if not metrics:
+            metrics = [HarmEngagementMetric()]
+
+        @decorator
+        def wrapper(func, instance, args, kwargs):
+
+            try:
+                self.validate(func=func, metrics=metrics,
+                              valid_metric_types=(HarmEngagementMetric))
+
+                metric_inputs = [EvaluatorFields.INPUT_FIELDS]
+
+                original_result = self.compute_helper(func=func, args=args, kwargs=kwargs,
+                                                      configuration=configuration,
+                                                      metrics=metrics,
+                                                      metric_inputs=metric_inputs,
+                                                      metric_outputs=[])
+
+                return original_result
+            except Exception as ex:
+                raise Exception(
+                    f"There was an error while evaluating harm engagement on {func.__name__},") from ex
+
+        return wrapper(func)

ibm_watsonx_gov/metrics/harm_engagement/harm_engagement_metric.py

@@ -0,0 +1,103 @@
+# ----------------------------------------------------------------------------------------------------
+# IBM Confidential
+# Licensed Materials - Property of IBM
+# 5737-H76, 5900-A3Q
+# © Copyright IBM Corp. 2025  All Rights Reserved.
+# US Government Users Restricted Rights - Use, duplication or disclosure restricted by
+# GSA ADPSchedule Contract with IBM Corp.
+# ----------------------------------------------------------------------------------------------------
+
+from typing import Annotated, Literal
+
+import pandas as pd
+from pydantic import Field
+
+from ibm_watsonx_gov.config.gen_ai_configuration import GenAIConfiguration
+from ibm_watsonx_gov.entities.enums import MetricGroup, TaskType
+from ibm_watsonx_gov.entities.evaluation_result import AggregateMetricResult
+from ibm_watsonx_gov.entities.metric import GenAIMetric
+from ibm_watsonx_gov.entities.metric_threshold import MetricThreshold
+from ibm_watsonx_gov.providers.detectors_provider import DetectorsProvider
+from ibm_watsonx_gov.utils.async_util import run_in_event_loop
+from ibm_watsonx_gov.utils.validation_util import validate_input
+
+HARM_ENGAGEMENT = "harm_engagement"
+
+
+class HarmEngagementMetric(GenAIMetric):
+    """
+    Defines the Harm Engagement metric class.
+
+    The Harm Engagement metric measures the risk of an engagement or endorsement with any requests that are harmful or unethical.
+    It is computed using the granite guardian model.
+
+    Examples:
+        1. Create Harm Engagement metric with default parameters and compute using metrics evaluator.
+            .. code-block:: python
+
+                metric = HarmEngagementMetric()
+                result = MetricsEvaluator().evaluate(data={"input_text": "...", metrics=[metric])
+
+        2. Create Harm Engagement metric with a custom threshold.
+            .. code-block:: python
+
+                threshold  = MetricThreshold(type="lower_limit", value=0.5)
+                metric = HarmEngagementMetric(threshold=threshold)
+    """
+    name: Annotated[Literal["harm_engagement"],
+                    Field(title="Name",
+                          description="The harm engagement metric name.",
+                          default=HARM_ENGAGEMENT, frozen=True)]
+    display_name: Annotated[Literal["Harm Engagement"],
+                            Field(title="Display Name",
+                                  description="The harm engagement metric display name.",
+                                  default="Harm Engagement", frozen=True)]
+    method: Annotated[Literal["granite_guardian"],
+                      Field(title="Method",
+                            description="The method used to compute harm engagement metric.",
+                            default="granite_guardian")]
+    tasks: Annotated[list[TaskType],
+                     Field(title="Tasks",
+                           description="The list of supported tasks.",
+                           default=TaskType.values(), frozen=True)]
+    thresholds: Annotated[list[MetricThreshold],
+                          Field(title="Thresholds",
+                                description="The metric thresholds.",
+                                default=[MetricThreshold(type="upper_limit", value=0.5)])]
+    group: Annotated[MetricGroup,
+                     Field(title="Group",
+                           description="The metric group.",
+                           default=MetricGroup.CONTENT_SAFETY, frozen=True)]
+
+    async def evaluate_async(
+        self,
+        data: pd.DataFrame,
+        configuration: GenAIConfiguration,
+        **kwargs
+    ) -> list[AggregateMetricResult]:
+
+        validate_input(data.columns.to_list(), configuration)
+        kwargs["detector_params"] = {"risk_name": HARM_ENGAGEMENT}
+        provider = DetectorsProvider(configuration=configuration,
+                                     metric_name=self.name,
+                                     metric_display_name=self.display_name,
+                                     metric_method=self.method,
+                                     metric_group=self.group,
+                                     thresholds=self.thresholds,
+                                     **kwargs)
+        aggregated_metric_result = await provider.evaluate_async(data=data)
+        return aggregated_metric_result
+
+    def evaluate(
+        self,
+        data: pd.DataFrame | dict,
+        configuration: GenAIConfiguration,
+        **kwargs,
+    ):
+        # If ran in sync mode, block until it is done
+        return run_in_event_loop(
+            self.evaluate_async,
+            data=data,
+            configuration=configuration,
+            **kwargs,
+        )

ibm_watsonx_gov/metrics/hit_rate/hit_rate_decorator.py

@@ -0,0 +1,59 @@
+# ----------------------------------------------------------------------------------------------------
+# IBM Confidential
+# Licensed Materials - Property of IBM
+# 5737-H76, 5900-A3Q
+# © Copyright IBM Corp. 2025  All Rights Reserved.
+# US Government Users Restricted Rights - Use, duplication or disclosure restricted by
+# GSA ADPSchedule Contract with IBM Corp.
+# ----------------------------------------------------------------------------------------------------
+from functools import partial
+from typing import Callable, Optional
+
+from wrapt import decorator
+
+from ibm_watsonx_gov.config.agentic_ai_configuration import \
+    AgenticAIConfiguration
+from ibm_watsonx_gov.entities.enums import EvaluatorFields
+from ibm_watsonx_gov.entities.metric import GenAIMetric
+from ibm_watsonx_gov.metrics.base_metric_decorator import BaseMetricDecorator
+from ibm_watsonx_gov.metrics.hit_rate.hit_rate_metric import HitRateMetric
+
+
+class HitRateDecorator(BaseMetricDecorator):
+    def evaluate_hit_rate(self,
+                          func: Optional[Callable] = None,
+                          *,
+                          configuration: Optional[AgenticAIConfiguration] = None,
+                          metrics: list[GenAIMetric] = []
+                          ) -> dict:
+        """
+        An evaluation decorator for computing hit rate metric on an agentic node.
+        """
+        if func is None:
+            return partial(self.evaluate_hit_rate, configuration=configuration, metrics=metrics)
+
+        if not metrics:
+            metrics = [HitRateMetric()]
+
+        @decorator
+        def wrapper(func, instance, args, kwargs):
+
+            try:
+                self.validate(func=func, metrics=metrics,
+                              valid_metric_types=(HitRateMetric,))
+
+                metric_inputs = [EvaluatorFields.INPUT_FIELDS]
+                metric_outputs = [EvaluatorFields.CONTEXT_FIELDS]
+
+                original_result = self.compute_helper(func=func, args=args, kwargs=kwargs,
+                                                      configuration=configuration,
+                                                      metrics=metrics,
+                                                      metric_inputs=metric_inputs,
+                                                      metric_outputs=metric_outputs)
+
+                return original_result
+            except Exception as ex:
+                raise Exception(
+                    f"There was an error while evaluating hit rate metric on {func.__name__},") from ex
+
+        return wrapper(func)

ibm_watsonx_gov/metrics/hit_rate/hit_rate_metric.py

@@ -0,0 +1,167 @@
+# ----------------------------------------------------------------------------------------------------
+# IBM Confidential
+# Licensed Materials - Property of IBM
+# 5737-H76, 5900-A3Q
+# © Copyright IBM Corp. 2025  All Rights Reserved.
+# US Government Users Restricted Rights - Use, duplication or disclosure restricted by
+# GSA ADPSchedule Contract with IBM Corp.
+# ----------------------------------------------------------------------------------------------------
+
+
+from typing import Annotated, Any, Literal
+
+import pandas as pd
+from pydantic import Field, TypeAdapter, field_validator
+
+from ibm_watsonx_gov.config import AgenticAIConfiguration, GenAIConfiguration
+from ibm_watsonx_gov.entities.enums import MetricGroup, TaskType
+from ibm_watsonx_gov.entities.evaluation_result import (AggregateMetricResult,
+                                                        RecordMetricResult)
+from ibm_watsonx_gov.entities.metric import GenAIMetric
+from ibm_watsonx_gov.entities.metric_threshold import MetricThreshold
+from ibm_watsonx_gov.metrics.context_relevance.context_relevance_metric import (
+    CONTEXT_RELEVANCE, ContextRelevanceMetric, ContextRelevanceResult)
+
+HIT_RATE = "hit_rate"
+HIT_RATE_DISPLAY_NAME = "Hit Rate"
+
+
+class HitRateResult(RecordMetricResult):
+    name: str = HIT_RATE
+    display_name: str = HIT_RATE_DISPLAY_NAME
+    group: MetricGroup = MetricGroup.RETRIEVAL_QUALITY
+
+
+class HitRateMetric(GenAIMetric):
+    """
+    Defines the Hit Rate metric class.
+
+    The Hit Rate metric whether there is at least one relevant context among the retrieved contexts.
+    The Context Relevance metric is computed as a pre requisite to compute this metric.
+
+    Examples:
+        1. Create Hit Rate metric with default parameters and compute using metrics evaluator.
+            .. code-block:: python
+
+                metric = HitRateMetric()
+                result = MetricsEvaluator().evaluate(data={"input_text": "...", "context": "..."},
+                                                    metrics=[metric])
+                # A list of contexts can also be passed as shown below
+                result = MetricsEvaluator().evaluate(data={"input_text": "...", "context": ["...", "..."]},
+                                                    metrics=[metric])
+
+        2. Create Hit Rate metric with a custom threshold.
+            .. code-block:: python
+
+                threshold  = MetricThreshold(type="lower_limit", value=0.5)
+                metric = HitRateMetric(method=method, threshold=threshold)
+
+        3. Create Hit Rate metric with llm_as_judge method.
+            .. code-block:: python
+
+                # Define LLM Judge using watsonx.ai
+                # To use other frameworks and models as llm_judge, see :module:`ibm_watsonx_gov.entities.foundation_model`
+                llm_judge = LLMJudge(model=WxAIFoundationModel(
+                                            model_id="ibm/granite-3-3-8b-instruct",
+                                            project_id="<PROJECT_ID>"
+                                    ))
+                cr_metric = ContextRelevanceMetric(llm_judge=llm_judge)
+                ap_metric = HitRateMetric()
+                result = MetricsEvaluator().evaluate(data={"input_text": "...", "context": ["...", "..."]},
+                                                    metrics=[cr_metric, ap_metric])
+    """
+    name: Annotated[Literal["hit_rate"],
+                    Field(title="Name",
+                          description="The hit rate metric name.",
+                          default=HIT_RATE, frozen=True)]
+    display_name: Annotated[Literal["Hit Rate"],
+                            Field(title="Display Name",
+                                  description="The hit rate metric display name.",
+                                  default=HIT_RATE_DISPLAY_NAME, frozen=True)]
+    tasks: Annotated[list[TaskType],
+                     Field(title="Tasks",
+                           description="The list of supported tasks.",
+                           default=[TaskType.RAG])]
+    thresholds: Annotated[list[MetricThreshold],
+                          Field(title="Thresholds",
+                                description="The metric thresholds.",
+                                default=[MetricThreshold(type="lower_limit", value=0.7)])]
+    metric_dependencies: Annotated[list[GenAIMetric],
+                                   Field(title="Metric dependencies",
+                                         description="The list of metric dependencies",
+                                         default=[ContextRelevanceMetric()])]
+    group: Annotated[MetricGroup,
+                     Field(title="Group",
+                           description="The metric group.",
+                           default=MetricGroup.RETRIEVAL_QUALITY, frozen=True)]
+
+    @field_validator("metric_dependencies", mode="before")
+    @classmethod
+    def metric_dependencies_validator(cls, value: Any):
+        if value:
+            value = [TypeAdapter(Annotated[ContextRelevanceMetric, Field(
+                discriminator="name")]).validate_python(
+                m) for m in value]
+        return value
+
+    def evaluate(
+            self,
+            data: pd.DataFrame,
+            configuration: GenAIConfiguration | AgenticAIConfiguration,
+            metrics_result: list[AggregateMetricResult],
+            **kwargs,
+    ) -> AggregateMetricResult:
+        record_level_metrics = []
+        scores = []
+
+        context_relevance_result: list[ContextRelevanceResult] = next(
+            (metric_result.record_level_metrics for metric_result in metrics_result if metric_result.name == CONTEXT_RELEVANCE), None)
+
+        if context_relevance_result is None:
+            raise Exception(
+                f"Failed to evaluate {self.name} metric. Missing context relevance metric result")
+
+        for relevance_result in context_relevance_result:
+            score = self.__compute(
+                relevance_scores=relevance_result.additional_info.get(
+                    "contexts_values", []),
+                threshold=self.thresholds[0].value,
+            )
+            scores.append(score)
+            record_level_metrics.append(
+                HitRateResult(
+                    method="",
+                    provider="",
+                    record_id=relevance_result.record_id,
+                    value=score,
+                    thresholds=self.thresholds
+                )
+            )
+
+        mean = sum(scores) / len(scores)
+        aggregate_metric_score = AggregateMetricResult(
+            name=self.name,
+            display_name=self.display_name,
+            method="",
+            provider="",
+            group=self.group,
+            min=min(scores),
+            max=max(scores),
+            mean=mean,
+            value=mean,
+            total_records=len(record_level_metrics),
+            record_level_metrics=record_level_metrics,
+            thresholds=self.thresholds
+        )
+
+        return aggregate_metric_score
+
+    def __compute(self, relevance_scores: list[float], threshold: float) -> float:
+        hit_rate_value = 0
+
+        for score in relevance_scores:
+            if score >= threshold:
+                hit_rate_value = 1
+                break  # Stop once a relevance is found
+
+        return hit_rate_value

ibm_watsonx_gov/metrics/input_token_count/__init__.py

@@ -0,0 +1,8 @@
+# ----------------------------------------------------------------------------------------------------
+# IBM Confidential
+# Licensed Materials - Property of IBM
+# 5737-H76, 5900-A3Q
+# © Copyright IBM Corp. 2025  All Rights Reserved.
+# US Government Users Restricted Rights - Use, duplication or disclosure restricted by
+# GSA ADPSchedule Contract with IBM Corp.
+# ----------------------------------------------------------------------------------------------------

ibm_watsonx_gov/metrics/input_token_count/input_token_count_decorator.py

@@ -0,0 +1,58 @@
+# ----------------------------------------------------------------------------------------------------
+# IBM Confidential
+# Licensed Materials - Property of IBM
+# 5737-H76, 5900-A3Q
+# © Copyright IBM Corp. 2025  All Rights Reserved.
+# US Government Users Restricted Rights - Use, duplication or disclosure restricted by
+# GSA ADPSchedule Contract with IBM Corp.
+# ----------------------------------------------------------------------------------------------------
+from functools import partial
+from typing import Callable, Optional
+
+from wrapt import decorator
+
+from ibm_watsonx_gov.config.agentic_ai_configuration import \
+    AgenticAIConfiguration
+from ibm_watsonx_gov.entities.enums import EvaluatorFields
+from ibm_watsonx_gov.entities.metric import GenAIMetric
+from ibm_watsonx_gov.metrics.base_metric_decorator import BaseMetricDecorator
+from ibm_watsonx_gov.metrics.input_token_count.input_token_count_metric import \
+    InputTokenCountMetric
+
+
+class InputTokenCountDecorator(BaseMetricDecorator):
+    def evaluate_input_token_count(self,
+                                   func: Optional[Callable] = None,
+                                   *,
+                                   configuration: Optional[AgenticAIConfiguration] = None,
+                                   metrics: list[GenAIMetric] = []
+                                   ) -> dict:
+        """
+        An evaluation decorator for computing total input token.
+        """
+        if func is None:
+            return partial(self.evaluate_input_token_count, configuration=configuration, metrics=metrics)
+
+        if not metrics:
+            metrics = [InputTokenCountMetric()]
+
+        @decorator
+        def wrapper(func, instance, args, kwargs):
+
+            try:
+                self.validate(func=func, metrics=metrics,
+                              valid_metric_types=(InputTokenCountMetric,))
+
+                metric_inputs = [EvaluatorFields.INPUT_TOKEN_COUNT_FIELDS]
+
+                original_result = self.compute_helper(func=func, args=args, kwargs=kwargs,
+                                                      configuration=configuration,
+                                                      metrics=metrics,
+                                                      metric_inputs=metric_inputs)
+
+                return original_result
+            except Exception as ex:
+                raise Exception(
+                    f"There was an error while tracking total input token count on {func.__name__},") from ex
+
+        return wrapper(func)