Skip to main content

Configuration

Overview​

The ConfigManager manages the configuration and dependency injection of the language models used within the library.

This documentation explains how the configuration system works, for both contributors and users of the library.

How Configuration Works​

For Library Users:​

Initialize Users must initialize the ConfigManager and configure it to use specific values before utilizing them in their application. This is achieved by invoking the configure() method and passing a configuration dictionary.

Current Configuration Dictionary:

config_dict = {
   'open_ai_chat_model': 'your_desired_chat_model',
   'open_ai_embedding_model': 'your_desired_embedding_model',
}

Initializing and Configuring:

config_manager = ConfigManager()
config_manager.configure(config_dict)

For Library Contributors:​

Understanding ConfigManager​

  • Singleton Pattern: ConfigManager employs a singleton pattern, ensuring that a single instance is utilized across the entire application, preserving a consistent configuration state.

  • Dependency Injection: Using the injector module, it defines classes (e.g. ChatModule or EmbeddingModule) to handle dependency injection and configuration for models (e.g. OpenAIChat or OpenAIEmbedding).

  • Lazy Initialization: Models are not instantiated until they are requested, ensuring configuration is only needed, when you intend to use a model that requires that specific configuration.

Additional Information​

Look at cognaize.melody.language_models.chat.open_ai_chat or cognaize.melody.language_models.embedding.open_ai_embedding modules to understand injection implementations and extend them as per your needs.