Give Feedback

digitalocean_dedicated_inference

Generated on 6 May 2026 from Terraform version v2.85.1

Copy page as Markdown View page as Markdown

Provides a DigitalOcean Dedicated Inference resource. This can be used to create, modify, and delete dedicated inference endpoints for running GPU-accelerated model inference.

Example Usage

resource "digitalocean_dedicated_inference" "example" {
  name   = "my-inference-endpoint"
  region = "tor1"

  model_deployments {
    model_slug     = "deepseek-r1-distill-qwen-14b"
    model_provider = "digitalocean"

    accelerators {
      accelerator_slug = "gpu-h100x1-80gb"
      scale            = 1
      type             = "nvidia_h100"
    }
  }
}

With Public Endpoint

resource "digitalocean_dedicated_inference" "public" {
  name                   = "my-public-inference"
  region                 = "tor1"
  enable_public_endpoint = true

  model_deployments {
    model_slug     = "deepseek-r1-distill-qwen-14b"
    model_provider = "digitalocean"

    accelerators {
      accelerator_slug = "gpu-h100x1-80gb"
      scale            = 1
      type             = "nvidia_h100"
    }
  }
}

With VPC

resource "digitalocean_dedicated_inference" "private" {
  name     = "my-private-inference"
  region   = "tor1"
  vpc_uuid = digitalocean_vpc.example.id

  model_deployments {
    model_slug     = "deepseek-r1-distill-qwen-14b"
    model_provider = "digitalocean"

    accelerators {
      accelerator_slug = "gpu-h100x1-80gb"
      scale            = 1
      type             = "nvidia_h100"
    }
  }
}

Argument Reference

The following arguments are supported:

name - (Required) A human-readable name for the dedicated inference endpoint.
region - (Required) The region slug where the dedicated inference endpoint will be deployed. Changing this forces a new resource.
enable_public_endpoint - (Optional) Whether to enable a public HTTPS endpoint for the dedicated inference endpoint. Defaults to false. This field is immutable after creation and changing it forces a new resource.
vpc_uuid - (Optional) The UUID of the VPC to deploy the dedicated inference endpoint into. Changing this forces a new resource.
model_deployments - (Required) The list of model deployments to run on the dedicated inference endpoint. Each model_deployments block supports:
- model_slug - (Required) The slug identifier for the model to deploy.
- model_provider - (Required) The provider of the model (e.g. digitalocean, huggingface).
- model_id - (Optional) The unique ID of the model.
- accelerators - (Required) The GPU accelerators to allocate for this model deployment. Each accelerators block supports:
  - accelerator_slug - (Required) The slug identifier for the GPU accelerator type.
  - scale - (Required) The number of accelerator units to allocate. Must be at least 1.
  - type - (Required) The accelerator type.
hugging_face_token - (Optional, Sensitive) A HuggingFace token for accessing gated models.

Attributes Reference

In addition to the arguments listed above, the following attributes are exported:

id - The unique ID of the dedicated inference endpoint.
status - The current status of the dedicated inference endpoint.
public_endpoint_fqdn - The fully-qualified domain name of the public endpoint, if enabled.
private_endpoint_fqdn - The fully-qualified domain name of the private endpoint.
created_at - The date and time when the dedicated inference endpoint was created.
updated_at - The date and time when the dedicated inference endpoint was last updated.

Import

Dedicated inference endpoints can be imported using their id, e.g.

terraform import digitalocean_dedicated_inference.example endpoint-id

Timeouts

digitalocean_dedicated_inference provides the following Timeouts configuration options:

create - (Default 60 minutes) Used for creating the dedicated inference endpoint.
update - (Default 60 minutes) Used for updating the dedicated inference endpoint.
delete - (Default 60 minutes) Used for deleting the dedicated inference endpoint.