in ui changing possibility of topk

app.py

@@ -233,16 +233,52 @@ def switch_model(model_name, vector_index):
         log_message(error_msg)
         return None, f"❌ {error_msg}"
 
+# Add these global variables near the top with other globals
+retrieval_params = {
+    'vector_top_k': 50,
+    'bm25_top_k': 50,
+    'similarity_cutoff': 0.55,
+    'hybrid_top_k': 100,
+    'rerank_top_k': 20
+}
+
+# MODIFIED: Update create_query_engine call signature
+def create_query_engine(vector_index, vector_top_k=50, bm25_top_k=50, 
+                       similarity_cutoff=0.55, hybrid_top_k=100):
+    try:
+        from config import CUSTOM_PROMPT
+        from index_retriever import create_query_engine as create_index_query_engine
+        
+        # Pass parameters to the index_retriever function
+        query_engine = create_index_query_engine(
+            vector_index=vector_index,
+            vector_top_k=vector_top_k,
+            bm25_top_k=bm25_top_k,
+            similarity_cutoff=similarity_cutoff,
+            hybrid_top_k=hybrid_top_k
+        )
+        
+        log_message(f"Query engine created with params: vector_top_k={vector_top_k}, "
+                   f"bm25_top_k={bm25_top_k}, cutoff={similarity_cutoff}, hybrid_top_k={hybrid_top_k}")
+        return query_engine
+        
+    except Exception as e:
+        log_message(f"Ошибка создания query engine: {str(e)}")
+        raise
+
+# MODIFIED: Update answer_question to use global retrieval_params
 def main_answer_question(question):
-    global query_engine, reranker, current_model, chunks_df
+    global query_engine, reranker, current_model, chunks_df, retrieval_params
     if not question.strip():
         return ("<div style='color: black;'>Пожалуйста, введите вопрос</div>",
                 "<div style='color: black;'>Источники появятся после обработки запроса</div>",
                 "<div style='color: black;'>Чанки появятся после обработки запроса</div>")
 
     try:
-        # Call the answer_question function which returns 3 values
-        answer_html, sources_html, chunks_html = answer_question(question, query_engine, reranker, current_model, chunks_df)
+        answer_html, sources_html, chunks_html = answer_question(
+            question, query_engine, reranker, current_model, chunks_df,
+            rerank_top_k=retrieval_params['rerank_top_k']
+        )
         return answer_html, sources_html, chunks_html
 
     except Exception as e:
@@ -251,6 +287,37 @@ def main_answer_question(question):
                 "<div style='color: black;'>Источники недоступны из-за ошибки</div>",
                 "<div style='color: black;'>Чанки недоступны из-за ошибки</div>")
 
+# NEW: Function to update retrieval parameters and recreate query engine
+def update_retrieval_params(vector_top_k, bm25_top_k, similarity_cutoff, hybrid_top_k, rerank_top_k):
+    global query_engine, vector_index, retrieval_params
+    
+    try:
+        retrieval_params['vector_top_k'] = vector_top_k
+        retrieval_params['bm25_top_k'] = bm25_top_k
+        retrieval_params['similarity_cutoff'] = similarity_cutoff
+        retrieval_params['hybrid_top_k'] = hybrid_top_k
+        retrieval_params['rerank_top_k'] = rerank_top_k
+        
+        # Recreate query engine with new parameters
+        if vector_index is not None:
+            query_engine = create_query_engine(
+                vector_index=vector_index,
+                vector_top_k=vector_top_k,
+                bm25_top_k=bm25_top_k,
+                similarity_cutoff=similarity_cutoff,
+                hybrid_top_k=hybrid_top_k
+            )
+            log_message(f"Параметры поиска обновлены: vector_top_k={vector_top_k}, "
+                       f"bm25_top_k={bm25_top_k}, cutoff={similarity_cutoff}, "
+                       f"hybrid_top_k={hybrid_top_k}, rerank_top_k={rerank_top_k}")
+            return f"✅ Параметры обновлены"
+        else:
+            return "❌ Система не инициализирована"
+    except Exception as e:
+        error_msg = f"Ошибка обновления параметров: {str(e)}"
+        log_message(error_msg)
+        return f"❌ {error_msg}"
+
 def retrieve_chunks(question: str, top_k: int = 20) -> list:
     from index_retriever import rerank_nodes
     global query_engine, reranker
@@ -362,24 +429,132 @@ def create_demo_interface(answer_question_func, switch_model_func, current_model
                         label="Релевантные чанки",
                         value="<div style='background-color: #2d3748; color: white; padding: 20px; border-radius: 10px; text-align: center;'>Здесь появятся релевантные чанки...</div>",
                     )
+        
+        # NEW TAB: Retrieval Parameters
+        with gr.Tab("⚙️ Параметры поиска"):
+            gr.Markdown("### Настройка параметров векторного поиска и переранжирования")
+            
+            with gr.Row():
+                with gr.Column():
+                    vector_top_k = gr.Slider(
+                        minimum=10,
+                        maximum=200,
+                        value=50,
+                        step=10,
+                        label="Vector Top K",
+                        info="Количество результатов из векторного поиска"
+                    )
+                
+                with gr.Column():
+                    bm25_top_k = gr.Slider(
+                        minimum=10,
+                        maximum=200,
+                        value=50,
+                        step=10,
+                        label="BM25 Top K",
+                        info="Количество результатов из BM25 поиска"
+                    )
             
-            switch_btn.click(
-                fn=switch_model_func,
-                inputs=[model_dropdown],
-                outputs=[model_status]
+            with gr.Row():
+                with gr.Column():
+                    similarity_cutoff = gr.Slider(
+                        minimum=0.0,
+                        maximum=1.0,
+                        value=0.55,
+                        step=0.05,
+                        label="Similarity Cutoff",
+                        info="Минимальный порог схожести для векторного поиска"
+                    )
+                
+                with gr.Column():
+                    hybrid_top_k = gr.Slider(
+                        minimum=10,
+                        maximum=300,
+                        value=100,
+                        step=10,
+                        label="Hybrid Top K",
+                        info="Количество результатов из гибридного поиска"
+                    )
+            
+            with gr.Row():
+                with gr.Column():
+                    rerank_top_k = gr.Slider(
+                        minimum=5,
+                        maximum=100,
+                        value=20,
+                        step=5,
+                        label="Rerank Top K",
+                        info="Количество результатов после переранжирования"
+                    )
+                
+                with gr.Column():
+                    update_btn = gr.Button("Применить параметры", variant="primary")
+                    update_status = gr.Textbox(
+                        value="Параметры готовы к применению",
+                        label="Статус",
+                        interactive=False
+                    )
+            
+            gr.Markdown("""
+            ### Рекомендации:
+            - **Vector Top K**: Увеличьте для более полного поиска по семантике (50-100)
+            - **BM25 Top K**: Увеличьте для лучшего поиска по ключевым словам (30-80)
+            - **Similarity Cutoff**: Снизьте для более мягких критериев (0.3-0.6), повысьте для строгих (0.7-0.9)
+            - **Hybrid Top K**: Объединённые результаты (100-150)
+            - **Rerank Top K**: Финальные результаты (10-30)
+            """)
+            
+            update_btn.click(
+                fn=update_retrieval_params,
+                inputs=[vector_top_k, bm25_top_k, similarity_cutoff, hybrid_top_k, rerank_top_k],
+                outputs=[update_status]
             )
             
-            ask_btn.click(
-                fn=answer_question_func,
-                inputs=[question_input],
-                outputs=[answer_output, sources_output, chunks_output]
+            # Display current parameters
+            gr.Markdown("### Текущие параметры:")
+            current_params_display = gr.Textbox(
+                value="Vector: 50 | BM25: 50 | Cutoff: 0.55 | Hybrid: 100 | Rerank: 20",
+                label="",
+                interactive=False,
+                lines=2
             )
             
-            question_input.submit(
-                fn=answer_question_func,
-                inputs=[question_input],
-                outputs=[answer_output, sources_output, chunks_output]
+            def display_current_params():
+                return f"""Vector Top K: {retrieval_params['vector_top_k']}
+BM25 Top K: {retrieval_params['bm25_top_k']}
+Similarity Cutoff: {retrieval_params['similarity_cutoff']}
+Hybrid Top K: {retrieval_params['hybrid_top_k']}
+Rerank Top K: {retrieval_params['rerank_top_k']}"""
+            
+            # Refresh params display on tab change
+            demo.load(
+                fn=display_current_params,
+                outputs=[current_params_display]
+            )
+            
+            update_btn.click(
+                fn=display_current_params,
+                outputs=[current_params_display]
             )
+        
+        # Original tab logic
+        switch_btn.click(
+            fn=switch_model_func,
+            inputs=[model_dropdown],
+            outputs=[model_status]
+        )
+        
+        ask_btn.click(
+            fn=answer_question_func,
+            inputs=[question_input],
+            outputs=[answer_output, sources_output, chunks_output]
+        )
+        
+        question_input.submit(
+            fn=answer_question_func,
+            inputs=[question_input],
+            outputs=[answer_output, sources_output, chunks_output]
+        )
     return demo
 
 
@@ -389,13 +564,6 @@ reranker = None
 vector_index = None
 current_model = DEFAULT_MODEL
 
-def main_answer_question(question):
-    global query_engine, reranker, current_model, chunks_df
-    answer_html, sources_html, chunks_html = answer_question(
-        question, query_engine, reranker, current_model, chunks_df
-    )
-    return answer_html, sources_html, chunks_html
-
 def main_switch_model(model_name):
     global query_engine, vector_index, current_model
     

index_retriever.py

@@ -65,24 +65,26 @@ def rerank_nodes(query, nodes, reranker, top_k=25, min_score_threshold=0.5):
         log_message(f"Ошибка переранжировки: {str(e)}")
         return nodes[:top_k]
 
-def create_query_engine(vector_index):
+# MODIFIED: Update create_query_engine function signature
+def create_query_engine(vector_index, vector_top_k=50, bm25_top_k=50, 
+                       similarity_cutoff=0.55, hybrid_top_k=100):
     try:
         from config import CUSTOM_PROMPT
         
         bm25_retriever = BM25Retriever.from_defaults(
             docstore=vector_index.docstore, 
-            similarity_top_k=60 
+            similarity_top_k=bm25_top_k  # NOW PARAMETERIZED
         )
         
         vector_retriever = VectorIndexRetriever(
             index=vector_index, 
-            similarity_top_k=60,
-            similarity_cutoff=0.45
+            similarity_top_k=vector_top_k,  # NOW PARAMETERIZED
+            similarity_cutoff=similarity_cutoff  # NOW PARAMETERIZED
         )
         
         hybrid_retriever = QueryFusionRetriever(
             [vector_retriever, bm25_retriever],
-            similarity_top_k=120,
+            similarity_top_k=hybrid_top_k,  # NOW PARAMETERIZED
             num_queries=1
         )
         
@@ -97,7 +99,9 @@ def create_query_engine(vector_index):
             response_synthesizer=response_synthesizer
         )
         
-        log_message("Query engine успешно создан")
+        log_message(f"Query engine created: vector_top_k={vector_top_k}, "
+                   f"bm25_top_k={bm25_top_k}, similarity_cutoff={similarity_cutoff}, "
+                   f"hybrid_top_k={hybrid_top_k}")
         return query_engine
         
     except Exception as e:

utils.py

@@ -197,8 +197,8 @@ def debug_search_tables(vector_index, search_term="С-25"):
 
 from documents_prep import normalize_text
 
-# MODIFIED: Update answer_question function
-def answer_question(question, query_engine, reranker, current_model, chunks_df=None):
+# MODIFIED: Update answer_question function signature
+def answer_question(question, query_engine, reranker, current_model, chunks_df=None, rerank_top_k=20):
     # NORMALIZE the question to convert C to С
     normalized_question = normalize_text(question)
     
@@ -226,8 +226,9 @@ def answer_question(question, query_engine, reranker, current_model, chunks_df=N
             log_message(f"  [{i+1}] {doc_id} - Table {table_num}: {table_title[:50]}")
         log_message(f"UNIQUE NODES: {len(unique_retrieved)} nodes")
         
-        # Simple reranking with NORMALIZED question
-        reranked_nodes = rerank_nodes(normalized_question, unique_retrieved, reranker, top_k=20)
+        # Simple reranking with NORMALIZED question and PARAMETERIZED top_k
+        reranked_nodes = rerank_nodes(normalized_question, unique_retrieved, reranker, 
+                                     top_k=rerank_top_k)  # NOW PARAMETERIZED
         
         # Direct query without formatting - use normalized question
         response = query_engine.query(normalized_question)
@@ -243,7 +244,7 @@ def answer_question(question, query_engine, reranker, current_model, chunks_df=N
         <h3 style='color: #63b3ed; margin-top: 0;'>Ответ (Модель: {current_model}):</h3>
         <div style='line-height: 1.6; font-size: 16px;'>{response.response}</div>
         <div style='margin-top: 15px; padding-top: 10px; border-top: 1px solid #4a5568; font-size: 14px; color: #a0aec0;'>
-        Время обработки: {processing_time:.2f} секунд
+        Время обработки: {processing_time:.2f} секунд | Переранжировано: {len(reranked_nodes)} документов
         </div>
         </div>"""
         log_message(f"Model Answer: {response.response}")