Update app.py

app.py

@@ -203,6 +203,8 @@ class PDFRAGSystem:
         """관련 청크 검색"""
         all_relevant_chunks = []
         
+        print(f"Searching chunks for query: '{query[:50]}...' in {len(doc_ids)} documents")
+        
         if self.embedder and self.embeddings_store:
             # 임베딩 기반 검색
             query_embedding = self.embedder.encode([query])[0]
@@ -218,58 +220,68 @@ class PDFRAGSystem:
                         sim = np.dot(query_embedding, emb) / (np.linalg.norm(query_embedding) * np.linalg.norm(emb))
                         similarities.append(sim)
                     
-                    # 상위 청크 선택
+                    # 상위 청크 선택 - 임계값 낮춤
                     top_indices = np.argsort(similarities)[-top_k:][::-1]
                     
                     for idx in top_indices:
-                        if similarities[idx] > 0.2:
+                        if similarities[idx] > 0.1:  # 임계값을 0.2에서 0.1로 낮춤
                             all_relevant_chunks.append({
                                 "content": chunks[idx],
                                 "doc_name": self.documents[doc_id]["metadata"]["file_name"],
                                 "similarity": similarities[idx]
                             })
+                            print(f"Found chunk with similarity: {similarities[idx]:.3f}")
         else:
             # 키워드 기반 검색
+            print("Using keyword-based search (embedder not available)")
             query_keywords = set(query.lower().split())
             
             for doc_id in doc_ids:
                 if doc_id in self.document_chunks:
                     chunks = self.document_chunks[doc_id]
-                    for i, chunk in enumerate(chunks[:5]):  # 처음 5개만
+                    for i, chunk in enumerate(chunks):  # 모든 청크 검색
                         chunk_lower = chunk.lower()
                         score = sum(1 for keyword in query_keywords if keyword in chunk_lower)
                         if score > 0:
                             all_relevant_chunks.append({
-                                "content": chunk[:500],
+                                "content": chunk[:800],  # 더 긴 청크 사용
                                 "doc_name": self.documents[doc_id]["metadata"]["file_name"],
                                 "similarity": score / len(query_keywords) if query_keywords else 0
                             })
         
         # 정렬 및 반환
         all_relevant_chunks.sort(key=lambda x: x.get('similarity', 0), reverse=True)
-        return all_relevant_chunks[:top_k]
+        result = all_relevant_chunks[:top_k]
+        print(f"Returning {len(result)} chunks")
+        return result
     
     def create_rag_prompt(self, query: str, doc_ids: List[str], top_k: int = 3) -> tuple:
         """RAG 프롬프트 생성 - 쿼리와 컨텍스트를 분리하여 반환"""
+        print(f"Creating RAG prompt for query: '{query[:50]}...' with docs: {doc_ids}")
+        
         relevant_chunks = self.search_relevant_chunks(query, doc_ids, top_k)
         
         if not relevant_chunks:
+            print("No relevant chunks found")
             return query, ""
         
+        print(f"Found {len(relevant_chunks)} relevant chunks")
+        
         # 컨텍스트 구성
         context_parts = []
-        context_parts.append("다음 문서 내용을 참고하여 답변해주세요:\n")
+        context_parts.append("Based on the following document context, please answer the question below:")
         context_parts.append("=" * 40)
         
         for i, chunk in enumerate(relevant_chunks, 1):
-            context_parts.append(f"\n[참고 {i} - {chunk['doc_name']}]")
-            content = chunk['content'][:300] if len(chunk['content']) > 300 else chunk['content']
+            context_parts.append(f"\n[Document Reference {i} - {chunk['doc_name']}]")
+            content = chunk['content'][:500] if len(chunk['content']) > 500 else chunk['content']
             context_parts.append(content)
+            print(f"Added chunk {i} with similarity: {chunk.get('similarity', 0):.3f}")
         
         context_parts.append("\n" + "=" * 40)
         
         context = "\n".join(context_parts)
-        enhanced_query = f"{context}\n\n질문: {query}"
+        enhanced_query = f"{context}\n\nQuestion: {query}\n\nAnswer based on the document context provided above:"
         
         return enhanced_query, context
 
@@ -304,7 +316,12 @@ def format_conversation_history(chat_history):
 @spaces.GPU()
 def generate_response(input_data, chat_history, max_new_tokens, system_prompt, temperature, top_p, top_k, repetition_penalty):
     """Generate response with optional RAG enhancement"""
-    global last_context
+    global last_context, rag_enabled, selected_docs, top_k_chunks
+    
+    # Debug logging
+    print(f"RAG Enabled: {rag_enabled}")
+    print(f"Selected Docs: {selected_docs}")
+    print(f"Available Docs: {list(rag_system.documents.keys())}")
     
     # Apply RAG if enabled
     if rag_enabled and selected_docs:
@@ -312,9 +329,11 @@ def generate_response(input_data, chat_history, max_new_tokens, system_prompt, t
         enhanced_input, context = rag_system.create_rag_prompt(input_data, doc_ids, top_k_chunks)
         last_context = context
         actual_input = enhanced_input
+        print(f"RAG Applied - Original: {len(input_data)} chars, Enhanced: {len(enhanced_input)} chars")
     else:
         actual_input = input_data
         last_context = ""
+        print("RAG Not Applied")
     
     # Prepare messages
     new_message = {"role": "user", "content": actual_input}
@@ -432,6 +451,9 @@ def update_rag_settings(enable, docs, k):
     selected_docs = docs if docs else []
     top_k_chunks = k
     
+    # Debug logging
+    print(f"RAG Settings Updated - Enabled: {rag_enabled}, Docs: {selected_docs}, Top-K: {top_k_chunks}")
+    
     status = "✅ Enabled" if enable and docs else "⭕ Disabled"
     status_html = f"<div class='pdf-status pdf-info'>🔍 RAG: <strong>{status}</strong></div>"