simplest version

utils.py

@@ -24,15 +24,9 @@ def format_sources(nodes):
         doc_id = meta.get('document_id', 'unknown')
         
         if doc_type == 'table':
-            table_id = meta.get('table_identifier', meta.get('table_number', 'unknown'))
+            table_num = meta.get('table_number', 'unknown')
             title = meta.get('table_title', '')
-            section = meta.get('section', '')
-            source = f"📊 {doc_id} - {table_id}"
-            if title:
-                source += f": {title}"
-            if section:
-                source += f" ({section})"
-            sources.append(source)
+            sources.append(f"📊 {doc_id} - Таблица {table_num}: {title}")
         elif doc_type == 'image':
             img_num = meta.get('image_number', 'unknown')
             sources.append(f"🖼️ {doc_id} - Рисунок {img_num}")
@@ -40,7 +34,7 @@ def format_sources(nodes):
             section = meta.get('section_id', '')
             sources.append(f"📄 {doc_id} - Раздел {section}")
     
-    return "\n".join(sources)  # Don't use set() to preserve order
+    return "\n".join(set(sources))
 
 def preprocess_query(question):
     import re
@@ -79,68 +73,52 @@ def preprocess_query(question):
 
 def answer_question(question, query_engine, reranker):
     try:
-        log_message(f"\n{'='*70}")
         log_message(f"Query: {question}")
-        log_message(f"{'='*70}")
         
         enhanced_query = preprocess_query(question)
         if enhanced_query != question:
             log_message(f"Enhanced query: {enhanced_query}")
         
         retrieved = query_engine.retriever.retrieve(enhanced_query)
-        log_message(f"\n📥 INITIAL RETRIEVAL: {len(retrieved)} nodes")
+        log_message(f"Retrieved {len(retrieved)} nodes")
         
-        # Detailed logging
-        doc_ids = {}
-        for n in retrieved:
-            doc_id = n.metadata.get('document_id', 'unknown')
-            if doc_id not in doc_ids:
-                doc_ids[doc_id] = {'tables': [], 'text': 0, 'images': 0}
-            
-            if n.metadata.get('type') == 'table':
-                table_id = n.metadata.get('table_identifier', n.metadata.get('table_number', ''))
-                doc_ids[doc_id]['tables'].append(table_id)
-            elif n.metadata.get('type') == 'image':
-                doc_ids[doc_id]['images'] += 1
-            else:
-                doc_ids[doc_id]['text'] += 1
-        
-        for doc_id, counts in doc_ids.items():
-            log_message(f"  📄 {doc_id}:")
-            if counts['tables']:
-                log_message(f"    Tables: {', '.join(set(counts['tables']))}")
-            if counts['text']:
-                log_message(f"    Text chunks: {counts['text']}")
-            if counts['images']:
-                log_message(f"    Images: {counts['images']}")
+        doc_ids = [n.metadata.get('document_id', 'unknown') for n in retrieved]
+        table_nums = [n.metadata.get('table_number', '') for n in retrieved if n.metadata.get('type') == 'table']
+        log_message(f"Retrieved from documents: {set(doc_ids)}")
+        if table_nums:
+            log_message(f"Retrieved tables: {set(table_nums)}")
         
         reranked = rerank_nodes(question, retrieved, reranker, top_k=25)
-        log_message(f"\n🔄 AFTER RERANKING: {len(reranked)} nodes")
+        log_message(f"Reranked to {len(reranked)} nodes")
         
-        # Detailed reranking results
-        doc_ids_reranked = {}
+        doc_ids_reranked = [n.metadata.get('document_id', 'unknown') for n in reranked]
+        table_nums_reranked = [n.metadata.get('table_number', '') for n in reranked if n.metadata.get('type') == 'table']
+        log_message(f"After reranking - documents: {set(doc_ids_reranked)}")
+        if table_nums_reranked:
+            log_message(f"After reranking - tables: {set(table_nums_reranked)}")
+        
+        context_parts = []
         for n in reranked:
-            doc_id = n.metadata.get('document_id', 'unknown')
-            if doc_id not in doc_ids_reranked:
-                doc_ids_reranked[doc_id] = {'tables': [], 'text': 0, 'images': 0}
+            meta = n.metadata
+            doc_id = meta.get('document_id', 'unknown')
+            doc_type = meta.get('type', 'text')
             
-            if n.metadata.get('type') == 'table':
-                table_id = n.metadata.get('table_identifier', n.metadata.get('table_number', ''))
-                doc_ids_reranked[doc_id]['tables'].append(table_id)
-            elif n.metadata.get('type') == 'image':
-                doc_ids_reranked[doc_id]['images'] += 1
+            if doc_type == 'table':
+                table_num = meta.get('table_number', 'unknown')
+                title = meta.get('table_title', '')
+                source_label = f"[ТАБЛИЦА {table_num} - {doc_id}]"
+                if title:
+                    source_label += f" {title}"
+            elif doc_type == 'image':
+                img_num = meta.get('image_number', 'unknown')
+                source_label = f"[РИСУНОК {img_num} - {doc_id}]"
             else:
-                doc_ids_reranked[doc_id]['text'] += 1
+                section = meta.get('section_id', '')
+                source_label = f"[{doc_id} - {section}]"
+            
+            context_parts.append(f"{source_label}\n{n.text}")
         
-        for doc_id, counts in doc_ids_reranked.items():
-            log_message(f"  📄 {doc_id}:")
-            if counts['tables']:
-                log_message(f"    Tables: {', '.join(set(counts['tables']))}")
-            if counts['text']:
-                log_message(f"    Text chunks: {counts['text']}")
-            if counts['images']:
-                log_message(f"    Images: {counts['images']}")
-        context = "\n\n" + ("="*70 + "\n\n").join(doc_ids_reranked)
+        context = "\n\n" + ("="*70 + "\n\n").join(context_parts)
         
         prompt = f"""Ты эксперт по технической документации.