Exploring Model Parameters in PHP

Welcome back! In the previous lesson, you learned how to send a simple message to OpenAI's language model and receive a response using PHP. Now, we will take a step further by exploring model parameters that allow you to customize the AI's responses. These parameters are crucial for tailoring the chatbot's behavior to meet specific needs. In this lesson, we will focus on three key parameters: max_completion_tokens, temperature, and top_p. Understanding these parameters will enable you to control the length, creativity, and word choice of the AI's responses, enhancing your chatbot's functionality.

The max_completion_tokens parameter sets a hard limit on the number of tokens the AI can generate in its response. A "token" can be a whole word or just part of a word. For example, "chatbot" might be one token, while "hello" could be split into two tokens: "hel" and "lo." It’s important to note that token counts vary across different models, words, and languages — so the same text might have a different token count depending on these factors.

When you set max_completion_tokens, you specify the maximum number of tokens the AI can produce. This is a strict limit, meaning the model will stop generating text once it reaches this count, even if it results in an incomplete sentence or word.

Here’s an example where we set max_completion_tokens to 100:

<?php
require 'vendor/autoload.php';

// Fetch API Key
$apiKey = $_ENV['OPENAI_API_KEY'] ?? getenv('OPENAI_API_KEY');
$baseUrl = $_ENV['OPENAI_BASE_URI'] ?? getenv('OPENAI_BASE_URI');

$client = \OpenAI::factory()
    ->withApiKey($apiKey)
    ->withBaseUri($baseUrl)
    ->make();

// Define a simple user message to test the API
$prompt = "Describe a sunset over the ocean";

// Get response with specific parameters
$response = $client->chat()->create([
    'model' => 'gpt-6-luna',
    'reasoning_effort' => 'none',
    'messages' => [['role' => 'user', 'content' => $prompt]],
    'max_completion_tokens' => 100
]);

// Process the response
$reply = trim($response['choices'][0]['message']['content']);
echo "Assistant: $reply\n";
?>

By setting max_completion_tokens to 100, you cap the AI’s response at 100 tokens, which may result in abrupt cut-offs if the model hasn't finished its thought. This parameter helps you control response length, making it useful for avoiding verbose answers and reducing the cost of API requests.

Note: API usage costs money. Each request consumes both input tokens (from the user's message) and output tokens (generated by the AI). Setting max_completion_tokens helps control response length and manage costs.
If the AI's response gets cut off due to reaching the max_completion_tokens limit, you can check response['choices'][0]['finish_reason']. If it returns 'length', the response was truncated, and you might need to adjust max_completion_tokens accordingly.

The temperature parameter is a fascinating aspect of AI interaction. It controls the randomness or creativity of the AI's responses. A lower temperature value, such as 0.2, makes the AI's output more deterministic and focused, often resulting in more predictable responses. Conversely, a higher temperature value, like 0.8, encourages the AI to generate more diverse and creative responses, which can be useful for tasks requiring imaginative outputs.

For example, consider the following code snippet where we set the temperature to 0.7:

<?php
// Get response with specific parameters
$response = $client->chat()->create([
    'model' => 'gpt-6-luna',
    'reasoning_effort' => 'none',
    'messages' => [['role' => 'user', 'content' => $prompt]],
    'temperature' => 0.7
]);

// Process the response
$reply = trim($response['choices'][0]['message']['content']);
echo "Assistant: $reply\n";
?>

With a temperature of 0.7, the AI is likely to provide a response that balances creativity and coherence. Experimenting with different temperature values will help you find the right balance for your specific use case.

The top_p parameter controls the AI's word choice through a technique called nucleus sampling. At each step, the model ranks every possible next word by probability, then considers only the smallest group of words whose probabilities add up to top_p. A low value, such as 0.3, keeps only the most likely candidates, producing safe and predictable wording. A high value, like 1.0, leaves the full range of candidates available, allowing more surprising and varied phrasing.

Consider the following code where we set top_p to 0.3:

<?php
// Get response with specific parameters
$response = $client->chat()->create([
    'model' => 'gpt-6-luna',
    'reasoning_effort' => 'none',
    'messages' => [['role' => 'user', 'content' => $prompt]],
    'top_p' => 0.3
]);

// Process the response
$reply = trim($response['choices'][0]['message']['content']);
echo "Assistant: $reply\n";
?>

With a top_p of 0.3, the AI sticks to the wording it considers most likely, which is useful when you want consistent, on-message answers — a customer service bot quoting support hours, for example.

Both temperature and top_p control how varied the AI's output is, but they do it differently: temperature reshapes the probabilities of all candidate words, while top_p trims which candidates are considered at all. Because they adjust the same underlying behavior, tuning both at once makes results hard to reason about — a change in one can mask a change in the other.

The practical guidance is to pick one and leave the other at its default. Use temperature when you want to dial overall creativity up or down; use top_p when you want to tighten or loosen word choice specifically.

In the following example, we rely on top_p alone and leave temperature unset:

<?php
// Get response with specific parameters
$response = $client->chat()->create([
    'model' => 'gpt-6-luna',
    'reasoning_effort' => 'none',
    'messages' => [['role' => 'user', 'content' => $prompt]],
    'top_p' => 1.0
]);

// Process the response
$reply = trim($response['choices'][0]['message']['content']);
echo "Assistant: $reply\n";
?>

Raising top_p to 1.0 opens up the full set of candidate words, so the same prompt can produce noticeably different phrasing each run. Try the same prompt at 0.3 and at 1.0 and compare how much the wording shifts.

Let's bring it all together with a complete code example that incorporates all the parameters we've discussed:

<?php
require 'vendor/autoload.php';

// Fetch API Key
$apiKey = $_ENV['OPENAI_API_KEY'] ?? getenv('OPENAI_API_KEY');
$baseUrl = $_ENV['OPENAI_BASE_URI'] ?? getenv('OPENAI_BASE_URI');

$client = \OpenAI::factory()
    ->withApiKey($apiKey)
    ->withBaseUri($baseUrl)
    ->make();

// Define a simple user message to test the API
$prompt = "Describe a sunset over the ocean";

// Get response with specific parameters
$response = $client->chat()->create([
    'model' => 'gpt-6-luna',
    'reasoning_effort' => 'none',
    'messages' => [['role' => 'user', 'content' => $prompt]],
    'max_completion_tokens' => 100,   // Limits response length
    'top_p' => 0.3  // Narrows word choice (use instead of temperature)
]);

// Process the response
$reply = trim($response['choices'][0]['message']['content']);
echo "Assistant: $reply\n";
?>

In this example, we cap the response length and narrow the word choice, leaving temperature at its default since top_p is already doing that job. By adjusting these parameters, you can fine-tune the chatbot's behavior to meet your specific requirements. When you run this code, you should see a response that reflects the balance of length and word choice that you've set.

In this lesson, we explored how to use model parameters to customize AI responses. You learned about the max_completion_tokens, temperature, and top_p parameters, saw how they can be applied in code, and learned why temperature and top_p should not be tuned at the same time. These tools allow you to control the length, creativity, and word choice of the AI's responses, enhancing your chatbot's functionality.

As you move on to the practice exercises, I encourage you to experiment with different parameter settings to see their effects firsthand. This hands-on practice will reinforce what you've learned and prepare you for the next unit, where we'll delve deeper into managing conversation history and message types. Keep up the great work, and enjoy the journey of creating your chatbot with OpenAI!