编程

ChatGPT 的 PHP 接口 - OpenAI PHP 客户端

3621 2023-04-05 15:37:00

OpenAI PHP 是一个社区维护的 PHP API 客户端,让你可以和 Open AI API 进行交互。如果您的业务依赖于此包,可以对开发该包的开发者表示支持:

开始

PHP 8.1+ 以上版本

首先,通过 Composer 安装 OpenAI:

composer require openai-php/client

注, Laravel 版为 openai-php/laravel

如果你的项目还没有集成 PSR-18 客户端,请确保 php-http/discovery 的 composer 插件能够运行或手动安装。

composer require guzzlehttp/guzzle

然后可以和 OpenAI 的 API 交互:

$yourApiKey = getenv('YOUR_API_KEY');
$client = OpenAI::client($yourApiKey);

$result = $client->completions()->create([
    'model' => 'text-davinci-003',
    'prompt' => 'PHP is',
]);

echo $result['choices'][0]['text']; // an open-source, widely-used, server-side scripting language.

如必要,也可以配置和创建其他的客户端

$yourApiKey = getenv('YOUR_API_KEY');

$client = OpenAI::factory()
    ->withApiKey($yourApiKey)
    ->withOrganization('your-organization') // default: null
    ->withBaseUri('openai.example.com/v1') // default: api.openai.com/v1
    ->withHttpClient($client = new \GuzzleHttp\Client([])) // default: HTTP client found using PSR-18 HTTP Client Discovery
    ->withHttpHeader('X-My-Header', 'foo')
    ->withQueryParam('my-param', 'bar')
    ->withStreamHandler(fn (RequestInterface $request): ResponseInterface => $client->send($request, [
        'stream' => true // Allows to provide a custom stream handler for the http client.
    ]))
    ->make();

使用

模型资源

list

列出当前可用模型,提供每个模型的基础信息,比如owner和availability。

$response = $client->models()->list();

$response->object; // 'list'

foreach ($response->data as $result) {
    $result->id; // 'text-davinci-003'
    $result->object; // 'model'
    // ...
}

$response->toArray(); // ['object' => 'list', 'data' => [...]]

retrieve

检索模型实例,提供该模型的基础信息,如owner和权限。

$response = $client->models()->retrieve('text-davinci-003');

$response->id; // 'text-davinci-003'
$response->object; // 'model'
$response->created; // 1642018370
$response->ownedBy; // 'openai'
$response->root; // 'text-davinci-003'
$response->parent; // null

foreach ($response->permission as $result) {
    $result->id; // 'modelperm-7E53j9OtnMZggjqlwMxW4QG7' 
    $result->object; // 'model_permission' 
    $result->created; // 1664307523 
    $result->allowCreateEngine; // false 
    $result->allowSampling; // true 
    $result->allowLogprobs; // true 
    $result->allowSearchIndices; // false 
    $result->allowView; // true 
    $result->allowFineTuning; // false 
    $result->organization; // '*' 
    $result->group; // null 
    $result->isBlocking; // false 
}

$response->toArray(); // ['id' => 'text-davinci-003', ...]

delete

.删除一个精调模型

$response = $client->models()->delete('curie:ft-acmeco-2021-03-03-21-44-20');

$response->id; // 'curie:ft-acmeco-2021-03-03-21-44-20'
$response->object; // 'model'
$response->deleted; // true

$response->toArray(); // ['id' => 'curie:ft-acmeco-2021-03-03-21-44-20', ...]

Completions 资源

create

为提供的提示或者参数创建一个completion。

$response = $client->completions()->create([
    'model' => 'text-davinci-003',
    'prompt' => 'Say this is a test',
    'max_tokens' => 6,
    'temperature' => 0
]);

$response->id; // 'cmpl-uqkvlQyYK7bGYrRHQ0eXlWi7'
$response->object; // 'text_completion'
$response->created; // 1589478378
$response->model; // 'text-davinci-003'

foreach ($response->choices as $result) {
    $result->text; // '\n\nThis is a test'
    $result->index; // 0
    $result->logprobs; // null
    $result->finishReason; // 'length' or null
}

$response->usage->promptTokens; // 5,
$response->usage->completionTokens; // 6,
$response->usage->totalTokens; // 11

$response->toArray(); // ['id' => 'cmpl-uqkvlQyYK7bGYrRHQ0eXlWi7', ...]

create streamed

为提供的提示和参数创建一个流式补全(streamed completion)。

$stream = $client->completions()->createStreamed([
        'model' => 'text-davinci-003',
        'prompt' => 'Hi',
        'max_tokens' => 10,
    ]);

foreach($stream as $response){
    $response->choices[0]->text;
}
// 1. iteration => 'I'
// 2. iteration => ' am'
// 3. iteration => ' very'
// 4. iteration => ' excited'
// ...

Chat 资源

create

为 chat 信息创建一个 completion。

$response = $client->chat()->create([
    'model' => 'gpt-3.5-turbo',
    'messages' => [
        ['role' => 'user', 'content' => 'Hello!'],
    ],
]);

$response->id; // 'chatcmpl-6pMyfj1HF4QXnfvjtfzvufZSQq6Eq'
$response->object; // 'chat.completion'
$response->created; // 1677701073
$response->model; // 'gpt-3.5-turbo-0301'

foreach ($response->choices as $result) {
    $result->index; // 0
    $result->message->role; // 'assistant'
    $result->message->content; // '\n\nHello there! How can I assist you today?'
    $result->finishReason; // 'stop'
}

$response->usage->promptTokens; // 9,
$response->usage->completionTokens; // 12,
$response->usage->totalTokens; // 21

$response->toArray(); // ['id' => 'chatcmpl-6pMyfj1HF4QXnfvjtfzvufZSQq6Eq', ...]

created streamed

为 chat 信息创建一个流式补全(streamed completion)。

$stream = $client->chat()->createStreamed([
    'model' => 'gpt-4',
    'messages' => [
        ['role' => 'user', 'content' => 'Hello!'],
    ],
]);

foreach($stream as $response){
    $response->choices[0]->toArray();
}
// 1. iteration => ['index' => 0, 'delta' => ['role' => 'assistant'], 'finish_reason' => null]
// 2. iteration => ['index' => 0, 'delta' => ['content' => 'Hello'], 'finish_reason' => null]
// 3. iteration => ['index' => 0, 'delta' => ['content' => '!'], 'finish_reason' => null]
// ...

audio 资源

transcribe

将音频(audio)资源用指定的语言记录。

$response = $client->audio()->transcribe([
    'model' => 'whisper-1',
    'file' => fopen('audio.mp3', 'r'),
    'response_format' => 'verbose_json',
]);

$response->task; // 'transcribe'
$response->language; // 'english'
$response->duration; // 2.95
$response->text; // 'Hello, how are you?'

foreach ($response->segments as $segment) {
    $segment->index; // 0
    $segment->seek; // 0
    $segment->start; // 0.0
    $segment->end; // 4.0
    $segment->text; // 'Hello, how are you?'
    $segment->tokens; // [50364, 2425, 11, 577, 366, 291, 30, 50564]
    $segment->temperature; // 0.0
    $segment->avgLogprob; // -0.45045216878255206
    $segment->compressionRatio; // 0.7037037037037037
    $segment->noSpeechProb; // 0.1076972484588623
    $segment->transient; // false
}

$response->toArray(); // ['task' => 'transcribe', ...]

translate

将音频翻译成英语。

$response = $client->audio()->translate([
    'model' => 'whisper-1',
    'file' => fopen('german.mp3', 'r'),
    'response_format' => 'verbose_json',
]);

$response->task; // 'translate'
$response->language; // 'english'
$response->duration; // 2.95
$response->text; // 'Hello, how are you?'

foreach ($response->segments as $segment) {
    $segment->index; // 0
    $segment->seek; // 0
    $segment->start; // 0.0
    $segment->end; // 4.0
    $segment->text; // 'Hello, how are you?'
    $segment->tokens; // [50364, 2425, 11, 577, 366, 291, 30, 50564]
    $segment->temperature; // 0.0
    $segment->avgLogprob; // -0.45045216878255206
    $segment->compressionRatio; // 0.7037037037037037
    $segment->noSpeechProb; // 0.1076972484588623
    $segment->transient; // false
}

$response->toArray(); // ['task' => 'translate', ...]

Edits 资源

create

为提供的输入(input)、说明(instruction)和参数创建一个新的编辑。

$response = $client->edits()->create([
    'model' => 'text-davinci-edit-001',
    'input' => 'What day of the wek is it?',
    'instruction' => 'Fix the spelling mistakes',
]);

$response->object; // 'edit'
$response->created; // 1589478378

foreach ($response->choices as $result) {
    $result->text; // 'What day of the week is it?'
    $result->index; // 0
}

$response->usage->promptTokens; // 25,
$response->usage->completionTokens; // 32,
$response->usage->totalTokens; // 57

$response->toArray(); // ['object' => 'edit', ...]

Embeddings 资源

create

创建一个 embedding 容器,以表示输入文本。

$response = $client->embeddings()->create([
    'model' => 'text-similarity-babbage-001',
    'input' => 'The food was delicious and the waiter...',
]);

$response->object; // 'list'

foreach ($response->embeddings as $embedding) {
    $embedding->object; // 'embedding'
    $embedding->embedding; // [0.018990106880664825, -0.0073809814639389515, ...]
    $embedding->index; // 0
}

$response->usage->promptTokens; // 8,
$response->usage->totalTokens; // 8

$response->toArray(); // ['data' => [...], ...]

文件资源

list

.返回用户组织的文件列表。

$response = $client->files()->list();

$response->object; // 'list'

foreach ($response->data as $result) {
    $result->id; // 'file-XjGxS3KTG0uNmNOK362iJua3'
    $result->object; // 'file'
    // ...
}

$response->toArray(); // ['object' => 'list', 'data' => [...]]

delete

删除一个文件。

$response = $client->files()->delete($file);

$response->id; // 'file-XjGxS3KTG0uNmNOK362iJua3'
$response->object; // 'file'
$response->deleted; // true

$response->toArray(); // ['id' => 'file-XjGxS3KTG0uNmNOK362iJua3', ...]

retrieve

返回一个指定文件的信息。

$response = $client->files()->retrieve('file-XjGxS3KTG0uNmNOK362iJua3');

$response->id; // 'file-XjGxS3KTG0uNmNOK362iJua3'
$response->object; // 'file'
$response->bytes; // 140
$response->createdAt; // 1613779657
$response->filename; // 'mydata.jsonl'
$response->purpose; // 'fine-tune'
$response->status; // 'succeeded'
$response->status_details; // null

$response->toArray(); // ['id' => 'file-XjGxS3KTG0uNmNOK362iJua3', ...]

upload

上传一个文件,该文件包含被用于各种 endpoints/features 的文档。

$response = $client->files()->upload([
        'purpose' => 'fine-tune',
        'file' => fopen('my-file.jsonl', 'r'),
    ]);

$response->id; // 'file-XjGxS3KTG0uNmNOK362iJua3'
$response->object; // 'file'
$response->bytes; // 140
$response->createdAt; // 1613779657
$response->filename; // 'mydata.jsonl'
$response->purpose; // 'fine-tune'
$response->status; // 'succeeded'
$response->status_details; // null

$response->toArray(); // ['id' => 'file-XjGxS3KTG0uNmNOK362iJua3', ...]

download

返回指定文件的内容。

$client->files()->download($file); // '{"prompt": "<prompt text>", ...'

FineTunes资源

create

创建一个作业,从给定的数据集中微调指定的模型。

$response = $client->fineTunes()->create([
    'training_file' => 'file-ajSREls59WBbvgSzJSVWxMCB',
    'validation_file' => 'file-XjSREls59WBbvgSzJSVWxMCa',
    'model' => 'curie',
    'n_epochs' => 4,
    'batch_size' => null,
    'learning_rate_multiplier' => null,
    'prompt_loss_weight' => 0.01,
    'compute_classification_metrics' => false,
    'classification_n_classes' => null,
    'classification_positive_class' => null,
    'classification_betas' => [],
    'suffix' => null,
]);

$response->id; // 'ft-AF1WoRqd3aJAHsqc9NY7iL8F'
$response->object; // 'fine-tune'
// ...

$response->toArray(); // ['id' => 'ft-AF1WoRqd3aJAHsqc9NY7iL8F', ...]

list

列出你组织中的微调(fine-tuning)作业。

$response = $client->fineTunes()->list();

$response->object; // 'list'

foreach ($response->data as $result) {
    $result->id; // 'ft-AF1WoRqd3aJAHsqc9NY7iL8F'
    $result->object; // 'fine-tune'
    // ...
}

$response->toArray(); // ['object' => 'list', 'data' => [...]]

retrieve

获取该微调作业的信息。

$response = $client->fineTunes()->retrieve('ft-AF1WoRqd3aJAHsqc9NY7iL8F');

$response->id; // 'ft-AF1WoRqd3aJAHsqc9NY7iL8F'
$response->object; // 'fine-tune'
$response->model; // 'curie'
$response->createdAt; // 1614807352
$response->fineTunedModel; // 'curie => ft-acmeco-2021-03-03-21-44-20'
$response->organizationId; // 'org-jwe45798ASN82s'
$response->resultFiles; // [
$response->status; // 'succeeded'
$response->validationFiles; // [
$response->trainingFiles; // [
$response->updatedAt; // 1614807865

foreach ($response->events as $result) {
    $result->object; // 'fine-tune-event' 
    $result->createdAt; // 1614807352
    $result->level; // 'info'
    $result->message; // 'Job enqueued. Waiting for jobs ahead to complete. Queue number =>  0.'
}

$response->hyperparams->batchSize; // 4 
$response->hyperparams->learningRateMultiplier; // 0.1 
$response->hyperparams->nEpochs; // 4 
$response->hyperparams->promptLossWeight; // 0.1

foreach ($response->resultFiles as $result) {
    $result->id; // 'file-XjGxS3KTG0uNmNOK362iJua3'
    $result->object; // 'file'
    $result->bytes; // 140
    $result->createdAt; // 1613779657
    $result->filename; // 'mydata.jsonl'
    $result->purpose; // 'fine-tune'
    $result->status; // 'succeeded'
    $result->status_details; // null
}

foreach ($response->validationFiles as $result) {
    $result->id; // 'file-XjGxS3KTG0uNmNOK362iJua3'
    // ...
}

foreach ($response->trainingFiles as $result) {
    $result->id; // 'file-XjGxS3KTG0uNmNOK362iJua3'
    // ...
}

$response->toArray(); // ['id' => 'ft-AF1WoRqd3aJAHsqc9NY7iL8F', ...]

cancel

立刻取消一个微调作业。

$response = $client->fineTunes()->cancel('ft-AF1WoRqd3aJAHsqc9NY7iL8F');

$response->id; // 'ft-AF1WoRqd3aJAHsqc9NY7iL8F'
$response->object; // 'fine-tune'
// ...
$response->status; // 'cancelled'
// ...

$response->toArray(); // ['id' => 'ft-AF1WoRqd3aJAHsqc9NY7iL8F', ...]

list events

Get fine-grained status updates for a fine-tune job.

$response = $client->fineTunes()->listEvents('ft-AF1WoRqd3aJAHsqc9NY7iL8F');

$response->object; // 'list'

foreach ($response->data as $result) {
    $result->object; // 'fine-tune-event' 
    $result->createdAt; // 1614807352
    // ...
}

$response->toArray(); // ['object' => 'list', 'data' => [...]]

list events streamed

Get streamed fine-grained status updates for a fine-tune job.

$stream = $client->fineTunes()->listEventsStreamed('ft-y3OpNlc8B5qBVGCCVsLZsDST');

foreach($stream as $response){
    $response->message;
}
// 1. iteration => 'Created fine-tune: ft-y3OpNlc8B5qBVGCCVsLZsDST'
// 2. iteration => 'Fine-tune costs $0.00'
// ...
// xx. iteration => 'Uploaded result file: file-ajLKUCMsFPrT633zqwr0eI4l'
// xx. iteration => 'Fine-tune succeeded'

Moderations 资源

create

如果文本违反OpenAI的内容策略,则进行分类。

$response = $client->moderations()->create([
    'model' => 'text-moderation-latest',
    'input' => 'I want to k*** them.',
]);

$response->id; // modr-5xOyuS
$response->model; // text-moderation-003

foreach ($response->results as $result) {
    $result->flagged; // true

    foreach ($result->categories as $category) {
        $category->category->value; // 'violence'
        $category->violated; // true
        $category->score; // 0.97431367635727
    }
}

$response->toArray(); // ['id' => 'modr-5xOyuS', ...]

Images 模型

create

通过给定的提示信息创建图像

$response = $client->images()->create([
    'prompt' => 'A cute baby sea otter',
    'n' => 1,
    'size' => '256x256',
    'response_format' => 'url',
]);

$response->created; // 1589478378

foreach ($response->data as $data) {
    $data->url; // 'https://oaidalleapiprodscus.blob.core.windows.net/private/...'
    $data->b64_json; // null
}

$response->toArray(); // ['created' => 1589478378, data => ['url' => 'https://oaidalleapiprodscus...', ...]]

edit

通过提供的原始图片和信息创建一个图片。

$response = $client->images()->edit([
    'image' => fopen('image_edit_original.png', 'r'),
    'mask' => fopen('image_edit_mask.png', 'r'),
    'prompt' => 'A sunlit indoor lounge area with a pool containing a flamingo',
    'n' => 1,
    'size' => '256x256',
    'response_format' => 'url',
]);

$response->created; // 1589478378

foreach ($response->data as $data) {
    $data->url; // 'https://oaidalleapiprodscus.blob.core.windows.net/private/...'
    $data->b64_json; // null
}

$response->toArray(); // ['created' => 1589478378, data => ['url' => 'https://oaidalleapiprodscus...', ...]]

variation

为给定的图片创建一个变体。

$response = $client->images()->variation([
    'image' => fopen('image_edit_original.png', 'r'),
    'n' => 1,
    'size' => '256x256',
    'response_format' => 'url',
]);

$response->created; // 1589478378

foreach ($response->data as $data) {
    $data->url; // 'https://oaidalleapiprodscus.blob.core.windows.net/private/...'
    $data->b64_json; // null
}

$response->toArray(); // ['created' => 1589478378, data => ['url' => 'https://oaidalleapiprodscus...', ...]]

测试

该包提供了 OpenAI\Client 类的伪实现,允许您伪造 API 响应。

要测试代码,请确保在测试用例中将 OpenAI\Client 类与 OpenAI\Testing\ClientFake 类交换。

伪造的响应将按照创建伪造客户端时提供的顺序返回。

所有响应都有一个 fake() 方法,该方法只提供与测试用例相关的参数,就可以轻松地创建响应对象。

use OpenAI\Testing\ClientFake;
use OpenAI\Responses\Completions\CreateResponse;

$client = new ClientFake([
    CreateResponse::fake([
        'choices' => [
            [
                'text' => 'awesome!',
            ],
        ],
    ]),
]);

$completion = $client->completions()->create([
    'model' => 'text-davinci-003',
    'prompt' => 'PHP is ',
]);

expect($completion['choices'][0]['text'])->toBe('awesome!');

在流式响应的情况下,你可以选择提供一个包含伪响应数据的资源。

use OpenAI\Testing\ClientFake;
use OpenAI\Responses\Chat\CreateStreamedResponse;

$client = new ClientFake([
    CreateStreamedResponse::fake(fopen('file.txt', 'r'););
]);

$completion = $client->chat()->createStreamed([
        'model' => 'gpt-3.5-turbo',
        'messages' => [
            ['role' => 'user', 'content' => 'Hello!'],
        ],
]);

expect($response->getIterator()->current())
        ->id->toBe('chatcmpl-6yo21W6LVo8Tw2yBf7aGf2g17IeIl');

在发送请求之后,有各种方法可以确保发送预期的请求:

// assert completion create request was sent
$client->assertSent(Completions::class, function (string $method, array $parameters): bool {
    return $method === 'create' &&
        $parameters['model'] === 'text-davinci-003' &&
        $parameters['prompt'] === 'PHP is ';
});
// or
$client->completions()->assertSent(function (string $method, array $parameters): bool {
    // ...
});

// assert 2 completion create requests were sent
$client->assertSent(Completions::class, 2);

// assert no completion create requests were sent
$client->assertNotSent(Completions::class);
// or
$client->completions()->assertNotSent();

// assert no requests were sent
$client->assertNothingSent();

要编写预期 API 请求失败的测试,可以提供一个 Throwable 对象作为响应。

$client = new ClientFake([
    new \OpenAI\Exceptions\ErrorException([
        'message' => 'The model `gpt-1` does not exist',
        'type' => 'invalid_request_error',
        'code' => null,
    ])
]);

// the `ErrorException` will be thrown
$completion = $client->completions()->create([
    'model' => 'text-davinci-003',
    'prompt' => 'PHP is ',
]);

开源许可 MIT license.

Laravel 版: openai-php/laravel