refactoring

src/text_generation/dependency_injection_container.py

@@ -4,10 +4,10 @@ from src.text_generation.adapters.embedding_model import EmbeddingModel
 from src.text_generation.adapters.text_generation_foundation_model import TextGenerationFoundationModel
 from src.text_generation.entrypoints.http_api_controller import HttpApiController
 from src.text_generation.entrypoints.server import RestApiServer
-from src.text_generation.services.language_models.text_generation_response_service import TextGenerationResponseService
-from src.text_generation.services.language_models.retrieval_augmented_generation_response_service import RetrievalAugmentedGenerationResponseService
-from src.text_generation.services.similarity_scoring.generated_text_guardrail_service import GeneratedTextGuardrailService
 from src.text_generation.services.logging.file_logging_service import FileLoggingService
+from src.text_generation.services.nlp.text_generation_response_service import TextGenerationResponseService
+from src.text_generation.services.nlp.retrieval_augmented_generation_response_service import RetrievalAugmentedGenerationResponseService
+from src.text_generation.services.guardrails.generated_text_guardrail_service import GeneratedTextGuardrailService
 
 
 class DependencyInjectionContainer(containers.DeclarativeContainer):

src/text_generation/entrypoints/http_api_controller.py

@@ -1,10 +1,10 @@
 import json
 import traceback
 
-from src.text_generation.services.language_models.text_generation_response_service import TextGenerationResponseService
-from src.text_generation.services.language_models.retrieval_augmented_generation_response_service import RetrievalAugmentedGenerationResponseService
+from src.text_generation.services.nlp.text_generation_response_service import TextGenerationResponseService
+from src.text_generation.services.nlp.retrieval_augmented_generation_response_service import RetrievalAugmentedGenerationResponseService
 from src.text_generation.services.logging.file_logging_service import FileLoggingService
-from src.text_generation.services.similarity_scoring.generated_text_guardrail_service import GeneratedTextGuardrailService
+from src.text_generation.services.guardrails.generated_text_guardrail_service import GeneratedTextGuardrailService
 
 
 class HttpApiController:

src/text_generation/services/guardrails/generated_text_guardrail_service.py

@@ -1,35 +1,17 @@
-import numpy
-from sklearn.metrics.pairwise import cosine_similarity
-
-from src.text_generation.ports.abstract_embedding_model import AbstractEmbeddingModel
 from src.text_generation.services.guardrails.abstract_generated_text_guardrail_service import AbstractGeneratedTextGuardrailService
+from src.text_generation.services.nlp.abstract_semantic_similarity_service import AbstractSemanticSimilarityService
 
 
 class GeneratedTextGuardrailService(AbstractGeneratedTextGuardrailService):
     def __init__(
-            self, 
-            embedding_model: AbstractEmbeddingModel,
+            self,
+            semantic_similarity_service: AbstractSemanticSimilarityService,
             comparison_texts: list[str]):
         super().__init__()
-        self.embeddings = embedding_model.embeddings
-        self.comparison_texts = comparison_texts
+        self.semantic_similarity_service = semantic_similarity_service
+        self.semantic_similarity_service.use_comparison_texts(comparison_texts)
+        self.cosine_similarity_risk_threshold: float = 0.5
 
     def analyze(self, model_generated_text: str) -> float:
-        # Get embeddings
-        query_embedding = self.embeddings.embed_query(model_generated_text)
-        doc_embeddings = self.embeddings.embed_documents(self.comparison_texts)
-
-        # Calculate similarity scores
-        query_embedding = numpy.array(query_embedding).reshape(1, -1)
-        doc_embeddings = numpy.array(doc_embeddings)
-
-        similarity_scores = cosine_similarity(query_embedding, doc_embeddings)[0]
-
-        scores = list()
-
-        # Results will be floating point values between -1 and 1
-        for i, score in enumerate(similarity_scores):
-            print(f"========  Text {i+1}: {score:.4f} | Score type: {type(score)}")
-            scores.append(score)
-
-        return max(scores)
+        score: float = self.semantic_similarity_service.analyze(text=model_generated_text)
+        return score >= self.cosine_similarity_risk_threshold

src/text_generation/services/guidelines/abstract_semantic_similarity_guidelines_service.py

@@ -0,0 +1,7 @@
+import abc
+
+
+class AbstractSemanticSimilarityGuidelinesService(abc.ABC):
+    @abc.abstractmethod
+    def analyze(self, prompt_input_text: str) -> float:
+        raise NotImplementedError

src/text_generation/services/guidelines/rag_guidelines_service.py

@@ -6,14 +6,13 @@ from src.text_generation.common.constants import Constants
 from src.text_generation.services.guidelines.abstract_rag_guidelines_service import AbstractRetrievalAugmentedGenerationGuidelinesService
 from src.text_generation.ports.abstract_embedding_model import AbstractEmbeddingModel
 
+
 class RetrievalAugmentedGenerationGuidelinesService(
-        AbstractRetrievalAugmentedGenerationGuidelinesService
-    ):
+        AbstractRetrievalAugmentedGenerationGuidelinesService):
 
     def __init__(
             self,
-            embedding_model: AbstractEmbeddingModel
-        ):
+            embedding_model: AbstractEmbeddingModel):
         self.constants = Constants()
         self.embedding_model = embedding_model
 
@@ -35,6 +34,7 @@ class RetrievalAugmentedGenerationGuidelinesService(
         )
         split_docs = text_splitter.split_documents(data)
 
+        # TODO: log?
         i = 1
         for doc in split_docs:
             print(f'{i}: {doc.page_content}\n\n')

src/text_generation/services/guidelines/semantic_similarity_guidelines_service.py

@@ -0,0 +1,7 @@
+import abc
+
+
+class SemanticSimilarityGuidelinesService(abc.ABC):
+    @abc.abstractmethod
+    def analyze(self, prompt_input_text: str) -> float:
+        raise NotImplementedError

src/text_generation/services/nlp/abstract_semantic_similarity_service.py

@@ -0,0 +1,11 @@
+import abc
+
+
+class AbstractSemanticSimilarityService(abc.ABC):
+    @abc.abstractmethod
+    def analyze(self, text: str) -> float:
+        raise NotImplementedError
+    
+    @abc.abstractmethod
+    def use_comparison_texts(self, comparison_texts: list[str]):
+        raise NotImplementedError

src/text_generation/services/nlp/fake_language_model_response_service.py

@@ -1,4 +1,4 @@
-from src.text_generation.services.language_models.abstract_language_model_response_service import AbstractLanguageModelResponseService
+from src.text_generation.services.nlp.abstract_language_model_response_service import AbstractLanguageModelResponseService
 
 
 class FakeLanguageModelResponseService(AbstractLanguageModelResponseService):

src/text_generation/services/nlp/retrieval_augmented_generation_response_service.py

@@ -3,7 +3,7 @@ from langchain.prompts import PromptTemplate
 
 from src.text_generation.ports.abstract_embedding_model import AbstractEmbeddingModel
 from src.text_generation.ports.abstract_foundation_model import AbstractFoundationModel
-from src.text_generation.services.language_models.abstract_language_model_response_service import AbstractLanguageModelResponseService
+from src.text_generation.services.nlp.abstract_language_model_response_service import AbstractLanguageModelResponseService
 from src.text_generation.services.guidelines.abstract_rag_guidelines_service import AbstractRetrievalAugmentedGenerationGuidelinesService
 from src.text_generation.services.utilities.abstract_response_processing_service import AbstractResponseProcessingService
 

src/text_generation/services/nlp/semantic_similarity_service.py

@@ -0,0 +1,34 @@
+from numpy import float64, array
+from sklearn.metrics.pairwise import cosine_similarity
+
+from src.text_generation.common.constants import Constants
+from src.text_generation.services.nlp.abstract_semantic_similarity_service import AbstractSemanticSimilarityService
+from src.text_generation.ports.abstract_embedding_model import AbstractEmbeddingModel
+
+
+class SemanticSimilarityService(AbstractSemanticSimilarityService):
+    def __init__(
+            self,
+            embedding_model: AbstractEmbeddingModel):
+        super().__init__()
+        self.embeddings = embedding_model.embeddings
+        self.constants = Constants()
+    
+    def use_comparison_texts(self, comparison_texts: list[str]):
+        self.comparison_texts = comparison_texts
+
+    def analyze(self, text: str) -> float:
+        query_embedding = self.embeddings.embed_query(text)
+        doc_embeddings = self.embeddings.embed_documents(self.comparison_texts)
+
+        query_embedding = array(query_embedding).reshape(1, -1)
+        doc_embeddings = array(doc_embeddings)
+        similarity_scores: list[float64] = cosine_similarity(query_embedding, doc_embeddings)[0]
+        scores = list()
+
+        # perfect alignment (similarity) results in a score of 1;
+        # opposite is -1
+        for _, score in enumerate(similarity_scores):
+            scores.append(score)
+
+        return max(scores)

src/text_generation/services/nlp/text_generation_response_service.py

@@ -2,7 +2,7 @@ from langchain.prompts import PromptTemplate
 from langchain_core.output_parsers import StrOutputParser
 from langchain_core.runnables import RunnablePassthrough
 
-from src.text_generation.services.language_models.abstract_language_model_response_service import AbstractLanguageModelResponseService
+from src.text_generation.services.nlp.abstract_language_model_response_service import AbstractLanguageModelResponseService
 from src.text_generation.ports.abstract_foundation_model import AbstractFoundationModel