OnnxEmbeddingFunction

OnnxEmbeddingFunction is a class in pymilvus that handles encoding text into embeddings using Open Neural Network Exchange (ONNX) embedding models to support embedding retrieval in Milvus.

pymilvus.model.dense.OnnxEmbeddingFunction

Constructor

Constructs an OnnxEmbeddingFunction for common use cases.

OnnxEmbeddingFunction(
    model_name: str = "GPTCache/paraphrase-albert-onnx",
    tokenizer_name: str = "GPTCache/paraphrase-albert-small-v2"
)

PARAMETERS:

• model_name (string)

  The repository ID on the Hugging Face Hub that contains the pre-trained ONNX model file. For example, in the provided code, it is set to GPTCache/paraphrase-albert-onnx by default. This repository should contain a compatible ONNX model for the desired natural language processing task, such as text classification, token classification, or feature extraction.

• tokenizer_name (string)

  The repository ID on the Hugging Face Hub that contains the tokenizer configuration compatible with the specified ONNX model. In the provided code, it is set to GPTCache/paraphrase-albert-small-v2 by default. The tokenizer handles text preprocessing, such as tokenization, padding, and encoding, ensuring compatibility with the ONNX model’s input format. The tokenizer should be pre-trained and compatible with the ONNX model for the same task.

Examples

from pymilvus.model.dense import OnnxEmbeddingFunction

onnx_ef = OnnxEmbeddingFunction(
    model_name="GPTCache/paraphrase-albert-onnx", # Defaults to `GPTCache/paraphrase-albert-onnx`
    tokenizer_name="GPTCache/paraphrase-albert-small-v2" # Defaults to `GPTCache/paraphrase-albert-small-v2`
)