Update app.py

app.py

@@ -40,17 +40,25 @@ def get_model():
         huggingfacehub_api_token=huggingface_token
     )
 
+def generate_chunked_response(model, prompt, max_tokens=500, max_chunks=5):
+    full_response = ""
+    for i in range(max_chunks):
+        chunk = model(prompt + full_response, max_new_tokens=max_tokens)
+        full_response += chunk
+        if chunk.strip().endswith((".", "!", "?")):
+            break
+    return full_response
+
 def response(database, model, question):
     prompt_val = ChatPromptTemplate.from_template(prompt)
     retriever = database.as_retriever()
-    parser = StrOutputParser()
-    chain = (
-        {'context': retriever, 'question': RunnablePassthrough()}
-        | prompt_val
-        | model
-        | parser
-    )
-    ans = chain.invoke(question)
+    
+    context = retriever.get_relevant_documents(question)
+    context_str = "\n".join([doc.page_content for doc in context])
+    
+    formatted_prompt = prompt_val.format(context=context_str, question=question)
+    
+    ans = generate_chunked_response(model, formatted_prompt)
     return ans
 
 def update_vectors(file):