LanceDB 中文文档

外观

本页尚未翻译

本页内容仍为英文原文，源自官方文档 integrations/embedding/voyageai。中文翻译正在分批进行中。

Voyage AI ​

Voyage AI provides cutting-edge embedding and rerankers.

Using voyageai API requires voyageai package, which can be installed using pip install voyageai. Voyage AI embeddings are used to generate embeddings for text data. The embeddings can be used for various tasks like semantic search, clustering, and classification. You also need to set the VOYAGE_API_KEY environment variable to use the Voyage AI API.

bash
export VOYAGE_API_KEY="your-api-key"

Supported models are:

  • voyage-4-large (best retrieval quality, 1024 default dimensions, supports 256/512/1024/2048)
  • voyage-4 (balanced general-purpose, 1024 default dimensions, supports 256/512/1024/2048)
  • voyage-4-lite (optimized for latency/cost, 1024 default dimensions, supports 256/512/1024/2048)
  • voyage-context-3
  • voyage-3.5
  • voyage-3.5-lite
  • voyage-3
  • voyage-3-lite
  • voyage-finance-2
  • voyage-multilingual-2
  • voyage-law-2
  • voyage-code-2
  • voyage-multimodal-3.5 (multimodal - supports text, images, and video)

说明

Multimodal Model: voyage-multimodal-3.5 supports text, images, and video inputs. It outputs 1024-dimensional embeddings by default, configurable via the output_dimension parameter (256, 512, 1024, 2048). See the Voyage AI multimodal embeddings documentation for more details.

Supported parameters (to be passed in create method) are:

ParameterTypeDefault ValueDescription
namestrNoneThe model ID of the model to use. Supported models: voyage-4-large, voyage-4, voyage-4-lite, voyage-3, voyage-3-lite, voyage-3.5, voyage-3.5-lite, voyage-context-3, voyage-finance-2, voyage-multilingual-2, voyage-law-2, voyage-code-2, voyage-multimodal-3.5
input_typestrNoneType of the input text. Default to None. Other options: query, document.
truncationboolTrueWhether to truncate the input texts to fit within the context length.
output_dimensionintNoneOutput embedding dimension. Only supported by voyage-multimodal-3.5. Valid options: 256, 512, 1024 (default), 2048.

For text models, compute_source_embeddings() accepts a single string, a list of strings, a one-dimensional NumPy array, or a PyArrow Array or ChunkedArray. Batch inputs retain their order.

Usage Example:

python
import tempfile
from pathlib import Path

import lancedb
from lancedb.embeddings import EmbeddingFunctionRegistry
from lancedb.pydantic import LanceModel, Vector

voyageai = (
    EmbeddingFunctionRegistry.get_instance().get("voyageai").create(name="voyage-3")
)

class TextModel(LanceModel):
    text: str = voyageai.SourceField()
    vector: Vector(voyageai.ndims()) = voyageai.VectorField()

data = [{"text": "hello world"}, {"text": "goodbye world"}]

db = lancedb.connect(str(Path(tempfile.mkdtemp()) / "voyageai-demo"))
tbl = db.create_table("test", schema=TextModel, mode="overwrite")

tbl.add(data)

Multimodal Example ​

The voyage-multimodal-3.5 model can embed text alongside images. You can use image URLs, local pathlib.Path objects, raw image bytes, or PIL Image objects.

For source embeddings, PIL images and images loaded from bytes or local paths are converted to RGB and encoded as JPEG. This supports RGBA, grayscale with alpha, and palette images; transparency is discarded. Composite transparent images onto your chosen background before embedding if the background color matters.

python
import tempfile
from pathlib import Path

import lancedb
from lancedb.embeddings import EmbeddingFunctionRegistry
from lancedb.pydantic import LanceModel, Vector

# Create multimodal embedding function with custom dimension
voyageai = (
    EmbeddingFunctionRegistry.get_instance()
    .get("voyageai")
    .create(name="voyage-multimodal-3.5", output_dimension=512)
)

class ImageModel(LanceModel):
    image_uri: str = voyageai.SourceField()
    vector: Vector(voyageai.ndims()) = voyageai.VectorField()

db = lancedb.connect(str(Path(tempfile.mkdtemp()) / "voyageai-multimodal"))
tbl = db.create_table("images", schema=ImageModel, mode="overwrite")

# Add images using URLs
tbl.add(
    [
        {
            "image_uri": "https://upload.wikimedia.org/wikipedia/commons/thumb/4/47/PNG_transparency_demonstration_1.png/300px-PNG_transparency_demonstration_1.png"
        },
    ]
)

# Search with text query
results = tbl.search("dice").limit(1).to_list()
print(results)