A Mini PHP Example: Attention as Weights

A toy weighted sum of context words

We are not implementing a real transformer in PHP – that would be too heavy for a mini example. But the idea of attention can be shown as a weighted sum: neighboring words receive different weights, and the final vector is assembled from their values.

Example of use


                
<?php

$context = [
    'API' => 0.6,
    'authentication' => 0.9,
    'for' => 0.1,
];

$values = [
    'API' => [1.0, 0.2],
    'authentication' => [0.9, 0.8],
    'for' => [0.1, 0.0],
];

$result = [0.0, 0.0];

foreach ($context as $word => $weight) {
    $result[0] += $weight * $values[$word][0];
    $result[1] += $weight * $values[$word][1];
}

print_r($result);

Example:


            Context weights:
API => 0.6
authentication => 0.9
for => 0.1

Word values:
API => [1, 0.2]
authentication => [0.9, 0.8]
for => [0.1, 0]

Result: Memory: 0.002 Mb Time running: < 0.001 sec.


                Array
(
    [0] => 1.42
    [1] => 0.84
)