ML Products on AWS Marketplace via SageMaker

Deploy Upstage AI models directly in your AWS account using Amazon SageMaker. Each model runs as a self-hosted inference endpoint within your AWS infrastructure, ensuring data privacy and compliance while leveraging AWS's managed infrastructure for AI inference.

Overview

AWS SageMaker Jumpstart provides a marketplace for deploying pre-trained AI models directly into your AWS environment. Upstage Solar models are available through SageMaker Jumpstart, allowing you to deploy and manage inference endpoints with full control over your data and infrastructure.

All inference happens within your AWS account using SageMaker's managed endpoints. Your data never leaves your environment.

Functional Differences from Upstage Console

• Payload Size: Supports up to 25MB per real-time inference.
• Endpoint Timeout: Default timeout is 60 seconds. This can be adjusted when creating your support ticket in AWS if needed.
• Processing Mode:
  • Real-time Inference: Synchronous endpoints for low-latency, on-demand requests. Supported by all models.
  • Batch Transform: Asynchronous, large-scale processing of inputs stored in S3. Currently supported by Document Parse, Document OCR, and Embedding.
• File Format Support: For Document Parsing and OCR, HWP files have limited compatibility.
  • Versions earlier than 251217.1 no longer support the hwp extension (since December 31, 2025).

Required Prerequisites

Before deploying models on SageMaker, ensure you have the following:

• AWS Account: Active AWS account with appropriate permissions
• SageMaker Permissions: IAM role with SageMaker execution permissions
  • AmazonSageMakerFullAccess (AWS Managed Policy)
  • aws-marketplace:ViewSubscriptions
  • aws-marketplace:Subscribe
  • aws-marketplace:Unsubscribe
• GPU Instance Quota: Sufficient quota for GPU instances
• VPC Configuration: (Optional) VPC setup for enhanced security

These permissions are typically held by account administrators. Contact your AWS admin if you lack these permissions.

Common Installation & Setup

Subscribe Upstage Sagemaker Product

1. Navigate to AWS Marketplace and search for Upstage.
2. Click Product list item on the Upstage listing to start subscribing the product.
3. Click View Purchase Options button on the product page.

4. See the offer details and click Subscribe button.

5. After a few minutes, a subscription success message will appear, and the Launch your software button will be activated. Click it to proceed to the next step.

Create Sagemaker Real-Time Endpoint

You can choose from multiple methods to create your SageMaker endpoint. Select your preferred deployment approach, and the interface will adapt to show relevant instructions and resources for that specific method.

Example using AWS python SDK to create an endpoint:

import boto3
import sagemaker
from sagemaker import ModelPackage, get_execution_role

role = get_execution_role()
sagemaker_session = sagemaker.Session()
sagemaker_runtime = boto3.client("sagemaker-runtime")

model_package_arn = "arn:aws:sagemaker:us-east-1:865070037744:model-package/document-parse-260630-1-cu118-43a9e49553c433509dfec1887d1e9883"
model_name = "Document-Parse-Endpoint"
instance_count = "1"
real_time_inference_instance_type = (
  "ml.g6.2xlarge"
)
# create a deployable model from the model package.
model = ModelPackage(
    role=role, model_package_arn=model_package_arn, sagemaker_session=sagemaker_session
)
endpoint_name = sagemaker.utils.name_from_base(model_name)
print(f"endpoint name: '{endpoint_name}'")
model.deploy(1, real_time_inference_instance_type, endpoint_name=endpoint_name)

Available Capabilities

import boto3
import json
import sseclient

runtime_sm_client = boto3.client("sagemaker-runtime")

# Prepare the request payload
payload = {
    "model": "solar-pro2",
    "messages": [
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "What is the capital of South Korea?"}
    ],
    "temperature": 0.7,
    "max_tokens": 1024
}

# Standard invocation
response = runtime_sm_client.invoke_endpoint(
    EndpointName=endpoint_name,
    ContentType="application/json",
    Body=json.dumps(payload),
)

result = json.loads(response["Body"].read().decode("utf-8"))
print("output:", result["choices"][0]["message"]["content"])

# Streaming invocation
payload["stream"] = True
response = runtime_sm_client.invoke_endpoint_with_response_stream(
    EndpointName=endpoint_name,
    ContentType="application/json",
    Body=json.dumps(payload),
)

# Process stream with SSEClient
client = sseclient.SSEClient((event["PayloadPart"]["Bytes"] for event in response["Body"]))
for event in client.events():
    if event.data == "[DONE]":
        break

    data = json.loads(event.data)
    if data.get("choices"):
        print(data["choices"][0]["delta"].get("content", ""), end="")

Support

For questions about AWS Jumpstart deployment or technical issues, please:

Technical Support

• Visit Upstage Support for technical issues
• Review the Troubleshooting section above for common issues
• Check service status and error messages for diagnostic information
• Contact AWS Support for infrastructure-related questions

Sales and Pricing

• Contact our sales team for enterprise deployment options
• Email contact@upstage.ai for volume discounts and custom solutions