OpenAIEmbeddingFunction

Description

An embedding function that utilizes the OpenAI API to compute vector embeddings, commonly used with vector databases. This class, part of pymilvus, handles encoding text into embeddings using OpenAI models to support embedding retrieval in Milvus.

Features

The OpenAIEmbeddingFunction offers flexible configuration for integrating with OpenAI's embedding services:

• Model Selection: Choose from various OpenAI models for encoding, including text-embedding-3-small, text-embedding-3-large, and text-embedding-ada-002 (default).
• API Key Management: Securely provide your OpenAI API key; the function also checks environment variables as a fallback.
• Custom Endpoint Support: Configure a custom base URL for the OpenAI API endpoint, defaulting to the public OpenAI API server.
• Embedding Dimensions Control: Specify the desired number of dimensions for the output embeddings, a feature supported by text-embedding-3 and later models.
• Extensible Configuration: Allows passing additional keyword arguments directly to the underlying OpenAI model initialization for advanced use cases.

Constructor Parameters

To initialize OpenAIEmbeddingFunction, the following parameters are available:

• model_name (string): Specifies the OpenAI model for encoding. Valid options are text-embedding-3-small, text-embedding-3-large, and text-embedding-ada-002 (default).
• api_key (string, optional): Your OpenAI API key. If unspecified, environment variables are checked.
• base_url (string, optional): The base URL of the OpenAI API endpoint. Defaults to None (public OpenAI API server).
• dimensions (int, optional): The number of dimensions the resulting output embeddings should have. Only supported in text-embedding-3 and later models.
• **kwargs: Allows additional keyword arguments to be passed to the model initialization.