README

PHP library for the OpenAI API

An unofficial library for OpenAI's API.

Used by

• DrupalGen - ChatGPT content generation for Drupal.
• GPT Fortune - Generates a fortune cookie message using ChatGPT.
• PHP AI Agents - AI agents written in PHP.

Requirements

• PHP 8.1 or above
• Composer
• An OpenAI API key

Quick set up

git clone git@github.com:ruscoe/openai-php.git

cd openai-php

composer install

Usage examples

The following examples assume you store your API key in an environment variable. To do this on a Linux or MacOS system, run:

export OPENAI_API_KEY=sk-XA0yN...

Be sure to substitute your own API key after OPENAI_API_KEY=.

Completions

This example asks the gpt-4o model to describe a keyboard. It instructs the OpenAI API to use more than the default number of tokens so a reasonable length description can be returned.

<?php

require __DIR__ . '/vendor/autoload.php';

// @see https://platform.openai.com/docs/api-reference/authentication
$api_key = getenv('OPENAI_API_KEY');

$api = new OpenAI\OpenAICompletions($api_key);

$parameters = [
    'max_completion_tokens' => 128,
];

$messages = [
    (object) ['role' => 'user', 'content' => 'Describe a keyboard']
];

$response = $api->create('gpt-4o', $messages, 1, $parameters);

var_dump($response);

The response:

object(stdClass)#35 (8) {
  ["id"]=>
  string(38) "chatcmpl-BLXYoMfybsQQjP1xlKjUtEL6chvHE"
  ["object"]=>
  string(15) "chat.completion"
  ["created"]=>
  int(1744473594)
  ["model"]=>
  string(17) "gpt-4o-2024-08-06"
  ["choices"]=>
  array(1) {
    [0]=>
    object(stdClass)#33 (4) {
      ["index"]=>
      int(0)
      ["message"]=>
      object(stdClass)#19 (4) {
        ["role"]=>
        string(9) "assistant"
        ["content"]=>
        string(604) "A keyboard is an input device used primarily to input text and commands into a computer or other electronic devices. Here are some key features and components of a typical keyboard:

1. **Keys**: A standard keyboard contains a variety of keys, each with a specific function. These include:
   - **Alphanumeric keys**: These are similar to those on a typewriter and include letters (A-Z), numbers (0-9), and punctuation symbols.
   - **Function keys**: Usually labeled F1 through F12, these keys are located at the top of the keyboard and offer shortcuts for various system and application functions.
   -"
        ["refusal"]=>
        NULL
        ["annotations"]=>
        array(0) {
        }
      }
      ["logprobs"]=>
      NULL
      ["finish_reason"]=>
      string(6) "length"
    }
  }
  ["usage"]=>
  object(stdClass)#29 (5) {
    ["prompt_tokens"]=>
    int(10)
    ["completion_tokens"]=>
    int(128)
    ["total_tokens"]=>
    int(138)
    ["prompt_tokens_details"]=>
    object(stdClass)#22 (2) {
      ["cached_tokens"]=>
      int(0)
      ["audio_tokens"]=>
      int(0)
    }
    ["completion_tokens_details"]=>
    object(stdClass)#20 (4) {
      ["reasoning_tokens"]=>
      int(0)
      ["audio_tokens"]=>
      int(0)
      ["accepted_prediction_tokens"]=>
      int(0)
      ["rejected_prediction_tokens"]=>
      int(0)
    }
  }
  ["service_tier"]=>
  string(7) "default"
  ["system_fingerprint"]=>
  string(13) "fp_b7faba9ef5"

Chat

This example sends a simple chat message.

<?php

require __DIR__ . '/vendor/autoload.php';

// @see https://platform.openai.com/docs/api-reference/authentication
$api_key = getenv('OPENAI_API_KEY');

$api = new OpenAI\OpenAIChat($api_key);

$messages = [
    (object) ['role' => 'user', 'content' => 'Hello, friend!'],
];

$response = $api->create('gpt-4o', $messages);

var_dump($response);

The response:

object(stdClass)#35 (8) {
  ["id"]=>
  string(38) "chatcmpl-BLXdCiHRU78QMKpj9q7D36Kb6UIVg"
  ["object"]=>
  string(15) "chat.completion"
  ["created"]=>
  int(1744473866)
  ["model"]=>
  string(17) "gpt-4o-2024-08-06"
  ["choices"]=>
  array(1) {
    [0]=>
    object(stdClass)#33 (4) {
      ["index"]=>
      int(0)
      ["message"]=>
      object(stdClass)#19 (4) {
        ["role"]=>
        string(9) "assistant"
        ["content"]=>
        string(34) "Hello! How can I assist you today?"
        ["refusal"]=>
        NULL
        ["annotations"]=>
        array(0) {
        }
      }
      ["logprobs"]=>
      NULL
      ["finish_reason"]=>
      string(4) "stop"
    }
  }
  ["usage"]=>
  object(stdClass)#29 (5) {
    ["prompt_tokens"]=>
    int(11)
    ["completion_tokens"]=>
    int(10)
    ["total_tokens"]=>
    int(21)
    ["prompt_tokens_details"]=>
    object(stdClass)#22 (2) {
      ["cached_tokens"]=>
      int(0)
      ["audio_tokens"]=>
      int(0)
    }
    ["completion_tokens_details"]=>
    object(stdClass)#20 (4) {
      ["reasoning_tokens"]=>
      int(0)
      ["audio_tokens"]=>
      int(0)
      ["accepted_prediction_tokens"]=>
      int(0)
      ["rejected_prediction_tokens"]=>
      int(0)
    }
  }
  ["service_tier"]=>
  string(7) "default"
  ["system_fingerprint"]=>
  string(13) "fp_b7faba9ef5"

Images

Create image

This example asks for two images of a jungle waterfall.

<?php

require __DIR__ . '/vendor/autoload.php';

// @see https://platform.openai.com/docs/api-reference/authentication
$api_key = getenv('OPENAI_API_KEY');

$api = new OpenAI\OpenAIImages($api_key);

$response = $api->createAsURL('A jungle waterfall', 2, '256x256');

var_dump($response);

The response

Create image variation

<?php

require __DIR__ . '/vendor/autoload.php';

// @see https://platform.openai.com/docs/api-reference/authentication
$api_key = getenv('OPENAI_API_KEY');

$api = new OpenAI\OpenAIImages($api_key);

$response = $api->createVariationAsURL('variation.png', 1, '256x256');

var_dump($response);

The source and response

Create image edit

<?php

require __DIR__ . '/vendor/autoload.php';

// @see https://platform.openai.com/docs/api-reference/authentication
$api_key = getenv('OPENAI_API_KEY');

$api = new OpenAI\OpenAIImages($api_key);

$response = $api->createEditAsURL('edit.png', 'a duck driving a car', 'edit_mask.png', 1, '256x256');

var_dump($response);

The source and response

Fine-tuning a model

First, take a look at the official documentation on fine-tuning.

Create and upload your training file. Example:

training.jsonl

{"messages": [{"role": "system", "content": "A chatbot that knows the difference between animals and rocks."}, {"role": "user", "content": "What is a dog?"}, {"role": "assistant", "content": "A dog is an animal."}]}
{"messages": [{"role": "system", "content": "A chatbot that knows the difference between animals and rocks."}, {"role": "user", "content": "What is a cat?"}, {"role": "assistant", "content": "A cat is an animal."}]}
{"messages": [{"role": "system", "content": "A chatbot that knows the difference between animals and rocks."}, {"role": "user", "content": "What is a bird?"}, {"role": "assistant", "content": "A bird is an animal."}]}
{"messages": [{"role": "system", "content": "A chatbot that knows the difference between animals and rocks."}, {"role": "user", "content": "What is a dolphin?"}, {"role": "assistant", "content": "A dolphin is an animal."}]}
{"messages": [{"role": "system", "content": "A chatbot that knows the difference between animals and rocks."}, {"role": "user", "content": "What is a lion?"}, {"role": "assistant", "content": "A lion is an animal."}]}
{"messages": [{"role": "system", "content": "A chatbot that knows the difference between animals and rocks."}, {"role": "user", "content": "What is limestone?"}, {"role": "assistant", "content": "Limestone is a rock."}]}
{"messages": [{"role": "system", "content": "A chatbot that knows the difference between animals and rocks."}, {"role": "user", "content": "What is marble?"}, {"role": "assistant", "content": "Marble is a rock."}]}
{"messages": [{"role": "system", "content": "A chatbot that knows the difference between animals and rocks."}, {"role": "user", "content": "What is granite?"}, {"role": "assistant", "content": "Granite is a rock."}]}
{"messages": [{"role": "system", "content": "A chatbot that knows the difference between animals and rocks."}, {"role": "user", "content": "What is shale?"}, {"role": "assistant", "content": "Shale is a rock."}]}
{"messages": [{"role": "system", "content": "A chatbot that knows the difference between animals and rocks."}, {"role": "user", "content": "What is basalt?"}, {"role": "assistant", "content": "Basalt is a rock."}]}

<?php

require __DIR__ . '/vendor/autoload.php';

// @see https://platform.openai.com/docs/api-reference/authentication
$api_key = getenv('OPENAI_API_KEY');

$api = new OpenAI\OpenAIFiles($api_key);

$file = $api->uploadFile('training.jsonl');

List your files to get the file ID:

$api = new OpenAI\OpenAIFiles($api_key);

$files = $api->getFiles();

var_dump($files);

Create a new training job using the file ID you've obtained:

$api = new OpenAI\OpenAIFineTunes($api_key);

$api->create('gpt-4o-mini-2024-07-18', 'file-mQy...');

Fine-tuning can take a while. Check the Fine-tuning UI for updates.

List models to get the name of your new fine-tuned model:

$api = new OpenAI\OpenAIModels($api_key);

$models = $api->getModels();

var_dump($models);

The model name will look something like this: ft:gpt-4o-mini-2024-07-18:personal::ABCABC

Create a new completion using your fine-tuned model.

$api = new OpenAI\OpenAICompletions($api_key);

$messages = [
    (object) ['role' => 'user', 'content' => 'What is a dog?'],
];

$response = $api->create('ft:gpt-4o-mini-2024-07-18:personal::ABCABC', $messages);

var_dump($response);

The response should be "A dog is an animal."

$api = new OpenAI\OpenAICompletions($api_key);

$messages = [
    (object) ['role' => 'user', 'content' => 'What is marble?'],
];

$response = $api->create('ft:gpt-4o-mini-2024-07-18:personal::ABCABC', $messages);

var_dump($response);

The response should be "Marble is a rock."

Available functions

Models

Completions

Chat

Edits

Images

Embeddings

Audio

Files

Fine-Tunes

Moderations

License

MIT