Managed Inference API

Introduction

Scaleway Managed Inference allows you to deploy and run machine learning models on Scaleway's infrastructure. This service provides scalable and efficient endpoints for your model inference needs. The Scaleway Inference API enables you to manage these endpoints and perform inference operations.

Concepts

Refer to our dedicated concepts page to find definitions of all concepts and terminology related to Managed Inference.

Quickstart

1. Configure your environment variables

   Note

   This is an optional step that seeks to simplify your usage of the Inference API. You can find your Project ID in the Scaleway console.

   export SCW_SECRET_KEY="<API secret key>"
   export SCW_DEFAULT_REGION="fr-par"
   export SCW_PROJECT_ID="<Scaleway Project ID>"
   

2. Create a model deployment: Run the following command to create a deployment. Customize the details in the payload (name, model, description, tags, etc.) to your needs:

   curl -X POST https://api.scaleway.com/inference/v1beta1/regions/$SCW_DEFAULT_REGION/deployments \
   -H "Content-Type: application/json" \
   -H "X-Auth-Token: $SCW_SECRET_KEY" \
   -d '{
     "project_id": "'"$SCW_PROJECT_ID"'",
     "name": "my-inference-deployment",
     "model_name": "meta/llama-3-8b-instruct:bf16",
     "node_type": "L4",
     "min_size": 1,
     "max_size": 3,
     "accept_eula": true,
     "endpoints": [
       {
         "public": {}
       }
     ]
   }'
   

   Parameter	Description	Valid values
   project_id	The Project in which the deployment should be created (string)	Any valid Scaleway Project ID, e.g., "b4bd99e0-b389-11ed-afa1-0242ac120002"
   name	A name of your choice for the deployment (string)	Any string containing only alphanumeric characters, dots, spaces, and dashes, e.g., "my-inference-deployment"
   model_name	The model to deploy (string)	Any valid model name in the format vendor/model-name:version, e.g., "meta/llama-3-8b-instruct:bf16"
   node_type	The type of node to use for the deployment (string)	Example: "L4"
   min_size	Minimum number of replicas for the deployment (integer)	Any integer, e.g., 1
   max_size	Maximum number of replicas for the deployment (integer)	Any integer, e.g., 3
   accept_eula	Indicates acceptance of the End User License Agreement (boolean)	true
   endpoints	Defines the endpoints for the deployment (array)	At least one endpoint, e.g., [ { "public": {} } ]

3. Create a model endpoint: Run the following command to create an inference endpoint for the deployment. Customize the details in the payload to your needs:

   Example for creating a public endpoint

   curl -X POST https://api.scaleway.com/inference/v1beta1/regions/$SCW_DEFAULT_REGION/endpoints \
   -H "Content-Type: application/json" \
   -H "X-Auth-Token: $SCW_SECRET_KEY" \
   -d '{
     "project_id": "'"$SCW_PROJECT_ID"'",
     "deployment_id": "your-deployment-id",
     "endpoint": {
       "disable_auth": false,
       "public": {}
     }
   }'
   

   Example for creating a private endpoint

   curl -X POST https://api.scaleway.com/inference/v1beta1/regions/$SCW_DEFAULT_REGION/endpoints \
   -H "Content-Type: application/json" \
   -H "X-Auth-Token: $SCW_SECRET_KEY" \
   -d '{
     "project_id": "'"$SCW_PROJECT_ID"'",
     "deployment_id": "your-deployment-id",
     "endpoint": {
       "disable_auth": false,
       "private_network": {
         "private_network_id": "your-private-network-id"
       }
     }
   }'
   

   Parameter	Description	Valid values
   project_id	The Project in which the endpoint should be created (string)	Any valid Scaleway Project ID, e.g., "b4bd99e0-b389-11ed-afa1-0242ac120002"
   deployment_id	The deployment ID to which the endpoint will be associated (string)	Any valid deployment ID, e.g. "bcb0976d-98d6-49c1-b6b5-17804941c0b7"
   disable_auth	Specifies whether to disable authentication (boolean)	true or false
   public	Public endpoint configuration (object)	{} for public endpoint
   private_network	Private endpoint configuration including the private network ID (object)	{ "private_network_id": "private-network-id" }

4. List your deployments: Run the following command to get a list of all the deployments in your account, with their details:

   curl -X GET \
     -H "Content-Type: application/json" \
     -H "X-Auth-Token: $SCW_SECRET_KEY" \
     "https://api.scaleway.com/inference/v1beta1/regions/$SCW_DEFAULT_REGION/deployments"
   

5. List your endpoints: Run the following command to get a list of all the inference endpoints in your account, with their details:

   curl -X GET \
     -H "Content-Type: application/json" \
     -H "X-Auth-Token: $SCW_SECRET_KEY" \
     "https://api.scaleway.com/inference/v1beta1/regions/$SCW_DEFAULT_REGION/endpoints"
   

6. Delete an endpoint: Run the following command to delete an inference endpoint, specified by its endpoint ID:

   curl -X DELETE \
     -H "X-Auth-Token: $SCW_SECRET_KEY" \
     -H "Content-Type: application/json" \
     "https://api.scaleway.com/inference/v1beta1/regions/$SCW_DEFAULT_REGION/endpoints/<endpoint-ID>"
   

   The expected successful response is empty.

   Important

   Managed Inference deployments must have at least one endpoint, either public or private.

Technical information

Region

Managed Inference endpoints are available in the following region:

Pagination

Most listing requests receive a paginated response. Requests against paginated endpoints accept two query arguments:

• page, a positive integer to choose which page to return.
• per_page, a positive integer lower or equal to 100 to select the number of items to return per page. The default value is 50.

Paginated endpoints usually also accept filters to search and sort results. These filters are documented along each endpoint documentation.

The X-Total-Count header contains the total number of items returned.

Creating a deployment: the model object

When creating a deployment, the model_id parameter is required. This specifies the model to deploy. Use the List Models endpoint to retrieve available model IDs.

Going further

For more help using Scaleway Inference, check out the following resources:

• Our main documentation
• The #inference-beta channel on our Slack Community
• Do not hesitate to reach out with any questions or issues you encounter during the public beta, our teams will help you through the dedicated Slack channel #inference-beta.

Models

Deployments

Node types

Access Control List

Endpoints