Update app.py

app.py

@@ -81,7 +81,8 @@ def get_chain(temperature):
         "text-generation",
         model=model, 
         tokenizer=tokenizer, 
-        max_length=2048,
+        max_length=4000,
+        max_new_tokens = 500,
         temperature=temperature,
         top_p=0.95,
         repetition_penalty=1.15