deepeval 3.7.5__py3-none-any.whl → 3.7.6__py3-none-any.whl

deepeval/metrics/multimodal_metrics/image_helpfulness/image_helpfulness.py

@@ -1,16 +1,17 @@
 import asyncio
 from typing import Optional, List, Tuple, Union
 
-from deepeval.metrics import BaseMultimodalMetric
+from deepeval.metrics import BaseMetric
 from deepeval.test_case import LLMTestCaseParams, LLMTestCase, MLLMImage
 from deepeval.metrics.multimodal_metrics.image_helpfulness.template import (
     ImageHelpfulnessTemplate,
 )
 from deepeval.metrics.utils import (
     construct_verbose_logs,
-    trimAndLoadJson,
-    check_mllm_test_case_params,
+    check_llm_test_case_params,
     initialize_model,
+    a_generate_with_schema_and_extract,
+    generate_with_schema_and_extract,
 )
 from deepeval.models import DeepEvalBaseLLM
 from deepeval.metrics.multimodal_metrics.image_helpfulness.schema import (
@@ -23,7 +24,7 @@ from deepeval.utils import (
 )
 
 
-class ImageHelpfulnessMetric(BaseMultimodalMetric):
+class ImageHelpfulnessMetric(BaseMetric):
 
     _required_params: List[LLMTestCaseParams] = [
         LLMTestCaseParams.INPUT,
@@ -54,8 +55,14 @@ class ImageHelpfulnessMetric(BaseMultimodalMetric):
         _in_component: bool = False,
         _log_metric_to_confident: bool = True,
     ) -> float:
-        check_mllm_test_case_params(
-            test_case, self._required_params, None, None, self, self.model
+        check_llm_test_case_params(
+            test_case,
+            self._required_params,
+            None,
+            None,
+            self,
+            self.model,
+            test_case.multimodal,
         )
         self.evaluation_cost = 0 if self.using_native_model else None
         with metric_progress_indicator(
@@ -156,8 +163,14 @@ class ImageHelpfulnessMetric(BaseMultimodalMetric):
         _in_component: bool = False,
         _log_metric_to_confident: bool = True,
     ) -> float:
-        check_mllm_test_case_params(
-            test_case, self._required_params, None, None, self, self.model
+        check_llm_test_case_params(
+            test_case,
+            self._required_params,
+            None,
+            None,
+            self,
+            self.model,
+            test_case.multimodal,
         )
         self.evaluation_cost = 0 if self.using_native_model else None
         with metric_progress_indicator(
@@ -262,20 +275,13 @@ class ImageHelpfulnessMetric(BaseMultimodalMetric):
             context_above, context_below
         )
         prompt = f"{instructions} \nImages: {image}"
-        if self.using_native_model:
-            res, cost = self.model.generate(prompt, schema=ReasonScore)
-            self.evaluation_cost += cost
-            return res.score, res.reasoning
-        else:
-            try:
-                res: ReasonScore = self.model.generate(
-                    prompt, schema=ReasonScore
-                )
-                return res.score, res.reasoning
-            except TypeError:
-                res = self.model.generate(prompt)
-                data = trimAndLoadJson(res, self)
-                return data["score"], data["reasoning"]
+        return generate_with_schema_and_extract(
+            metric=self,
+            prompt=prompt,
+            schema_cls=ReasonScore,
+            extract_schema=lambda s: (s.score, s.reasoning),
+            extract_json=lambda data: (data["score"], data["reasoning"]),
+        )
 
     async def a_evaluate_image_helpfulness(
         self,
@@ -287,20 +293,13 @@ class ImageHelpfulnessMetric(BaseMultimodalMetric):
             context_above, context_below
         )
         prompt = f"{instructions} \nImages: {image}"
-        if self.using_native_model:
-            res, cost = await self.model.a_generate(prompt, schema=ReasonScore)
-            self.evaluation_cost += cost
-            return res.score, res.reasoning
-        else:
-            try:
-                res: ReasonScore = await self.model.a_generate(
-                    prompt, schema=ReasonScore
-                )
-                return res.score, res.reasoning
-            except TypeError:
-                res = await self.model.a_generate(prompt)
-                data = trimAndLoadJson(res, self)
-                return data["score"], data["reasoning"]
+        return await a_generate_with_schema_and_extract(
+            metric=self,
+            prompt=prompt,
+            schema_cls=ReasonScore,
+            extract_schema=lambda s: (s.score, s.reasoning),
+            extract_json=lambda data: (data["score"], data["reasoning"]),
+        )
 
     def get_image_context(
         self, image_index: int, actual_output: List[Union[str, MLLMImage]]
@@ -335,7 +334,7 @@ class ImageHelpfulnessMetric(BaseMultimodalMetric):
             if isinstance(element, MLLMImage)
         ]
 
-    def calculate_score(self, scores: List[float]):
+    def calculate_score(self, scores: List[float]) -> float:
         return sum(scores) / len(scores)
 
     def is_successful(self) -> bool:
@@ -344,7 +343,7 @@ class ImageHelpfulnessMetric(BaseMultimodalMetric):
         else:
             try:
                 self.success = self.score >= self.threshold
-            except:
+            except TypeError:
                 self.success = False
         return self.success
 

deepeval/metrics/multimodal_metrics/image_reference/image_reference.py

@@ -1,16 +1,17 @@
 import asyncio
 from typing import Optional, List, Tuple, Union
 
-from deepeval.metrics import BaseMultimodalMetric
+from deepeval.metrics import BaseMetric
 from deepeval.test_case import LLMTestCaseParams, LLMTestCase, MLLMImage
 from deepeval.metrics.multimodal_metrics.image_reference.template import (
     ImageReferenceTemplate,
 )
 from deepeval.metrics.utils import (
     construct_verbose_logs,
-    trimAndLoadJson,
-    check_mllm_test_case_params,
+    check_llm_test_case_params,
     initialize_model,
+    a_generate_with_schema_and_extract,
+    generate_with_schema_and_extract,
 )
 from deepeval.models import DeepEvalBaseLLM
 from deepeval.metrics.multimodal_metrics.image_reference.schema import (
@@ -23,7 +24,7 @@ from deepeval.utils import (
 )
 
 
-class ImageReferenceMetric(BaseMultimodalMetric):
+class ImageReferenceMetric(BaseMetric):
 
     _required_params: List[LLMTestCaseParams] = [
         LLMTestCaseParams.INPUT,
@@ -54,8 +55,14 @@ class ImageReferenceMetric(BaseMultimodalMetric):
         _in_component: bool = False,
         _log_metric_to_confident: bool = True,
     ) -> float:
-        check_mllm_test_case_params(
-            test_case, self._required_params, None, None, self, self.model
+        check_llm_test_case_params(
+            test_case,
+            self._required_params,
+            None,
+            None,
+            self,
+            self.model,
+            test_case.multimodal,
         )
         self.evaluation_cost = 0 if self.using_native_model else None
         with metric_progress_indicator(
@@ -156,8 +163,14 @@ class ImageReferenceMetric(BaseMultimodalMetric):
         _in_component: bool = False,
         _log_metric_to_confident: bool = True,
     ) -> float:
-        check_mllm_test_case_params(
-            test_case, self._required_params, None, None, self, self.model
+        check_llm_test_case_params(
+            test_case,
+            self._required_params,
+            None,
+            None,
+            self,
+            self.model,
+            test_case.multimodal,
         )
         self.evaluation_cost = 0 if self.using_native_model else None
         with metric_progress_indicator(
@@ -262,20 +275,13 @@ class ImageReferenceMetric(BaseMultimodalMetric):
             context_above, context_below
         )
         prompt = f"{instructions} \nImages: {image}"
-        if self.using_native_model:
-            res, cost = self.model.generate(prompt, schema=ReasonScore)
-            self.evaluation_cost += cost
-            return res.score, res.reasoning
-        else:
-            try:
-                res: ReasonScore = self.model.generate(
-                    prompt, schema=ReasonScore
-                )
-                return res.score, res.reasoning
-            except TypeError:
-                res = self.model.generate(prompt)
-                data = trimAndLoadJson(res, self)
-                return data["score"], data["reasoning"]
+        return generate_with_schema_and_extract(
+            metric=self,
+            prompt=prompt,
+            schema_cls=ReasonScore,
+            extract_schema=lambda s: (s.score, s.reasoning),
+            extract_json=lambda data: (data["score"], data["reasoning"]),
+        )
 
     async def a_evaluate_image_reference(
         self,
@@ -287,20 +293,13 @@ class ImageReferenceMetric(BaseMultimodalMetric):
             context_above, context_below
         )
         prompt = f"{instructions} \nImages: {image}"
-        if self.using_native_model:
-            res, cost = await self.model.a_generate(prompt, schema=ReasonScore)
-            self.evaluation_cost += cost
-            return res.score, res.reasoning
-        else:
-            try:
-                res: ReasonScore = await self.model.a_generate(
-                    prompt, schema=ReasonScore
-                )
-                return res.score, res.reasoning
-            except TypeError:
-                res = await self.model.a_generate(prompt)
-                data = trimAndLoadJson(res, self)
-                return data["score"], data["reasoning"]
+        return await a_generate_with_schema_and_extract(
+            metric=self,
+            prompt=prompt,
+            schema_cls=ReasonScore,
+            extract_schema=lambda s: (s.score, s.reasoning),
+            extract_json=lambda data: (data["score"], data["reasoning"]),
+        )
 
     def get_image_context(
         self, image_index: int, actual_output: List[Union[str, MLLMImage]]
@@ -335,7 +334,7 @@ class ImageReferenceMetric(BaseMultimodalMetric):
             if isinstance(element, MLLMImage)
         ]
 
-    def calculate_score(self, scores: List[float]):
+    def calculate_score(self, scores: List[float]) -> float:
         return sum(scores) / len(scores)
 
     def is_successful(self) -> bool:
@@ -344,7 +343,7 @@ class ImageReferenceMetric(BaseMultimodalMetric):
         else:
             try:
                 self.success = self.score >= self.threshold
-            except:
+            except TypeError:
                 self.success = False
         return self.success
 

deepeval/metrics/multimodal_metrics/text_to_image/text_to_image.py

@@ -3,7 +3,7 @@ from typing import Optional, List, Tuple, Union
 import math
 import textwrap
 
-from deepeval.metrics import BaseMultimodalMetric
+from deepeval.metrics import BaseMetric
 from deepeval.test_case import LLMTestCaseParams, LLMTestCase, MLLMImage
 from deepeval.metrics.multimodal_metrics.text_to_image.template import (
     TextToImageTemplate,
@@ -14,9 +14,10 @@ from deepeval.utils import (
 )
 from deepeval.metrics.utils import (
     construct_verbose_logs,
-    trimAndLoadJson,
-    check_mllm_test_case_params,
+    check_llm_test_case_params,
     initialize_model,
+    a_generate_with_schema_and_extract,
+    generate_with_schema_and_extract,
 )
 from deepeval.models import DeepEvalBaseLLM
 from deepeval.metrics.multimodal_metrics.text_to_image.schema import ReasonScore
@@ -28,7 +29,7 @@ required_params: List[LLMTestCaseParams] = [
 ]
 
 
-class TextToImageMetric(BaseMultimodalMetric):
+class TextToImageMetric(BaseMetric):
     def __init__(
         self,
         model: Optional[Union[str, DeepEvalBaseLLM]] = None,
@@ -50,8 +51,14 @@ class TextToImageMetric(BaseMultimodalMetric):
         _show_indicator: bool = True,
         _in_component: bool = False,
     ) -> float:
-        check_mllm_test_case_params(
-            test_case, required_params, 0, 1, self, self.model
+        check_llm_test_case_params(
+            test_case,
+            required_params,
+            0,
+            1,
+            self,
+            self.model,
+            test_case.multimodal,
         )
 
         self.evaluation_cost = 0 if self.using_native_model else None
@@ -97,7 +104,7 @@ class TextToImageMetric(BaseMultimodalMetric):
                     steps=[
                         f"Semantic Consistency Scores:\n{self.SC_scores}",
                         f"Semantic Consistency Reasoning:\n{self.SC_reasoning}",
-                        f"Perceptual Quality Scores:\n{self.SC_scores}",
+                        f"Perceptual Quality Scores:\n{self.PQ_scores}",
                         f"Perceptual Quality Reasoning:\n{self.PQ_reasoning}",
                         f"Score: {self.score}\nReason: {self.reason}",
                     ],
@@ -110,8 +117,14 @@ class TextToImageMetric(BaseMultimodalMetric):
         _show_indicator: bool = True,
         _in_component: bool = False,
     ) -> float:
-        check_mllm_test_case_params(
-            test_case, required_params, 0, 1, self, self.model
+        check_llm_test_case_params(
+            test_case,
+            required_params,
+            0,
+            1,
+            self,
+            self.model,
+            test_case.multimodal,
         )
 
         self.evaluation_cost = 0 if self.using_native_model else None
@@ -150,7 +163,7 @@ class TextToImageMetric(BaseMultimodalMetric):
                 steps=[
                     f"Semantic Consistency Scores:\n{self.SC_scores}",
                     f"Semantic Consistency Reasoning:\n{self.SC_reasoning}",
-                    f"Perceptual Quality Scores:\n{self.SC_scores}",
+                    f"Perceptual Quality Scores:\n{self.PQ_scores}",
                     f"Perceptual Quality Reasoning:\n{self.PQ_reasoning}",
                     f"Score: {self.score}\nReason: {self.reason}",
                 ],
@@ -174,8 +187,7 @@ class TextToImageMetric(BaseMultimodalMetric):
         text_prompt: str,
         actual_image_output: MLLMImage,
     ) -> Tuple[List[int], str]:
-        images: List[MLLMImage] = []
-        images.append(actual_image_output)
+        images: List[MLLMImage] = [actual_image_output]
         prompt = f"""
             {
                 TextToImageTemplate.generate_semantic_consistency_evaluation_results(
@@ -185,28 +197,20 @@ class TextToImageMetric(BaseMultimodalMetric):
             Images:
             {images}
         """
-        if self.using_native_model:
-            res, cost = await self.model.a_generate(prompt, ReasonScore)
-            self.evaluation_cost += cost
-            return res.score, res.reasoning
-        else:
-            try:
-                res: ReasonScore = await self.model.a_generate(
-                    prompt, schema=ReasonScore
-                )
-                return res.score, res.reasoning
-            except TypeError:
-                res = await self.model.a_generate(prompt, input_text=prompt)
-                data = trimAndLoadJson(res, self)
-                return data["score"], data["reasoning"]
+        return await a_generate_with_schema_and_extract(
+            metric=self,
+            prompt=prompt,
+            schema_cls=ReasonScore,
+            extract_schema=lambda s: (s.score, s.reasoning),
+            extract_json=lambda data: (data["score"], data["reasoning"]),
+        )
 
     def _evaluate_semantic_consistency(
         self,
         text_prompt: str,
         actual_image_output: MLLMImage,
     ) -> Tuple[List[int], str]:
-        images: List[MLLMImage] = []
-        images.append(actual_image_output)
+        images: List[MLLMImage] = [actual_image_output]
         prompt = f"""
             {
                 TextToImageTemplate.generate_semantic_consistency_evaluation_results(
@@ -216,20 +220,13 @@ class TextToImageMetric(BaseMultimodalMetric):
             Images:
             {images}
         """
-        if self.using_native_model:
-            res, cost = self.model.generate(prompt, ReasonScore)
-            self.evaluation_cost += cost
-            return res.score, res.reasoning
-        else:
-            try:
-                res: ReasonScore = self.model.generate(
-                    prompt, schema=ReasonScore
-                )
-                return res.score, res.reasoning
-            except TypeError:
-                res = self.model.generate(prompt)
-                data = trimAndLoadJson(res, self)
-                return data["score"], data["reasoning"]
+        return generate_with_schema_and_extract(
+            metric=self,
+            prompt=prompt,
+            schema_cls=ReasonScore,
+            extract_schema=lambda s: (s.score, s.reasoning),
+            extract_json=lambda data: (data["score"], data["reasoning"]),
+        )
 
     async def _a_evaluate_perceptual_quality(
         self, actual_image_output: MLLMImage
@@ -242,20 +239,13 @@ class TextToImageMetric(BaseMultimodalMetric):
             Images:
             {images}
         """
-        if self.using_native_model:
-            res, cost = await self.model.a_generate(prompt, ReasonScore)
-            self.evaluation_cost += cost
-            return res.score, res.reasoning
-        else:
-            try:
-                res: ReasonScore = await self.model.a_generate(
-                    prompt, schema=ReasonScore
-                )
-                return res.score, res.reasoning
-            except TypeError:
-                res = await self.model.a_generate(prompt)
-                data = trimAndLoadJson(res, self)
-                return data["score"], data["reasoning"]
+        return await a_generate_with_schema_and_extract(
+            metric=self,
+            prompt=prompt,
+            schema_cls=ReasonScore,
+            extract_schema=lambda s: (s.score, s.reasoning),
+            extract_json=lambda data: (data["score"], data["reasoning"]),
+        )
 
     def _evaluate_perceptual_quality(
         self, actual_image_output: MLLMImage
@@ -268,22 +258,15 @@ class TextToImageMetric(BaseMultimodalMetric):
             Images:
             {images}
         """
-        if self.using_native_model:
-            res, cost = self.model.generate(prompt + images, ReasonScore)
-            self.evaluation_cost += cost
-            return res.score, res.reasoning
-        else:
-            try:
-                res: ReasonScore = self.model.generate(
-                    prompt, schema=ReasonScore
-                )
-                return res.score, res.reasoning
-            except TypeError:
-                res = self.model.generate(prompt)
-                data = trimAndLoadJson(res, self)
-                return data["score"], data["reasoning"]
+        return generate_with_schema_and_extract(
+            metric=self,
+            prompt=prompt,
+            schema_cls=ReasonScore,
+            extract_schema=lambda s: (s.score, s.reasoning),
+            extract_json=lambda data: (data["score"], data["reasoning"]),
+        )
 
-    def _calculate_score(self) -> List[str]:
+    def _calculate_score(self) -> float:
         min_SC_score = min(self.SC_scores)
         min_PQ_score = min(self.PQ_scores)
         return math.sqrt(min_SC_score * min_PQ_score) / 10
@@ -293,14 +276,12 @@ class TextToImageMetric(BaseMultimodalMetric):
             self.success = False
         else:
             try:
-                self.score >= self.threshold
-            except:
+                self.success = self.score >= self.threshold
+            except TypeError:
                 self.success = False
         return self.success
 
-    def _generate_reason(
-        self,
-    ) -> Tuple[List[float], str]:
+    def _generate_reason(self) -> str:
         return textwrap.dedent(
             f"""
             The overall score is {self.score:.2f} because the lowest score from semantic consistency was {min(self.SC_scores)}