Update app.py

app.py

@@ -121,7 +121,7 @@ def get_conversation_chain(vectorstore):
     # llm = CTransformers(model="llama-2-7b-chat.ggmlv3.q2_K.bin", model_type="llama", config=config)
 
     llm = LlamaCpp(model_path=model_path,
-                   input={"temperature": 0.75, "max_length": 2000, "top_p": 1},
+                   input={"n_ctx":2048,"temperature": 0.75, "max_length": 2000, "top_p": 1},
                    verbose=True, )
     memory = ConversationBufferMemory(
         memory_key='chat_history', return_messages=True)