sabatinomasala/replicate-php

PHP client for the Replicate API

1.2.0 2024-08-09 15:11 UTC

Requires

• php: ^8.1
• saloonphp/saloon: ^2.5

Requires (Dev)

• laravel/pint: ^1.2
• spatie/ray: ^1.28

MIT

• SabatinoMasala

• machine
• learning
• artificial
• intelligence
• ai
• replicate
• replicate-php
• SabatinoMasala

README

This is a PHP client for Replicate. It lets you run models from your PHP code and do various other things on Replicate.

Credits

This package is based on the work of https://github.com/sawirricardo/replicate-php

Video

If you're a more visual learner, you can check out this video this package is showcased:

Installation

You can install the package via composer:

composer require sabatinomasala/replicate-php

Getting an API token

You can get an API token from https://replicate.com/account/api-tokens

Creating a prediction

You can create a prediction as follows:

// Create a prediction
$prediction = $client->predictions()->create('stability-ai/sdxl:7762fd07cf82c948538e41f63f77d685e02b063e37e496e96eefd46c929f9bdc', [
    'prompt' => 'a cat wearing a cowboy hat',
]);

$id = $prediction->json('id');

// Fetch prediction
$prediction = $client->predictions()->get($id);
dd($prediction->json())

Running a model

You can run a model and wait on the output as follows:

$token = env('REPLICATE_TOKEN');
$client = new SabatinoMasala\Replicate\Replicate($token);
$output = $client->run('stability-ai/sdxl:7762fd07cf82c948538e41f63f77d685e02b063e37e496e96eefd46c929f9bdc', [
    'prompt' => 'a cat wearing a cowboy hat',
], function($prediction) {
    // You can log the current state of the prediction
    \Log::info('Progress', $prediction->json());
});

dd($output[0]);

Chaining multiple models

Because 'run' returns the output of the model, you can chain multiple models together like this:

$output = $client->run('stability-ai/sdxl:7762fd07cf82c948538e41f63f77d685e02b063e37e496e96eefd46c929f9bdc', [
    'prompt' => 'a cat wearing a cowboy hat',
]);

$caption = $client->run('salesforce/blip:2e1dddc8621f72155f24cf2e0adbde548458d3cab9f00c0139eea840d0ac4746', [
    'task' => 'image_captioning',
    'image' => $output[0],
]);

dd($caption); // Caption: a drawing of a cat wearing a cowboy hat

Usage with official models

Offical models, like Llama, have a different API endpoint. This client automatically switches endpoints, based on the input you give:

// This will do an API call to https://api.replicate.com/v1/models/meta/meta-llama-3-70b-instruct/predictions instead of https://api.replicate.com/v1/predictions
$output = $replicate->run('meta/meta-llama-3-70b-instruct', [
    'prompt' => 'I want to travel to Japan, give me an itinerary',
    'max_tokens' => 1000,
], function($prediction) {
    \Log::info($prediction->json('output'));
});

Credits

• Original creator of replicate-php: sawirricardo
• sabatinomasala

License

The MIT License (MIT). Please see License File for more information.