Start trained model deployment API

edit

Starts a new trained model deployment.

Request

edit

POST _ml/trained_models/<model_id>/deployment/_start

Prerequisites

edit

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

Description

edit

Currently only pytorch models are supported for deployment. Once deployed the model can be used by the Inference processor in an ingest pipeline or directly in the Infer trained model API.

A model can be deployed multiple times by using deployment IDs. A deployment ID must be unique and should not match any other deployment ID or model ID, unless it is the same as the ID of the model being deployed. If deployment_id is not set, it defaults to the model_id.

You can enable adaptive allocations to automatically scale model allocations up and down based on the actual resource requirement of the processes.

Manually scaling inference performance can be achieved by setting the parameters number_of_allocations and threads_per_allocation.

Increasing threads_per_allocation means more threads are used when an inference request is processed on a node. This can improve inference speed for certain models. It may also result in improvement to throughput.

Increasing number_of_allocations means more threads are used to process multiple inference requests in parallel resulting in throughput improvement. Each model allocation uses a number of threads defined by threads_per_allocation.

Model allocations are distributed across machine learning nodes. All allocations assigned to a node share the same copy of the model in memory. To avoid thread oversubscription which is detrimental to performance, model allocations are distributed in such a way that the total number of used threads does not surpass the node’s allocated processors.

Path parameters

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

Query parameters

edit
deployment_id

(Optional, string) A unique identifier for the deployment of the model.

Defaults to model_id.

timeout
(Optional, time) Controls the amount of time to wait for the model to deploy. Defaults to 30 seconds.
wait_for
(Optional, string) Specifies the allocation status to wait for before returning. Defaults to started. The value starting indicates deployment is starting but not yet on any node. The value started indicates the model has started on at least one node. The value fully_allocated indicates the deployment has started on all valid nodes.

Request body

edit
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 1.
cache_size
(Optional, byte value) The inference cache size (in memory outside the JVM heap) per node for the model. In serverless, the cache is disabled by default. Otherwise, the default value is the size of the model as reported by the model_size_bytes field in the Get trained models stats. To disable the cache, 0b can be provided.
number_of_allocations
(Optional, integer) The total number of allocations this model is assigned across machine learning nodes. Increasing this value generally increases the throughput. Defaults to 1. If adaptive_allocations is enabled, do not set this value, because it’s automatically set.
priority

(Optional, string) The priority of the deployment. The default value is normal. There are two priority settings:

  • normal: Use this for deployments in production. The deployment allocations are distributed so that node processors are not oversubscribed.
  • low: Use this for testing model functionality. The intention is that these deployments are not sent a high volume of input. The deployment is required to have a single allocation with just one thread. Low priority deployments may be assigned on nodes that already utilize all their processors but will be given a lower CPU priority than normal deployments. Low priority deployments may be unassigned in order to satisfy more allocations of normal priority deployments.

Heavy usage of low priority deployments may impact performance of normal priority deployments.

queue_capacity
(Optional, integer) Controls how many inference requests are allowed in the queue at a time. Every machine learning node in the cluster where the model can be allocated has a queue of this size; when the number of requests exceeds the total value, new requests are rejected with a 429 error. Defaults to 1024. Max allowed value is 1000000.
threads_per_allocation
(Optional, 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. Defaults to 1. Must be a power of 2. Max allowed value is 32.

Examples

edit

The following example starts a new deployment for a elastic__distilbert-base-uncased-finetuned-conll03-english trained model:

resp = client.ml.start_trained_model_deployment(
    model_id="elastic__distilbert-base-uncased-finetuned-conll03-english",
    wait_for="started",
    timeout="1m",
)
print(resp)
const response = await client.ml.startTrainedModelDeployment({
  model_id: "elastic__distilbert-base-uncased-finetuned-conll03-english",
  wait_for: "started",
  timeout: "1m",
});
console.log(response);
POST _ml/trained_models/elastic__distilbert-base-uncased-finetuned-conll03-english/deployment/_start?wait_for=started&timeout=1m

The API returns the following results:

{
    "assignment": {
        "task_parameters": {
            "model_id": "elastic__distilbert-base-uncased-finetuned-conll03-english",
            "model_bytes": 265632637,
            "threads_per_allocation" : 1,
            "number_of_allocations" : 1,
            "queue_capacity" : 1024,
            "priority": "normal"
        },
        "routing_table": {
            "uckeG3R8TLe2MMNBQ6AGrw": {
                "routing_state": "started",
                "reason": ""
            }
        },
        "assignment_state": "started",
        "start_time": "2022-11-02T11:50:34.766591Z"
    }
}

Using deployment IDs

edit

The following example starts a new deployment for the my_model trained model with the ID my_model_for_ingest. The deployment ID an be used in inference API calls or in inference processors.

resp = client.ml.start_trained_model_deployment(
    model_id="my_model",
    deployment_id="my_model_for_ingest",
)
print(resp)
const response = await client.ml.startTrainedModelDeployment({
  model_id: "my_model",
  deployment_id: "my_model_for_ingest",
});
console.log(response);
POST _ml/trained_models/my_model/deployment/_start?deployment_id=my_model_for_ingest

The my_model trained model can be deployed again with a different ID:

resp = client.ml.start_trained_model_deployment(
    model_id="my_model",
    deployment_id="my_model_for_search",
)
print(resp)
const response = await client.ml.startTrainedModelDeployment({
  model_id: "my_model",
  deployment_id: "my_model_for_search",
});
console.log(response);
POST _ml/trained_models/my_model/deployment/_start?deployment_id=my_model_for_search

Setting adaptive allocations

edit

The following example starts a new deployment of the my_model trained model with the ID my_model_for_search and enables adaptive allocations with the minimum number of 3 allocations and the maximum number of 10.

resp = client.ml.start_trained_model_deployment(
    model_id="my_model",
    deployment_id="my_model_for_search",
)
print(resp)
const response = await client.ml.startTrainedModelDeployment({
  model_id: "my_model",
  deployment_id: "my_model_for_search",
});
console.log(response);
POST _ml/trained_models/my_model/deployment/_start?deployment_id=my_model_for_search
{
  "adaptive_allocations": {
    "enabled": true,
    "min_number_of_allocations": 3,
    "max_number_of_allocations": 10
  }
}