› › ›

Create trained model vocabulary API

Creates a trained model vocabulary. This is supported only for natural language processing (NLP) models.

Request

PUT _ml/trained_models/<model_id>/vocabulary/

Prerequisites

Requires the manage_ml cluster privilege. This privilege is included in the machine_learning_admin built-in role.

Description

The vocabulary is stored in the index as described in inference_config.*.vocabulary of the trained model definition.

Path parameters

<model_id>: (Required, string) The unique identifier of the trained model.

Request body

vocabulary: (array) The model vocabulary. Must not be empty.
merges: (Optional, array) The model merges used in byte-pair encoding. The merges must be sub-token pairs, space delimited, and in order of preference. Example: ["f o", "fo o"]. Must be provided for RoBERTa and BART style models.
scores: (Optional, array) Vocabulary value scores used by sentence-piece tokenization. Must have the same length as vocabulary. Required for unigram sentence-piece tokenized models like XLMRoberta and T5.

Examples

The following example shows how to create a model vocabulary for a previously stored trained model configuration.

PUT _ml/trained_models/elastic__distilbert-base-uncased-finetuned-conll03-english/vocabulary
{
  "vocabulary": [
    "[PAD]",
    "[unused0]",
    ...
  ]
}

The API returns the following results:

{
    "acknowledged": true
}

© 2023-2025 Elasticsearch
As of September 2024, Elasticsearch is available under a choice of three licenses: the Server Side Public License (SSPL), the Elastic License, or the AGPLv3 (OSI approved).
Elasticsearch and the Elasticsearch logo are trademarks of Elasticsearch B.V., registered in the U.S. and in other countries.
https://www.elastic.co/guide/en/elasticsearch/reference/8.17/put-trained-model-vocabulary.html