Manab's Notes - 08-Function calling

Manab's Notes

Function calling lets you connect models to external tools and APIs. Instead of generating text responses, the model determines when to call specific functions and provides the necessary parameters to execute real-world actions.

LLMs determine when to use function calling by analyzing the user prompt to detect if it requires executing an external function or tool.

Signup to weather api : https://www.weatherapi.com/ and copy the api key

Call the url

curl --location 'http://api.weatherapi.com/v1/current.json?key=e30eda3442bf4a16851213341261502&q=SO534NG&aqi=yes'

Sample response: Its inside the subpage

https://www.thebasu.co.uk/ms-azure/ai/08-function-calling/08-sampleresponse

Now integrate function call api with the prompt flow. Go to the foundry portal

Go to the hub based project and prompt flow

Create a new standard prompt flow

Add a new LLM - function_calling

Configure the LLM and click on Function calling

Copy the below function definition and copy it

[{"name":"get_weather_description","description":"get the air quality index of a particular location","parameters":{"type":"object","properties":{"location":{"type":"string","description":"The city and state, e.g. San Francisco, CA"}},"required":["location"]}}]

The description is the driver for LLM to take decision if the call will go the function

Type auto for the second parameter

Type the below prompt

# system:

you are a helpful AI assistant, meant to assist the user query based upon the user query.

# user:

Set the input of the LLM

Set the output block

Now need to write a python code to capture the content from the API response

Copy the below code

from promptflow import tool

import requests

import json

api_key="YOUR_API_KEY"

# The inputs section will change based on the arguments of the tool function, after you save the code

# Adding type to arguments and return value will help the system show the types properly

# Please update the function name/signature per need

def get_weather_description(location: str)->str:

url="http://api.weatherapi.com/v1/current.json?q=" + str(location) + "&key=" + str(api_key) + "&aqi=yes" #setting the url to be sent API request to

response = requests.get(url).json() #sending GET request to that URL

#storing values of different AQI parameters

aqi = response["current"]["air_quality"]

co_concentration = aqi["co"]

no2_concentration = aqi["no2"]

o3_concentration = aqi["o3"]

so2_concentration = aqi["so2"]

pm2_5_concentration = aqi["pm2_5"]

pm10_concentration = aqi["pm10"]

final_response = "CO concentration : {}, NO2 concentration : {}, O3 concentration: {}, SO2 concentration: {}, PM2.5 concentration: {}, PM10 concentration: {}".format(co_concentration, no2_concentration, o3_concentration, so2_concentration, pm2_5_concentration, pm10_concentration )

return final_response

@tool

def run_functions(response_message: dict)->str:

function_call = response_message.get("function_call", None)

if function_call and "name" in function_call and "arguments" in function_call:

function_name = function_call["name"]

function_args = json.loads(function_call["arguments"])

print(function_args)

result = globals()[function_name](**function_args)

else:

print("No function call")

if isinstance(response_message, dict):