4.7.7 ML_EMBED_ROW

The ML_EMBED_ROW routine uses the specified embedding model to encode the specified text or query into a vector embedding. The routine returns a VECTOR that contains a numerical representation of the specified text.

ML_EMBED_ROW Syntax

mysql> select sys.ML_EMBED_ROW('Text', [options]);

options: {
  JSON_OBJECT('key','value'[,'key','value'] ...)
    'key','value': {
    ['model_id', {'all_minilm_l12_v2'|'multilingual-e5-small'|'cohere.embed-english-v3.0'|'cohere.embed-multilingual-v3.0'}]
    ['truncate', {true|false}]
    }
}

Following are ML_EMBED_ROW parameters:

• Text: specifies the text to encode.

• options: specifies optional parameters as key-value pairs in JSON format. It can include the following parameters:

  • model_id: specifies the embedding model to use for encoding the text. Default value is all_minilm_l12_v2. Possible values are:

    • all_minilm_l12_v2: for encoding English text.

    • multilingual-e5-small: for encoding text in supported languages other than English (en). This embedding model is available in HeatWave 9.0.1-u1 and later versions.

    • cohere.embed-english-v3.0: for encoding English text. This embedding model is available in HeatWave 9.0.1-u1 and later versions.

    • cohere.embed-multilingual-v3.0: for encoding text in supported languages other than English. This embedding model is available in HeatWave 9.0.1-u1 and later versions.

    To view the lists of supported models, see HeatWave In-Database Embedding Models and OCI Generative AI Service Embedding Models. To view the list of supported languages, see Languages.

  • truncate: specifies whether to truncate inputs longer than the maximum token size. Default value is true.

Syntax Examples

• Embedding an English query using the all_minilm_l12_v2 embedding model, and store the generated embedding in the @text_embedding session variable:

  mysql> select sys.ML_EMBED_ROW("What is artificial intelligence?", JSON_OBJECT("model_id", "all_minilm_l12_v2")) into @text_embedding;