LLaMa.CPP

¶Health Check

The Health API allows you to verify whether the Llama.cpp server is reachable and operational. This is particularly useful for monitoring or troubleshooting purposes, ensuring that your application can interact with the server before making further requests (e.g., completions, embeddings, etc.).

Tip

Incorporate health checks into your deployment pipeline or monitoring system to proactively detect issues with the Llama.cpp server.

¶Example: Health Check

¶PHP Code

use Partitech\PhpMistral\Clients\LlamaCpp\LlamaCppClient;
use Partitech\PhpMistral\MistralClientException;

$llamacppUrl = getenv('LLAMACPP_URL');
$llamacppApiKey = getenv('LLAMACPP_API_KEY');

$client = new LlamaCppClient(apiKey: $llamacppApiKey, url: $llamacppUrl);

try {
    // Perform the health check (returns a boolean)
    $isHealthy = $client->health();
    
    // Display the result
    print_r($isHealthy);
    
} catch (MistralClientException $e) {
    echo $e->getMessage();
}

¶Example Output

1

• 1 (true): The server is healthy and responding correctly.
• (false): The server is unreachable or unhealthy.

Warning

A false result indicates that the server is either not running, misconfigured, or unreachable (e.g., network issues, incorrect URL or port).

¶Use Cases

• Startup checks: Ensure the Llama.cpp server is running before launching an application.
• Health monitoring: Integrate with monitoring tools (e.g., Prometheus, Grafana) for real-time server status.
• Error handling: Avoid sending inference requests when the server is down, and notify administrators proactively.

¶Error Handling

If the server is unreachable or there is an internal error during the check, the method will throw a MistralClientException. Always wrap health checks in a try-catch block to handle such situations gracefully.

try {
    $isHealthy = $client->health();
} catch (MistralClientException $e) {
    // Log or notify about the server issue
    echo 'Health check failed: ' . $e->getMessage();
}

Note

This health check performs a lightweight status query and does not validate model readiness (e.g., whether a model is loaded). Use this method for basic server availability checks.