Code Changes - Reranking Implementation

app.py

@@ -169,7 +169,7 @@ initial_plot = plot_subset_metrics(initial_subset) if initial_subset else "No da
 def generate_advance_report(subset_dropdown, dataset_type_dropdown, chunking_dropdown, 
                             embed_dropdown, generator_dropdown,
                             chunk_count, retriever_type, noOfQuestions,
-                            reranking_checkbox, evaluator_dropdown):
+                            reranking_checkbox, reranking_dropdown, evaluator_dropdown):
     
     export_data = {
             "metadata": {
@@ -188,6 +188,7 @@ def generate_advance_report(subset_dropdown, dataset_type_dropdown, chunking_dro
                 "noOfQuestions": noOfQuestions,
                 "retriever_type": retriever_type,
                 "reranking": reranking_checkbox,
+                "reranking_method": reranking_dropdown if reranking_checkbox else None,
                 "evaluator_model": evaluator_dropdown
             }
         }
@@ -244,7 +245,7 @@ def generate_advance_report(subset_dropdown, dataset_type_dropdown, chunking_dro
 def generate_file(subset_dropdown, dataset_type_dropdown, 
                         chunking_dropdown, embed_dropdown, generator_dropdown,
                         chunk_count, retriever_type,
-                        reranking_checkbox, evaluator_dropdown,
+                        reranking_checkbox, reranking_dropdown, evaluator_dropdown,
                         orig_ans_display, y_metrics_display,
                         gen_ans_display, y_pred_metrics_display,
                         chunks_retrieved_display, 
@@ -268,6 +269,7 @@ def generate_file(subset_dropdown, dataset_type_dropdown,
                 "chunk_count": chunk_count,
                 "retriever_type": retriever_type,
                 "reranking": reranking_checkbox,
+                "reranking_method": reranking_dropdown if reranking_checkbox else None,
                 "evaluator_model": evaluator_dropdown
             },
             "results": {
@@ -291,7 +293,7 @@ def generate_file(subset_dropdown, dataset_type_dropdown,
     return json_str, fileName
 
 def run_rag_pipeline_multiple_questions(subset, chunking, embed_model, retriever, noOfQuestions, retriever_type, 
-                                        chunk_count, reranking, evaluator):
+                                        chunk_count, reranking, reranking_dropdown, evaluator):
     print(f"Running RAG Pipeline for {noOfQuestions} questions in subset: {subset}")
     global advanced_analysis
     try:
@@ -302,6 +304,9 @@ def run_rag_pipeline_multiple_questions(subset, chunking, embed_model, retriever
         logger.error(f"Failed to load data: {e}")
         return None
     
+    ranking_method = reranking_dropdown if reranking else None
+    print(f"Using reranking: {reranking}, method: {ranking_method}")
+
     print("Starting RAG pipeline for {noOfQuestions} questions!!!")
 
     ragSystemObject = RAGSystem(
@@ -310,7 +315,8 @@ def run_rag_pipeline_multiple_questions(subset, chunking, embed_model, retriever
         strategy=chunking,
         chunks=[],  # Not needed for loading
         generator_model_name=retriever,
-        retriever_model_name=embed_model
+        retriever_model_name=embed_model,
+        reranker_model_name=ranking_method
     )
 
     # 3. Load or use stored vector DB
@@ -427,42 +433,6 @@ def run_rag_pipeline(subset, question, custom_question, chunking, embed_model, r
 
     y_pred_metrics, evaluator_json_output = _evaluate_using_groq(context_docs, final_question, generated_answer)
 
-
-    # response_sentences = form_response_sentences(generated_answer)
-
-    # # print(f"\nResponse Sentences: {response_sentences}")
-
-    # print(f"Length of Response Sentences: {len(response_sentences)}")
-    # print(f"Length of Document Sentences : {len(document_sentences)}")
-
-    # y_pred_metrics = {
-    #     "relevance_score": "NA",
-    #     "utilization_score": "NA",
-    #     "completeness_score": "NA",
-    #     "adherence_score": "NA"
-    # }
-
-    # # Call evaluator with the right variables
-    # try:
-    #     grok_api_key = os.environ.get("GROQ_API_KEY")  # Safely loaded from HF Secrets
-
-    #     evaluator = RAGEvaluator(
-    #         use_groq=True,
-    #         groq_api_key=grok_api_key,
-    #         groq_model="llama3-70b-8192"
-    #     )
-        
-    #     result = evaluator.evaluate(document_sentences, question, response_sentences)
-    #     print(f"\nResult----\n: {result}")
-    #     if result is not None:
-    #         y_pred_metrics = evaluator.extract_trace_metrics_from_json(result, len(document_sentences))
-    #         evaluator_json_output = json.dumps(result, indent=4)
-    #         print(f"Result: {evaluator_json_output}")
-    #         print(f"Metrics: {y_pred_metrics}")
-    #     else:
-    #         print("No result obtained for this question")
-    # except Exception as e:
-    #     print(f"Exception Raised in evaluation / extract_trace_metrics_from_json. Details: {e}")
     
     # Format as list of lists
     formatted_chunks = [
@@ -609,7 +579,7 @@ with gr.Blocks(
                         subset_dropdown,
                         chunking_dropdown, embed_dropdown, generator_dropdown,
                         noOfQuestions, retriever_type, chunk_count,
-                        reranking_checkbox, evaluator_dropdown
+                        reranking_checkbox, reranking_dropdown, evaluator_dropdown
                     ] 
                 )
 
@@ -635,7 +605,7 @@ with gr.Blocks(
         inputs=[subset_dropdown, dataset_type_dropdown, 
                 chunking_dropdown, embed_dropdown, generator_dropdown,
                 chunk_count, retriever_type,
-                reranking_checkbox, evaluator_dropdown,
+                reranking_checkbox, reranking_dropdown, evaluator_dropdown,
                 orig_ans_display, y_metrics_display,
                 gen_ans_display, y_pred_metrics_display,
                 chunks_retrieved_display, 
@@ -649,7 +619,7 @@ with gr.Blocks(
         inputs=[subset_dropdown, dataset_type_dropdown, 
                 chunking_dropdown, embed_dropdown, generator_dropdown,
                 chunk_count, retriever_type, noOfQuestions,
-                reranking_checkbox, evaluator_dropdown
+                reranking_checkbox, reranking_dropdown, evaluator_dropdown
                 ],
             outputs=[json_output, download_file]
         )