SmartEmbedModel is a versatile and convenient interface for utilizing various embedding models via API and locally. It provides a unified way to work with different embedding models, making it easy to switch between models or use multiple models in your projects.

• Support for multiple embedding models
• Easy-to-use API for embedding text
• Batch processing capabilities
• Token counting functionality
• Flexible configuration options
• Support for both local and API-based models

SmartEmbedModel can be configured with various options (opts{}):

• adapters{}: Available adapters. Default adapters include:
  • transformers
  • openai
  • transformers_iframe
  • transformers_worker

• model_key: The specific model identifier.
• model_config{}: Overrides defaults in models.json.
  • adapter: Selects the adapter to use (e.g., transformers_iframe).

• settings{}: Overrides defaults in settings_config.
  • model_key: The model key to use if not set directly.

• batch_size: Number of inputs per batch.
• max_tokens: Maximum tokens the model can handle.
• use_gpu: Enable GPU acceleration.
• gpu_batch_size: Batch size when GPU is enabled.

The available models are defined in the models.json file. Some of the included models are:

• TaylorAI/bge-micro-v2
• andersonbcdefg/bge-small-4096
• Xenova/jina-embeddings-v2-base-zh
• text-embedding-3-small
• text-embedding-3-large
• And more...

SmartEmbedModel uses adapters to interface with different embedding models. The available adapters are:

• transformers: Use the transformers library to interface with embedding models
  • transformers_iframe: Use the transformers library in an iframe
  • transformers_worker: Use the transformers library in a web worker
• openai: Use the openai library to interface with embedding models

Base class for all adapters. Adapters must implement the following methods:

• load(): Initialize and load the model.
• count_tokens(input: string | Object): Promise<number | Object>: Count tokens in the input.
• embed(input: string | Object): Promise<Object>: Embed a single input.
• embed_batch(inputs: Array<Object>): Promise<Array<Object>>: Embed a batch of inputs.
• unload(): Clean up and unload the model.

Uses OpenAI's API for embeddings.

const embed_model = new SmartEmbedModel({
  model_key: 'text-embedding-ada-002',
  settings: {
    api_key: 'YOUR_OPENAI_API_KEY',
  },
  adapters: {
    openai: SmartEmbedOpenAIAdapter,
  },
});

This project is licensed under the MIT License - see the LICENSE file for details.

Brian Joseph Petro (🌴 Brian)