Function Calling in AI Applications

Function calling allows you to integrate Large Language Models (LLMs) with external tools for effective usage and interaction with APIs.

LLMs like GPT-4 and GPT-3.5 can detect when a function needs to be called and generate JSON arguments for those functions. These functions act as tools in your AI application, and multiple functions can be defined in a single request.

Basic Sequence of Steps for Function Calling

1. Initial Call: Call the model with the user query and a set of functions defined in the functions parameter.

2. Function Detection: The model can choose to call one or more functions. The content will be a stringified JSON object in your custom schema (note: the model may hallucinate parameters).

3. Parsing and Execution: Parse the string into JSON in your code, and call your function with the provided arguments if they exist.

4. Response Handling: Call the model again, appending the function response as a new message. Let the model summarize the results back to the user.

Example: Sending Email with Gemini

def multiply(a:float, b:float):
    """returns a * b."""
    return a*b



def send_email(destination_email: str, message_text: str):
    """Send message to user_email with message_text"""

    email = ''
    password = ''
    dest_email = destination_email
    subject = 'Test'
    email_text = message_text

    message = 'From: {}\nTo: {}\nSubject: {}\n\n{}'.format(email,
                                                        dest_email, 
                                                        subject, 
                                                        email_text)

    server = smtp.SMTP_SSL('smtp.yandex.com')
    server.set_debuglevel(1)
    server.ehlo(email)
    server.login(email, password)
    server.auth_plain()
    server.sendmail(email, dest_email, message)
    server.quit()

model = genai.GenerativeModel(model_name='gemini-1.5-flash',
                              tools=[multiply, send_email])

chat = model.start_chat(enable_automatic_function_calling=True)

response = chat.send_message('Send a message to email about meeting on June 13. And tell him to call my number back. Be kind')
response.text

Contributing

We welcome contributions from the community. Please read our contributing guidelines and code of conduct.

License

function-calling is licensed under the MIT License. See the LICENSE file for more information.

Contact

For any questions, issues, or suggestions, please open an issue on our GitHub repository.