chat = llm.predict({
  "inputs":tok.apply_chat_template(messages,tokenize=False,add_generation_prompt=True), # convert messages to model input
  "parameters":payload
})
 
print(chat[0]["generated_text"])