k1

Author: = Enea_Gore

modules/text/module_text_llm/module_text_llm/__main__.py

@@ -13,7 +13,7 @@
 from module_text_llm.approach_controller import generate_suggestions
 from module_text_llm.helpers.detect_suspicios_submission import hybrid_suspicion_score, llm_check
 from module_text_llm.helpers.feedback_icl.store_feedback_icl import store_feedback_icl
-
+from module_text_llm.few_shot_chain_of_thought_approach import FewShotChainOfThoughtConfig
 #Test Demo
 @submissions_consumer
 def receive_submissions(exercise: Exercise, submissions: List[Submission]):
@@ -42,15 +42,19 @@ def process_incoming_feedback(exercise: Exercise, submission: Submission, feedba
 async def suggest_feedback(exercise: Exercise, submission: Submission, is_graded: bool, module_config: Configuration) -> List[Feedback]:
     logger.info("suggest_feedback: %s suggestions for submission %d of exercise %d were requested, with approach: %s",
                 "Graded" if is_graded else "Non-graded", submission.id, exercise.id, module_config.approach.__class__.__name__)
-    logger.info("suggest_feedback: %s suggestions for submission %d of exercise %d were requested",
-                "Graded" if is_graded else "Non-graded", submission.id, exercise.id)
-    is_sus, score = hybrid_suspicion_score(submission.text, threshold=0.8)
-    if is_sus:
-        logger.info("Suspicious submission detected with score %f", score)
-        is_suspicious,suspicios_text = await llm_check(submission.text)
-        if is_suspicious:
-            logger.info("Suspicious submission detected by LLM with text %s", suspicios_text)
-            return [Feedback(title="Instructors need to review this submission", description="This Submission potentially violates the content policy!", credits=-1.0, exercise_id=exercise.id, submission_id=submission.id, is_graded=is_graded)]
+
+    if not is_graded:    
+        is_sus, score = hybrid_suspicion_score(submission.text, threshold=0.8)
+        if is_sus:
+            logger.info("Suspicious submission detected with score %f", score)
+            is_suspicious,suspicios_text = await llm_check(submission.text)
+            if is_suspicious:
+                logger.info("Suspicious submission detected by LLM with text %s", suspicios_text)
+                return [Feedback(title="Instructors need to review this submission", description="This Submission potentially violates the content policy!", credits=-1.0, exercise_id=exercise.id, submission_id=submission.id, is_graded=is_graded)]
+        module_config.approach = FewShotChainOfThoughtConfig()
+        return await generate_suggestions(exercise, submission, module_config.approach, module_config.debug, is_graded)
+
+    
     return await generate_suggestions(exercise, submission, module_config.approach, module_config.debug, is_graded)
 
 

modules/text/module_text_llm/module_text_llm/helpers/feedback_icl/store_feedback_icl.py

@@ -21,33 +21,17 @@ def store_feedback_icl(submission: Submission, exercise: Exercise, feedbacks: Li
 
 def save_embedding_with_metadata(embedding,submission, exercise_id, metadata):
     embeddings_file = f"embeddings_{exercise_id}.npy"
-    full_return = []
-    print("inside reference")
-    print(type(metadata))
-    print(metadata.keys())
-    # print()
-    # print(metadata["feedbacks"])
-    # for data in metadata["feedbacks"]:
-    # if "index_start" in metadata.keys() and "index_end" in metadata.keys():
-    #     print(metadata["index_start"])
-    #     print(metadata["index_end"])
-        
+
     if metadata["index_start"] is not None and metadata["index_end"] is not None:
         reference = submission.text[metadata["index_start"]:metadata["index_end"]]
         metadata["text_reference"] = reference
-    #     print(reference)
-    #     full_return.append(metadata)
-    print(full_return)
+
     try:
-        print("inside try")
         if os.path.exists(embeddings_file):
-            # Load existing data
             existing_data = np.load(embeddings_file, allow_pickle=True).item()
-            # Append the new embedding and metadata
             existing_data['embeddings'] = np.vstack((existing_data['embeddings'], embedding))
             existing_data['metadata'].append(metadata)
         else:
-            # Create a new dictionary with embeddings and metadata
             existing_data = {
                 'embeddings': np.array([embedding], dtype=np.float32),
                 'metadata': [metadata]
@@ -78,7 +62,7 @@ def save_embeddings_to_file(embeddings, filename="keyword_embeddings.npy"):
     print(f"Embeddings saved to {filename}")
 
 
-def query_embedding(query_embedding, exercise_id, k=5):
+def query_embedding(query_embedding, exercise_id, k=1):
     """
     Query the top-k most similar embeddings to the provided query_embedding
     for a given exercise ID. Return the corresponding metadata for these embeddings.