Customizing the Chat Response

The default response type of the language model messages is str. However, you can overwrite the send method to return a different type.
Just like tasks, you can control the type hints by declaring the return type of the send method.

from typing import List
import declarai
gpt_35 = declarai.openai(model="gpt-3.5-turbo")

@gpt_35.experimental.chat
class SQLBot:
    """
    You are a sql assistant."""
    ...

    def send(self, operation: str) -> List[str]:
        ...

sql_bot = SQLBot()
print(sql_bot.send(message="Offer two sql queries that use the 'SELECT' operation"))
> [
    "SELECT * FROM table_name;",
    "SELECT column_name FROM table_name;"
]

Warning

As with tasks, the message is sent along with the expected return types. This means that if not careful, a message conflicting with the expected results could cause weird behavior in the llm responses.
For more best-practices, see here.