Update app.py

app.py

@@ -323,7 +323,7 @@ def respond(message, history, model, temperature, num_calls, use_web_search, sel
             embed = get_embeddings()
             if os.path.exists("faiss_database"):
                 database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
-                retriever = database.as_retriever(search_kwargs={"k": 5})
+                retriever = database.as_retriever(search_kwargs={"k": 10})
                 
                 # Filter relevant documents based on user selection
                 all_relevant_docs = retriever.get_relevant_documents(message)
@@ -455,33 +455,20 @@ def get_response_from_pdf(query, model, selected_docs, num_calls=3, temperature=
     if os.path.exists("faiss_database"):
         logging.info("Loading FAISS database")
         database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
-        
-        # Log the contents of the FAISS database
-        logging.info(f"FAISS database contains {len(database.docstore._dict)} documents")
-        for doc_id, doc in database.docstore._dict.items():
-            logging.info(f"Document ID: {doc_id}, Source: {doc.metadata['source']}")
     else:
         logging.warning("No FAISS database found")
         yield "No documents available. Please upload PDF documents to answer questions."
         return
 
-    retriever = database.as_retriever(search_kwargs={"k": 20})
+    retriever = database.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)}")
     
-    # Log details of retrieved documents
-    for i, doc in enumerate(relevant_docs):
-        logging.info(f"Relevant doc {i}: Source: {doc.metadata['source']}, Content preview: {doc.page_content[:100]}...")
-    
     # 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)}")
     
-    # Log details of filtered documents
-    for i, doc in enumerate(filtered_docs):
-        logging.info(f"Filtered doc {i}: Source: {doc.metadata['source']}, Content preview: {doc.page_content[:100]}...")
-
     if not filtered_docs:
         logging.warning(f"No relevant information found in the selected documents: {selected_docs}")
         yield "No relevant information found in the selected documents. Please try selecting different documents or rephrasing your query."
@@ -489,56 +476,40 @@ def get_response_from_pdf(query, model, selected_docs, num_calls=3, temperature=
 
     for doc in filtered_docs:
         logging.info(f"Document source: {doc.metadata['source']}")
-        logging.info(f"Document content preview: {doc.page_content[:100]}...")
-
-    # Implement a sliding window approach for context
-    max_context_length = 4000  # Adjust based on your model's capacity
-    context_chunks = []
-    current_chunk = ""
-    for doc in filtered_docs:
-        if len(current_chunk) + len(doc.page_content) > max_context_length:
-            context_chunks.append(current_chunk)
-            current_chunk = doc.page_content
-        else:
-            current_chunk += "\n" + doc.page_content
-    if current_chunk:
-        context_chunks.append(current_chunk)
+        logging.info(f"Document content preview: {doc.page_content[:100]}...")  # Log first 100 characters of each document
 
-    logging.info(f"Number of context chunks: {len(context_chunks)}")
+    context_str = "\n".join([doc.page_content for doc in filtered_docs])
+    logging.info(f"Total context length: {len(context_str)}")
 
-    for i, context_str in enumerate(context_chunks):
-        logging.info(f"Processing context chunk {i+1}/{len(context_chunks)}")
-        logging.info(f"Context chunk length: {len(context_str)}")
-
-        if model == "@cf/meta/llama-3.1-8b-instruct":
-            logging.info("Using Cloudflare API")
-            for response in get_response_from_cloudflare(prompt="", context=context_str, query=query, num_calls=num_calls, temperature=temperature, search_type="pdf"):
-                yield response
-        else:
-            logging.info("Using Hugging Face API")
-            prompt = f"""Using the following context from the PDF documents:
+    if model == "@cf/meta/llama-3.1-8b-instruct":
+        logging.info("Using Cloudflare API")
+        # Use Cloudflare API with the retrieved context
+        for response in get_response_from_cloudflare(prompt="", context=context_str, query=query, num_calls=num_calls, temperature=temperature, search_type="pdf"):
+            yield response
+    else:
+        logging.info("Using Hugging Face API")
+        # Use Hugging Face API
+        prompt = f"""Using the following context from the PDF documents:
 {context_str}
 Write a detailed and complete response that answers the following user question: '{query}'"""
         
-            client = InferenceClient(model, token=huggingface_token)
+        client = InferenceClient(model, token=huggingface_token)
         
-            response = ""
-            for j in range(num_calls):
-                logging.info(f"API call {j+1}/{num_calls}")
-                for message in client.chat_completion(
-                    messages=[{"role": "user", "content": prompt}],
-                    max_tokens=10000,
-                    temperature=temperature,
-                    stream=True,
-                ):
-                    if message.choices and message.choices[0].delta and message.choices[0].delta.content:
-                        chunk = message.choices[0].delta.content
-                        response += chunk
-                        yield response  # Yield partial response
+        response = ""
+        for i in range(num_calls):
+            logging.info(f"API call {i+1}/{num_calls}")
+            for message in client.chat_completion(
+                messages=[{"role": "user", "content": prompt}],
+                max_tokens=10000,
+                temperature=temperature,
+                stream=True,
+            ):
+                if message.choices and message.choices[0].delta and message.choices[0].delta.content:
+                    chunk = message.choices[0].delta.content
+                    response += chunk
+                    yield response  # Yield partial response
         
-            logging.info("Finished generating response for this context chunk")
-
-    logging.info("Finished processing all context chunks")
+        logging.info("Finished generating response")
 
 def vote(data: gr.LikeData):
     if data.liked: