WeaviateEx.API.Vectorizers.Text2VecTransformers (WeaviateEx v0.7.4)

Text2Vec-Transformers vectorizer configuration.

Uses local transformer models for text embedding. This module requires a running transformers inference service.

Example

Text2VecTransformers.new(
  pooling_strategy: :masked_mean,
  inference_url: "http://localhost:8080"
)

Passage/Query Models

For asymmetric search (different embeddings for queries vs documents):

Text2VecTransformers.new(
  passage_inference_url: "http://localhost:8081",
  query_inference_url: "http://localhost:8082"
)

Summary

Types

pooling_strategy()

t()

Functions

from_api(map)

Parse configuration from API response.

new(opts \\ [])

Create a new Text2Vec-Transformers configuration.

to_api(config)

Convert configuration to API format.

vectorizer_name()

Returns the vectorizer name for the API.

Types

pooling_strategy()

@type pooling_strategy() :: :masked_mean | :cls

t()

@type t() :: %WeaviateEx.API.Vectorizers.Text2VecTransformers{
  inference_url: String.t() | nil,
  passage_inference_url: String.t() | nil,
  pooling_strategy: pooling_strategy() | nil,
  query_inference_url: String.t() | nil,
  vectorize_collection_name: boolean()
}

Functions

from_api(map)

@spec from_api(map()) :: t()

Parse configuration from API response.

new(opts \\ [])

@spec new(keyword()) :: t()

Create a new Text2Vec-Transformers configuration.

Options

:pooling_strategy - Pooling strategy (:masked_mean, :cls)
:inference_url - URL of the transformers inference service
:passage_inference_url - URL for passage/document inference
:query_inference_url - URL for query inference
:vectorize_collection_name - Include collection name (default: true)

to_api(config)

@spec to_api(t()) :: map()

Convert configuration to API format.

vectorizer_name()

@spec vectorizer_name() :: String.t()

Returns the vectorizer name for the API.