log full completion result with semantic similarity comparison results

src/text_generation/services/guardrails/reflexion_security_guidelines_service.py

@@ -42,7 +42,7 @@ class ReflexionSecurityGuardrailsService(
         self.llm_configuration_introspection_service = llm_configuration_introspection_service
 
         # constants
-        self.COSINE_SIMILARITY_RISK_THRESHOLD = 0.75
+        self.COSINE_SIMILARITY_RISK_THRESHOLD = 0.8
 
         # set up semantic similarity service and supporting texts
         self.example_prompt_injection_completions = prompt_injection_example_service.get_all_completions()
@@ -128,7 +128,7 @@ class ReflexionSecurityGuardrailsService(
             
             # update completion result with similarity scoring threshold and result
             result.guardrails_result.cosine_similarity_risk_threshold = self.COSINE_SIMILARITY_RISK_THRESHOLD
-            result.guardrails_result.cosine_similarity_score = similarity_result.max
+            result.guardrails_result.append_semantic_similarity_result(semantic_similarity_result=similarity_result)
 
             # return raw result if the completion comparison score didn't exceed threshold
             if not result.guardrails_result.is_completion_malicious():

src/text_generation/services/logging/test_run_logging_service.py

@@ -7,6 +7,7 @@ import time
 from datetime import datetime
 from typing import Any, Dict, List
 
+from src.text_generation.domain.text_generation_completion_result import TextGenerationCompletionResult
 from src.text_generation.services.logging.abstract_test_run_logging_service import AbstractTestRunLoggingService
 
 
@@ -36,22 +37,18 @@ class TestRunLoggingService(AbstractTestRunLoggingService):
     def log_results(
             self, 
             id: str,
-            prompt: str,
-            final_completion: str,
+            text_generation_completion_result: TextGenerationCompletionResult,
+            final_completion_text_score: float,
             is_rag_few_shot_enabled: bool,
             is_cot_enabled: bool,
             is_reflexion_enabled: bool,
-            original_llm_config: dict,
-            scores: List[float],
-            mean: float,
-            max: float):
+            original_llm_config: dict):
         with self._lock:
             logs = self._read_logs()
             log_entry = {
                 "id": id,
                 "timestamp": datetime.now().isoformat(),
-                "original_prompt": prompt,
-                "final_completion": final_completion,
+                "final_completion_text_score": final_completion_text_score,
                 "mitigations_enabled": {
                     "guidelines": {
                         "rag_with_few_shot_examples": is_rag_few_shot_enabled,
@@ -62,11 +59,7 @@ class TestRunLoggingService(AbstractTestRunLoggingService):
                     }
                 },
                 "original_llm_config": original_llm_config,
-                "cosine_similarity": {
-                    "mean": mean,
-                    "max": max,
-                    "all_scores": scores
-                }
+                "text_generation_completion_result": text_generation_completion_result.__dict__
             }
             logs.append(log_entry)
             self._write_logs(logs)

src/text_generation/services/nlp/text_generation_completion_service.py

@@ -57,7 +57,7 @@ class TextGenerationCompletionService(
         self.reflexion_guardrails: AbstractGeneratedTextGuardrailService = reflexion_guardrails
 
         # constants
-        self.COSINE_SIMILARITY_RISK_THRESHOLD = 0.75
+        self.COSINE_SIMILARITY_RISK_THRESHOLD = 0.8
 
         # default guidelines settings
         self._use_guidelines = False
@@ -121,8 +121,8 @@ class TextGenerationCompletionService(
         
         # update completion result with similarity scoring threshold and result
         completion_result.guidelines_result.cosine_similarity_risk_threshold = self.COSINE_SIMILARITY_RISK_THRESHOLD
-        completion_result.guidelines_result.cosine_similarity_score = similarity_result.max
-        
+        completion_result.guidelines_result.append_semantic_similarity_result(semantic_similarity_result=similarity_result)
+
         # return raw result if the completion comparison score didn't exceed threshold
         if not completion_result.guidelines_result.is_completion_malicious():
             print(f'Guidelines-based completion was NOT malicious. Score: {completion_result.guidelines_result.cosine_similarity_score}')