Introduction to LLM Calls

Welcome to the first lesson of our course on building your own AI Cooking Helper. In this lesson, we will explore the concept of making basic LLM (Large Language Model) calls. LLMs, such as OpenAI's models, are powerful tools that can generate human-like text responses. They are integral to AI applications, enabling them to understand and naturally respond to user inputs. By the end of this lesson, you will understand how to make a basic LLM call and interpret its output.

Let's start by understanding the structure of the code used to make an LLM call. We will build this step-by-step.

First, we need to import the necessary libraries and set up the OpenAI client. This client will allow us to interact with the OpenAI API.

import os
from openai import OpenAI

client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"), base_url=os.getenv("OPENAI_BASE_URL"))

Here, we import the os module to access environment variables and the OpenAI class from the openai library. We then create an OpenAI client using the api_key and base_url stored in the environment variables. This client is essential for making requests to the OpenAI API.

Next, we need to define the prompts that will guide the model's behavior. There are two types of prompts: system prompts and user prompts.

system_prompt = "You are a coding assistant that talks like a pirate."
user_prompt = "What is the best cooking recipe in the world?"

• System Prompt: This sets the context for the model. In our example, the system_prompt instructs the model to act as a coding assistant that speaks like a pirate.
• User Prompt: This is the input from the user. Here, the user is asking what the best cooking recipe is.

These prompts are crucial as they shape the model's responses, ensuring they are relevant and contextually appropriate.

To control the model's output, we configure parameters such as reasoning_effort.

reasoning_effort = "medium"

• Reasoning Effort: For reasoning models like gpt-6-luna, the amount of reasoning and depth applied to generate an answer is controlled by the reasoning_effort parameter rather than temperature. Typical effort values are "low", "medium", and "high". A "low" effort produces faster, more concise responses with less reasoning overhead, while "high" allows the model to think more thoroughly through complex problems. In our example, "medium" strikes a balance between depth and response time.
• Model Selection: Instead of using a variable, we will pass the model name as a literal string directly into our request. We will use gpt-6-luna, which is a powerful and efficient model for these tasks.