ELSER inference service

edit

Creates an inference endpoint to perform an inference task with the elser service. You can also deploy ELSER by using the Elasticsearch inference service.

The API request will automatically download and deploy the ELSER model if it isn’t already downloaded.

Deprecated in 8.16

The elser service is deprecated and will be removed in a future release. Use the Elasticsearch inference service instead, with model_id included in the service_settings.

Request

edit

PUT /_inference/<task_type>/<inference_id>

Path parameters

edit
<inference_id>
(Required, string) The unique identifier of the inference endpoint.
<task_type>

(Required, string) The type of the inference task that the model will perform.

Available task types:

  • sparse_embedding.

Request body

edit
chunking_settings

(Optional, object) Chunking configuration object. Refer to Configuring chunking to learn more about chunking.

max_chunking_size
(Optional, integer) Specifies the maximum size of a chunk in words. Defaults to 250. This value cannot be higher than 300 or lower than 20 (for sentence strategy) or 10 (for word strategy).
overlap
(Optional, integer) Only for word chunking strategy. Specifies the number of overlapping words for chunks. Defaults to 100. This value cannot be higher than the half of max_chunking_size.
sentence_overlap
(Optional, integer) Only for sentence chunking strategy. Specifies the numnber of overlapping sentences for chunks. It can be either 1 or 0. Defaults to 1.
strategy
(Optional, string) Specifies the chunking strategy. It could be either sentence or word.
service
(Required, string) The type of service supported for the specified task type. In this case, elser.
service_settings

(Required, object) Settings used to install the inference model.

These settings are specific to the elser service.

adaptive_allocations

(Optional, object) Adaptive allocations configuration object. If enabled, the number of allocations of the model is set based on the current load the process gets. When the load is high, a new model allocation is automatically created (respecting the value of max_number_of_allocations if it’s set). When the load is low, a model allocation is automatically removed (respecting the value of min_number_of_allocations if it’s set). If adaptive_allocations is enabled, do not set the number of allocations manually.

enabled
(Optional, Boolean) If true, adaptive_allocations is enabled. Defaults to false.
max_number_of_allocations
(Optional, integer) Specifies the maximum number of allocations to scale to. If set, it must be greater than or equal to min_number_of_allocations.
min_number_of_allocations
(Optional, integer) Specifies the minimum number of allocations to scale to. If set, it must be greater than or equal to 0. If not defined, the deployment scales to 0.
num_allocations
(Required, integer) The total number of allocations this model is assigned across machine learning nodes. Increasing this value generally increases the throughput. If adaptive_allocations is enabled, do not set this value, because it’s automatically set.
num_threads
(Required, integer) Sets the number of threads used by each model allocation during inference. This generally increases the speed per inference request. The inference process is a compute-bound process; threads_per_allocations must not exceed the number of available allocated processors per node. Must be a power of 2. Max allowed value is 32.

ELSER service example with adaptive allocations

edit

When adaptive allocations are enabled, the number of allocations of the model is set automatically based on the current load.

For more information on how to optimize your ELSER endpoints, refer to the ELSER recommendations section in the model documentation. To learn more about model autoscaling, refer to the trained model autoscaling page.

The following example shows how to create an inference endpoint called my-elser-model to perform a sparse_embedding task type and configure adaptive allocations.

The request below will automatically download the ELSER model if it isn’t already downloaded and then deploy the model.

resp = client.inference.put(
    task_type="sparse_embedding",
    inference_id="my-elser-model",
    inference_config={
        "service": "elser",
        "service_settings": {
            "adaptive_allocations": {
                "enabled": True,
                "min_number_of_allocations": 3,
                "max_number_of_allocations": 10
            },
            "num_threads": 1
        }
    },
)
print(resp)
const response = await client.inference.put({
  task_type: "sparse_embedding",
  inference_id: "my-elser-model",
  inference_config: {
    service: "elser",
    service_settings: {
      adaptive_allocations: {
        enabled: true,
        min_number_of_allocations: 3,
        max_number_of_allocations: 10,
      },
      num_threads: 1,
    },
  },
});
console.log(response);
PUT _inference/sparse_embedding/my-elser-model
{
  "service": "elser",
  "service_settings": {
    "adaptive_allocations": {
      "enabled": true,
      "min_number_of_allocations": 3,
      "max_number_of_allocations": 10
    },
    "num_threads": 1
  }
}

ELSER service example without adaptive allocations

edit

The following example shows how to create an inference endpoint called my-elser-model to perform a sparse_embedding task type. Refer to the ELSER model documentation for more info.

If you want to optimize your ELSER endpoint for ingest, set the number of threads to 1 ("num_threads": 1). If you want to optimize your ELSER endpoint for search, set the number of threads to greater than 1.

The request below will automatically download the ELSER model if it isn’t already downloaded and then deploy the model.

resp = client.inference.put(
    task_type="sparse_embedding",
    inference_id="my-elser-model",
    inference_config={
        "service": "elser",
        "service_settings": {
            "num_allocations": 1,
            "num_threads": 1
        }
    },
)
print(resp)
const response = await client.inference.put({
  task_type: "sparse_embedding",
  inference_id: "my-elser-model",
  inference_config: {
    service: "elser",
    service_settings: {
      num_allocations: 1,
      num_threads: 1,
    },
  },
});
console.log(response);
PUT _inference/sparse_embedding/my-elser-model
{
  "service": "elser",
  "service_settings": {
    "num_allocations": 1,
    "num_threads": 1
  }
}

Example response:

{
  "inference_id": "my-elser-model",
  "task_type": "sparse_embedding",
  "service": "elser",
  "service_settings": {
    "num_allocations": 1,
    "num_threads": 1
  },
  "task_settings": {}
}

You might see a 502 bad gateway error in the response when using the Kibana Console. This error usually just reflects a timeout, while the model downloads in the background. You can check the download progress in the Machine Learning UI. If using the Python client, you can set the timeout parameter to a higher value.