Ai Lm Studio Platform Laravel Package

Getting Started

Minimal Steps to First Use

1. Install the Package Add to your composer.json:

   composer require symfony/ai-lm-studio-platform
   

2. Configure LM Studio Provider Update config/packages/ai.yaml:

   framework:
       ai:
           providers:
               lm_studio:
                   platform: lm_studio
                   endpoint: http://localhost:1234/v1/  # Default LM Studio OpenAI endpoint
                   model: "TheBloke/Llama-2-7B-Chat-GGUF"  # Your LM Studio model
                   auth: null  # LM Studio typically doesn't require auth
   

3. First API Call Use Symfony AI’s ChatCompletion service in a controller or command:

   use Symfony\AI\Chat\ChatCompletion;
   use Symfony\AI\Message;
   
   public function __invoke(ChatCompletion $chatCompletion)
   {
       $response = $chatCompletion->create([
           'model' => 'lm_studio',  // Route to LM Studio provider
           'messages' => [
               new Message('user', 'Explain Laravel dependency injection in 3 bullet points.')
           ],
       ]);
   
       return $response->getChoices()[0]->getMessage()->getContent();
   }
   

4. Verify LM Studio is Running Ensure LM Studio is deployed locally (e.g., via Docker) and accessible at the configured endpoint. Test with:

   curl http://localhost:1234/v1/models
   

Where to Look First

• Symfony AI Documentation – Core concepts for providers, messages, and responses.
• LM Studio API Docs – Endpoint specifications (e.g., /chat/completions, /embeddings).
• symfony/ai-lm-studio-platform Tests – Example usage in tests/.

First Use Case: Local AI Prototyping

Replace cloud-based AI calls (e.g., OpenAI) with LM Studio for offline development:

// Before (OpenAI)
$response = $chatCompletion->create(['model' => 'gpt-3.5-turbo', ...]);

// After (LM Studio)
$response = $chatCompletion->create(['model' => 'lm_studio', ...]);

Use Case: Test AI features without API costs or latency during local development.

Implementation Patterns

Core Workflow: Provider Routing

Leverage Symfony AI’s provider abstraction (v0.8.0+) to dynamically route requests:

# config/packages/ai.yaml
framework:
    ai:
        providers:
            lm_studio:
                platform: lm_studio
                endpoint: http://localhost:1234/v1/
                model: "mistral-7b"
            openai:
                platform: openai
                api_key: "%env(OPENAI_API_KEY)%"
                model: "gpt-3.5-turbo"

Dynamic Routing in Code:

$provider = $request->get('provider', 'lm_studio'); // Default to LM Studio
$response = $chatCompletion->create([
    'model' => $provider,
    'messages' => [...],
]);

Integration Tips

1. Environment-Specific Config Use Symfony’s %kernel.environment% to switch providers:

   # config/packages/ai_{env}.yaml
   framework:
       ai:
           providers:
               default:
                   platform: "%env(AI_PROVIDER)%"  # 'lm_studio' or 'openai'
   

2. Model Fallbacks Combine providers for resilience:

   try {
       $response = $chatCompletion->create(['model' => 'lm_studio', ...]);
   } catch (Exception $e) {
       $response = $chatCompletion->create(['model' => 'openai', ...]);
   }
   

3. LM Studio-Specific Features Extend the provider for custom LM Studio endpoints (e.g., non-OpenAI routes):

   // src/AI/LmStudioCustomProvider.php
   use Symfony\AI\Provider\AbstractProvider;
   
   class LmStudioCustomProvider extends AbstractProvider
   {
       public function getCustomEndpoint(): string
       {
           return $this->endpoint . 'custom/route'; // LM Studio-specific
       }
   }
   

   Register in config/services.yaml:

   services:
       Symfony\AI\Provider\LmStudioCustomProvider: ~
   

4. Rate Limiting LM Studio may throttle requests. Use Symfony’s RateLimiter:

   use Symfony\Component\RateLimiter\RateLimiterFactory;
   
   $factory = new RateLimiterFactory();
   $limiter = $factory->create($this->rateLimiter->create('lm_studio', 10, '1 minute'));
   $limiter->consume(); // Enforce limit
   

Common Patterns

Gotchas and Tips

Pitfalls

1. LM Studio API Changes

   • LM Studio’s OpenAI-compatible endpoints may deprecate routes without notice. Monitor LM Studio releases.
   • Fix: Subscribe to LM Studio’s changelog or wrap calls in try-catch blocks.

2. Resource Exhaustion

   • LM Studio consumes significant GPU/CPU. Running large models (e.g., llama-13b) may crash your machine.
   • Fix: Use Docker with resource limits or smaller models (e.g., mistral-7b).

3. No Built-in Retries

   • The package does not retry failed requests to LM Studio. Network issues or OOM errors will propagate.
   • Fix: Implement retry logic with exponential backoff:

     use Symfony\Component\HttpClient\RetryStrategy;
     
     $client = HttpClient::create([
         'timeout' => 30,
         'retry' => RetryStrategy::fromOptions([
             'max_retries' => 3,
             'delay' => 100,
         ]),
     ]);
     

4. Model Name Mismatches

   • LM Studio’s model names (e.g., "TheBloke/Llama-2-7B-Chat-GGUF") do not match OpenAI’s. Misconfiguration will return 404.
   • Fix: Verify model names in LM Studio’s UI or API:

     curl http://localhost:1234/v1/models
     

5. Symfony AI Version Lock

   • The package requires Symfony AI v0.9+. Downgrading breaks provider routing.
   • Fix: Pin Symfony AI in composer.json:

     "symfony/ai": "^0.9"
     

6. No Input Sanitization

   • Raw prompts are sent to LM Studio without escaping. Malicious input (e.g., SQLi) may leak.
   • Fix: Sanitize prompts before sending:

     $sanitizedPrompt = filter_var($userInput, FILTER_SANITIZE_STRING);
     

Debugging Tips

1. Enable HTTP Debugging Configure Symfony’s HttpClient to log requests:

   # config/packages/http_client.yaml
   framework:
       http_client:
           plugins:
               - Symfony\Component\HttpClient\EventListener\ProfilerListener
   

   View logs in the Symfony Profiler under "HTTP Client".

2. LM Studio Logs Check LM Studio’s logs (Docker: docker logs lmstudio) for errors like:

   • CUDA out of memory → Reduce model size or allocate more GPU RAM.
   • Invalid API request → Validate endpoint/payload against LM Studio’s OpenAI docs.

3. Provider-Specific Errors Catch Symfony\AI\Exception\ProviderException to handle LM Studio failures:

   try {
       $response = $chatCompletion->create(['model' => 'lm_studio', ...]);
   } catch (ProviderException $e) {
       if (str_contains($e->getMessage(), 'LM Studio')) {
           // Log and fallback
       }
   }
   

Extension Points

1. Custom Provider Logic Extend AbstractProvider