# generation args
max_new_tokens: 1024
temperature: 0
stream: true