Subchapter 8.4

Sagemaker

references/sagemaker.mdMarkdown16 KBView on GitHub

Amazon SageMaker

The managed middle ground: Deepgram runs inside your AWS account and VPC, but AWS handles instance provisioning, scaling, and container management. You subscribe to a Deepgram listing on AWS Marketplace and deploy a SageMaker Endpoint — no Quay credentials, no .dg model files, no driver installs.

(public) ships a skill with 13 deterministic scripts covering preflight, product selection, Marketplace subscribe, IAM execution role, quota check, deploy, invoke test, autoscaling, update, and teardown. Install it and use its scripts rather than hand-writing calls:

Sagemaker · Self Hosted · deepgram/skills · Skills Docs

deepgram-sagemaker

aws sagemaker create-*

Product	Recommended	Also supported	Not supported
Nova-3 STT	`ml.g6.2xlarge`	`ml.g7.2xlarge`, `ml.g7e.2xlarge`, `ml.g6e.2xlarge`, `ml.g5.2xlarge`, `ml.g4dn.2xlarge`	—
Flux STT	`ml.g6.2xlarge`	`ml.g7.2xlarge`, `ml.g7e.2xlarge`, `ml.g6e.2xlarge`, `ml.g5.2xlarge`	`ml.g4dn.*` (no `sm_75` kernel)
Aura-2 TTS	`ml.g6.12xlarge`	`ml.g7.12xlarge`, `ml.g7e.12xlarge`, `ml.g5.12xlarge`, `ml.g6e.12xlarge`, `ml.g4dn.12xlarge`	Single-GPU types — Aura-2 needs 2+ GPUs
Flux TTS (Aura-3)	`ml.g6e.2xlarge`	`ml.g7.2xlarge`, `ml.g7e.2xlarge`, `ml.g6.2xlarge`	`ml.g5.`, `ml.g4dn.`

SDK	Package	Latest published	Repository
Python	PyPI `deepgram-sagemaker`	0.4.0	deepgram-python-sdk-transport-sagemaker (opens in a new tab)
Java	Maven Central `com.deepgram:deepgram-sagemaker`	0.1.3	deepgram-java-sdk-transport-sagemaker (opens in a new tab)
JavaScript	npm `@deepgram/sagemaker`	0.1.2	deepgram-js-sdk-transport-sagemaker (opens in a new tab)

import asyncio
from deepgram import AsyncDeepgramClient
from deepgram.core.events import EventType
from deepgram_sagemaker import SageMakerTransportFactory

factory = SageMakerTransportFactory(
    endpoint_name="my-deepgram-endpoint",
    region="us-west-2",
)
client = AsyncDeepgramClient(api_key="unused", transport_factory=factory)

async def main():
    async with client.listen.v1.connect(model="nova-3") as connection:
        connection.on(EventType.MESSAGE, lambda msg: print(msg))
        await connection.start_listening()

asyncio.run(main())

from deepgram_sagemaker import SageMakerConfig, SageMakerTransportFactory

config = SageMakerConfig(
    endpoint_name="my-deepgram-endpoint",
    region="us-east-2",
    connection_timeout=5.0,          # default 30.0
    connection_acquire_timeout=15.0, # default 60.0
)
factory = SageMakerTransportFactory(config=config)

import { DeepgramClient } from "@deepgram/sdk";
import { createSageMakerTransportFactory } from "@deepgram/sagemaker";

const transportFactory = createSageMakerTransportFactory({
  endpointName: "my-deepgram-endpoint",
  region: "us-west-2",
});

const client = new DeepgramClient({ apiKey: "unused", transportFactory });

const socket = await client.listen.v1.createConnection({ model: "nova-3" });
socket.on("message", (message) => console.log(message));
socket.connect();
socket.sendMedia(new Uint8Array([1, 2, 3]));

SageMakerTransportFactory factory = new SageMakerTransportFactory(
    SageMakerConfig.builder()
        .endpointName("my-deepgram-endpoint")
        .region("us-west-2")
        .build()
);

DeepgramClient client = DeepgramClient.builder()
    .apiKey("unused")
    .transportFactory(factory)
    .build();

V1WebSocketClient ws = client.listen().v1().v1WebSocket();
ws.onResults(results -> System.out.println(
    results.getChannel().getAlternatives().get(0).getTranscript()));
ws.connect(V1ConnectOptions.builder().model(ListenV1Model.NOVA3).build());
ws.sendMedia(audioBytes);
ws.close();

Sagemaker

Amazon SageMaker

Sagemaker

Amazon SageMaker

When SageMaker fits

Product listings

Instance types

Invocation modes

SDK transports

Python

JavaScript / TypeScript

Java

Validating an endpoint

Operations

Sources