Day 1 - Module 1: Exercises

These exercises are based on the concepts and code presented in Module 1: Basic LLM Interaction & Tool Calling (src/01-basics/).

Exercise 1.1: Changing Language and Persona

Goal: Practice modifying the system prompt to change the LLM's behavior.

1. Open the src/01-basics/hello-world.py script.
2. Modify the system message content from "antworte alles in französisch" to instruct the model to respond like a pirate (e.g., "Answer everything like a pirate.").
3. Modify the user message content to ask a different question (e.g., "What is the weather like today?").
4. Run the script (python hello-world.py) and observe the pirate-themed response.

Exercise 1.2: Using the Time Tool for a Different City

Goal: Practice changing the user input to trigger a tool call with different arguments.

1. Open the src/01-basics/tool-calling.py script.
2. Locate the initial user message:

   messages=[
       {
           "role": "user",
           "content": "What time is it in my current location?",
       }
   ]
   

3. Change the content to ask for the time in a specific city, for example, Tokyo: "What time is it in Tokyo?".
4. Run the script (python tool-calling.py).
5. Observe the output. Does the get_current_time function get called with Asia/Tokyo (or a similar valid timezone identifier derived by the LLM)? Does the final model response include the correct time for Tokyo?

Exercise 1.3 (Advanced): Adding a Simple Calculator Tool

Goal: Practice defining and integrating a new tool for the LLM to use.

Tool calling allows the LLM to interact with external functions or APIs to retrieve information or perform actions. The diagram below illustrates the concept:

1. Open the src/01-basics/tool-calling.py script.
2. Define a new Python function: Add a function that performs simple addition.

   def add_numbers(a: int, b: int) -> str:
       \"\"\"Adds two integer numbers together.\"\"\"
       print(f"Calling function 'add_numbers' with {a} and {b}")
       return str(a + b)
   

3. Define the tool schema: Create a new dictionary describing the add_numbers function to the LLM, similar to the get_current_time schema.

   add_tool = {
       "type": "function",
       "function": {
           "name": "add_numbers",
           "description": "Adds two integer numbers together.",
           "parameters": {
               "type": "object",
               "properties": {
                   "a": {
                       "type": "integer",
                       "description": "The first number.",
                   },
                   "b": {
                       "type": "integer",
                       "description": "The second number.",
                   }
               },
               "required": ["a", "b"],
           },
       },
   }
   

4. Update the API calls:
   • Modify the tools list passed to client.chat.completions.create to include both the original time tool (tool) and the new add_tool:

     tools=[tool, add_tool]
     

   • Ensure this updated tools list is used in both the initial API call and the second call made after handling a potential tool invocation.
5. Update the user query: Change the initial user message to ask an addition question, e.g., "What is 123 plus 456?".
6. Run the script: python3.11 tool-calling.py.
7. Observe the output. Does the LLM correctly identify the need for the add_numbers tool? Is the function called with the correct arguments? Does the final response contain the sum?