1936: VLLM client

CHANGELOG.md

@@ -8,6 +8,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+- VLLM client and plugin to support VLLM communication.
+
 ## [0.0.7] - 2024-04-07
 
 - Parse stream data from LLM correctly in relation to newlines

src/Client/Client.php

@@ -0,0 +1,139 @@
+<?php
+
+namespace Drupal\llm_services\Client;
+
+use Drupal\llm_services\Exceptions\CommunicationException;
+use Drupal\llm_services\Model\Payload;
+use GuzzleHttp\ClientInterface;
+use GuzzleHttp\Exception\GuzzleException;
+use GuzzleHttp\RequestOptions;
+use Psr\Http\Message\ResponseInterface;
+
+/**
+ * Base client to communicate with a LLM framework.
+ */
+abstract class Client {
+
+  /**
+   * Cache for stream parsing.
+   *
+   * @var string
+   *
+   * @see parse()
+   */
+  protected string $parserCache = '';
+
+  /**
+   * Default constructor.
+   *
+   * @param string $url
+   *   The URL to the LLM serving framework.
+   * @param int $port
+   *   The port that the framework is using.
+   * @param \GuzzleHttp\ClientInterface $client
+   *   The http client used to interact with the framework.
+   * @param string $username
+   *   Basic auth username (default: empty string).
+   * @param string $password
+   *   Basic auth password (default: empty string).
+   * @param int $connectTimeout
+   *   The timeout for connoting to the API.
+   * @param int $waitTimeout
+   *   The timeout for wait on content from the API.
+   */
+  public function __construct(
+    protected string $url,
+    protected int $port,
+    protected ClientInterface $client,
+    protected string $username = '',
+    protected string $password = '',
+    protected int $connectTimeout = 10,
+    protected int $waitTimeout = 300,
+  ) {
+  }
+
+  /**
+   * Take all payload messages and change them into an array.
+   *
+   * This array of messages is used to give the model some chat context to make
+   * the interaction appear more like real char with a person.
+   *
+   * @param \Drupal\llm_services\Model\Payload $payload
+   *   The payload sent to the chat function.
+   *
+   * @return array{content: string, role: string}[]
+   *   Array of messages to send to the model.
+   */
+  protected function chatMessagesAsArray(Payload $payload): array {
+    $messages = [];
+    foreach ($payload->getMessages() as $message) {
+      $messages[] = [
+        'content' => $message->content,
+        'role' => $message->role->value,
+      ];
+    }
+
+    return $messages;
+  }
+
+  /**
+   * Make request to LLM framework.
+   *
+   * @param string $method
+   *   The method to use (GET/POST).
+   * @param string $uri
+   *   The API endpoint to call.
+   * @param array<string, mixed> $options
+   *   Extra options and/or payload to post.
+   *
+   * @return \Psr\Http\Message\ResponseInterface
+   *   The response object.
+   *
+   * @throws \Drupal\llm_services\Exceptions\CommunicationException
+   */
+  protected function call(string $method, string $uri, array $options = []): ResponseInterface {
+    try {
+      // Add basic auth if given.
+      if (!empty($this->username)) {
+        $auth = 'Basic ' . base64_encode($this->username . ':' . $this->password);
+        if (isset($options['headers'])) {
+          $options['headers']['Authorization'] = $auth;
+        }
+        else {
+          $options['headers'] = ['Authorization' => $auth];
+        }
+      }
+
+      // Add default configuration options (shared between clients).
+      $options = array_merge([
+        RequestOptions::CONNECT_TIMEOUT => $this->connectTimeout,
+        RequestOptions::TIMEOUT => $this->waitTimeout,
+        RequestOptions::STREAM => TRUE,
+      ], $options);
+
+      $response = $this->client->request($method, $this->getUrl($uri), $options);
+      if ($response->getStatusCode() !== 200) {
+        throw new CommunicationException('Request failed', $response->getStatusCode());
+      }
+    }
+    catch (GuzzleException $exception) {
+      throw new CommunicationException('Request failed: ' . $exception->getMessage(), $exception->getCode(), $exception);
+    }
+
+    return $response;
+  }
+
+  /**
+   * Returns a URL string with the given URI appended to the base URL.
+   *
+   * @param string $uri
+   *   The URI to append to the base URL. Default is an empty string.
+   *
+   * @return string
+   *   The complete URL string.
+   */
+  protected function getUrl(string $uri = ''): string {
+    return $this->url . ':' . $this->port . ($uri ? '/' . ltrim($uri, '/') : '');
+  }
+
+}

src/Client/Ollama.php

@@ -2,49 +2,12 @@
 
 namespace Drupal\llm_services\Client;
 
-use Drupal\llm_services\Exceptions\CommunicationException;
 use Drupal\llm_services\Model\Payload;
-use GuzzleHttp\ClientInterface;
-use GuzzleHttp\Exception\GuzzleException;
-use GuzzleHttp\RequestOptions;
-use Psr\Http\Message\ResponseInterface;
 
 /**
  * Client to communicate with Ollama.
  */
-class Ollama {
-
-  /**
-   * Cache for stream parsing.
-   *
-   * @var string
-   *
-   * @see parse()
-   */
-  private string $parserCache = '';
-
-  /**
-   * Default constructor.
-   *
-   * @param string $url
-   *   The URL of the Ollama server.
-   * @param int $port
-   *   The port that Ollama is listening at.
-   * @param \GuzzleHttp\ClientInterface $client
-   *   The http client used to interact with ollama.
-   * @param string $username
-   *   Basic auth username (default: empty string).
-   * @param string $password
-   *   Basic auth password (default: empty string).
-   */
-  public function __construct(
-    private readonly string $url,
-    private readonly int $port,
-    private readonly ClientInterface $client,
-    private readonly string $username = '',
-    private readonly string $password = '',
-  ) {
-  }
+class Ollama extends Client {
 
   /**
    * List all models currently installed in Ollama.
@@ -95,9 +58,6 @@ public function install(string $modelName): \Generator {
       'headers' => [
         'Content-Type' => 'application/json',
       ],
-      RequestOptions::CONNECT_TIMEOUT => 10,
-      RequestOptions::TIMEOUT => 300,
-      RequestOptions::STREAM => TRUE,
     ]);
 
     $body = $response->getBody();
@@ -132,9 +92,6 @@ public function completion(Payload $payload): \Generator {
       'headers' => [
         'Content-Type' => 'application/json',
       ],
-      RequestOptions::CONNECT_TIMEOUT => 10,
-      RequestOptions::TIMEOUT => 300,
-      RequestOptions::STREAM => TRUE,
     ]);
 
     $body = $response->getBody();
@@ -167,9 +124,6 @@ public function chat(Payload $payload): \Generator {
       'headers' => [
         'Content-Type' => 'application/json',
       ],
-      RequestOptions::CONNECT_TIMEOUT => 10,
-      RequestOptions::TIMEOUT => 300,
-      RequestOptions::STREAM => TRUE,
     ]);
 
     $body = $response->getBody();
@@ -179,32 +133,6 @@ public function chat(Payload $payload): \Generator {
     }
   }
 
-  /**
-   * Take all payload messages and change them into an array.
-   *
-   * This array of messages is used to give the model some chat context to make
-   * the interaction appear more like real char with a person.
-   *
-   * @param \Drupal\llm_services\Model\Payload $payload
-   *   The payload sent to the chat function.
-   *
-   * @return array{content: string, role: string}[]
-   *   Array of messages to send to Ollama.
-   *
-   * @see https://github.com/ollama/ollama/blob/main/docs/api.md#chat-request-with-history
-   */
-  private function chatMessagesAsArray(Payload $payload): array {
-    $messages = [];
-    foreach ($payload->getMessages() as $message) {
-      $messages[] = [
-        'content' => $message->content,
-        'role' => $message->role->value,
-      ];
-    }
-
-    return $messages;
-  }
-
   /**
    * Parse LLM stream.
    *
@@ -345,56 +273,4 @@ private function parseDataToStrings(string $data): array {
     return $results;
   }
 
-  /**
-   * Make request to Ollama.
-   *
-   * @param string $method
-   *   The method to use (GET/POST).
-   * @param string $uri
-   *   The API endpoint to call.
-   * @param array<string, mixed> $options
-   *   Extra options and/or payload to post.
-   *
-   * @return \Psr\Http\Message\ResponseInterface
-   *   The response object.
-   *
-   * @throws \Drupal\llm_services\Exceptions\CommunicationException
-   */
-  private function call(string $method, string $uri, array $options = []): ResponseInterface {
-    try {
-      // Add basic auth if given.
-      if (!empty($this->username)) {
-        $auth = 'Basic ' . base64_encode($this->username . ':' . $this->password);
-        if (isset($options['headers'])) {
-          $options['headers']['Authorization'] = $auth;
-        }
-        else {
-          $options['headers'] = ['Authorization' => $auth];
-        }
-      }
-      $response = $this->client->request($method, $this->getUrl($uri), $options);
-      if ($response->getStatusCode() !== 200) {
-        throw new CommunicationException('Request failed', $response->getStatusCode());
-      }
-    }
-    catch (GuzzleException $exception) {
-      throw new CommunicationException('Request failed: ' . $exception->getMessage(), $exception->getCode(), $exception);
-    }
-
-    return $response;
-  }
-
-  /**
-   * Returns a URL string with the given URI appended to the base URL.
-   *
-   * @param string $uri
-   *   The URI to append to the base URL. Default is an empty string.
-   *
-   * @return string
-   *   The complete URL string.
-   */
-  private function getUrl(string $uri = ''): string {
-    return $this->url . ':' . $this->port . ($uri ? '/' . ltrim($uri, '/') : '');
-  }
-
 }

src/Client/OllamaChatResponse.php

@@ -62,7 +62,13 @@ public function getRole(): MessageRoles {
   }
 
   /**
-   * {@inheritdoc}
+   * Get images.
+   *
+   * @todo Not sure yet but this must be an base64 encode image is on is sendt
+   *       to the LLMs
+   *
+   * @return array<string>
+   *   String of base64 encoded images.
    */
   public function getImages(): array {
     return $this->images;