Update app.py

app.py

@@ -460,25 +460,32 @@ def get_response_from_pdf(query, model, selected_docs, num_calls=3, temperature=
         yield "No documents available. Please upload PDF documents to answer questions."
         return
 
-    retriever = database.as_retriever(search_kwargs={"k": 20})
-    logging.info(f"Retrieving relevant documents for query: {query}")
-    relevant_docs = retriever.get_relevant_documents(query)
-    logging.info(f"Number of relevant documents retrieved: {len(relevant_docs)}")
-    
-    # Filter relevant_docs based on selected documents
-    filtered_docs = [doc for doc in relevant_docs if doc.metadata["source"] in selected_docs]
-    logging.info(f"Number of filtered documents: {len(filtered_docs)}")
+    # Pre-filter the docstore
+    filtered_docstore = {
+        k: v for k, v in database.docstore._dict.items() 
+        if v.metadata["source"] in selected_docs
+    }
     
-    if not filtered_docs:
-        logging.warning(f"No relevant information found in the selected documents: {selected_docs}")
+    logging.info(f"Number of documents after pre-filtering: {len(filtered_docstore)}")
+
+    if not filtered_docstore:
+        logging.warning(f"No documents found for the selected sources: {selected_docs}")
         yield "No relevant information found in the selected documents. Please try selecting different documents or rephrasing your query."
         return
 
-    for doc in filtered_docs:
+    # Create a new FAISS index with only the selected documents
+    filtered_db = FAISS(embed.embed_query, filtered_docstore, database.index)
+    
+    retriever = filtered_db.as_retriever(search_kwargs={"k": 10})
+    logging.info(f"Retrieving relevant documents for query: {query}")
+    relevant_docs = retriever.get_relevant_documents(query)
+    logging.info(f"Number of relevant documents retrieved: {len(relevant_docs)}")
+
+    for doc in relevant_docs:
         logging.info(f"Document source: {doc.metadata['source']}")
         logging.info(f"Document content preview: {doc.page_content[:100]}...")  # Log first 100 characters of each document
 
-    context_str = "\n".join([doc.page_content for doc in filtered_docs])
+    context_str = "\n".join([doc.page_content for doc in relevant_docs])
     logging.info(f"Total context length: {len(context_str)}")
 
     if model == "@cf/meta/llama-3.1-8b-instruct":