DeepSeek-R1 Models Now Fully Managed on Amazon Bedrock: A Comprehensive Guide

As of January 30, the innovative DeepSeek-R1 models have been made available on Amazon Bedrock. This advancement is accessible through both the Amazon Bedrock Marketplace and the Amazon Bedrock Custom Model Import. Since their release, thousands of users have integrated these models into the Amazon Bedrock ecosystem. Customers have particularly appreciated the robust safety measures and comprehensive tools that facilitate the secure deployment of artificial intelligence (AI). In an exciting development, Amazon has expanded the range of options available for utilizing DeepSeek in Amazon Bedrock, introducing a new serverless solution that simplifies the process.

Amazon Web Services Leads the Way

Amazon Web Services (AWS) has become the first cloud service provider to offer the DeepSeek-R1 model as a fully managed and generally available product. This development allows businesses to accelerate innovation and deliver tangible value without the need to manage complex infrastructure. The DeepSeek-R1 capabilities can now be integrated into generative AI applications via a single API within Amazon Bedrock’s managed services, providing access to an extensive array of features and tools.

Understanding DeepSeek-R1’s Capabilities

According to DeepSeek, the DeepSeek-R1 model is publicly accessible under the MIT license. It is designed with strong capabilities in areas such as reasoning, coding, and natural language understanding. These capabilities enable a range of applications, including intelligent decision support, software development, mathematical problem-solving, scientific analysis, data insights, and comprehensive knowledge management systems.

Implementing DeepSeek-R1: Key Considerations

When implementing AI solutions like DeepSeek-R1, several considerations are vital. These include data privacy requirements, checking for bias in outputs, and monitoring results. Here are some key aspects to keep in mind:

Data Security

Amazon Bedrock provides enterprise-grade security, monitoring, and cost control features crucial for responsibly deploying AI at scale. These features ensure complete control over data, with user inputs and model outputs not shared with any model providers. Key security features include data encryption at rest and in transit, fine-grained access controls, secure connectivity options, and various compliance certifications. You can access these features by default when communicating with the DeepSeek-R1 model in Amazon Bedrock.

Responsible AI

Amazon Bedrock Guardrails allow users to implement safeguards tailored to application requirements and responsible AI policies. These safeguards include content filtering, sensitive information filtering, and customizable security controls to prevent hallucinations using contextual grounding and automated reasoning checks. This means you can control the interaction between users and the DeepSeek-R1 model in Bedrock by filtering undesirable and harmful content in your generative AI applications.

Model Evaluation

Evaluating and comparing models to identify the optimal fit for your use case is straightforward with Amazon Bedrock model evaluation tools. Users can choose automatic evaluation with predefined metrics such as accuracy, robustness, and toxicity or opt for human evaluation workflows for subjective or custom metrics such as relevance, style, and alignment to brand voice. Model evaluation offers built-in curated datasets, or users can bring their own datasets.

Recommendations for DeepSeek-R1 Integration

Integrating Amazon Bedrock Guardrails and utilizing Amazon Bedrock model evaluation features with your DeepSeek-R1 model is strongly recommended to ensure robust protection for your generative AI applications. Additional information can be found in resources like "Protect your DeepSeek model deployments with Amazon Bedrock Guardrails" and "Evaluate the performance of Amazon Bedrock resources."

Getting Started with DeepSeek-R1 in Amazon Bedrock

For those unfamiliar with the DeepSeek-R1 models, you can begin by accessing the Amazon Bedrock console. Navigate to "Model access" under "Bedrock configurations" in the left navigation pane. To use the fully managed DeepSeek-R1 model, request access under "DeepSeek." Once access is granted, you can start using the model in Amazon Bedrock.

To test the DeepSeek-R1 model, choose "Chat/Text" under "Playgrounds" in the left menu pane. Select the model by choosing "DeepSeek" as the category and "DeepSeek-R1" as the model, then apply the selection.

Example Use Case: Financial Decision-Making

Consider a scenario where a family has $5,000 to save for their vacation next year. They face a choice between placing the money in a savings account earning 2% interest annually or a certificate of deposit earning 4% interest annually, with the latter option restricting access to the funds until the vacation. If they anticipate needing $1,000 for emergency expenses during the year, how should they allocate their money between the two options to maximize their vacation fund?

This example demonstrates the model’s ability to handle complex reasoning, yielding precise results.

Accessing DeepSeek-R1 via APIs

To interact with the model using code, users can access it through the AWS Command Line Interface (AWS CLI) and AWS SDK. The model supports both the InvokeModel and Converse API. For instance, using the AWS CLI, the following command allows users to send a text message to the DeepSeek-R1 model:

bash aws bedrock-runtime invoke-model \ --model-id us.deepseek-r1-v1:0 \ --body "{\"messages\":[{\"role\":\"user\",\"content\":[{\"type\":\"text\",\"text\":\"[n\"}]}],max_tokens\":2000,\"temperature\":0.6,\"top_k\":250,\"top_p\":0.9,\"stop_sequences\":[\"\\n\\nHuman:\"]}" \ --cli-binary-format raw-in-base64-out \ --region us-west-2 \ invoke-model-output.txt 

For Python users, the following code snippet illustrates how to send a text message to the DeepSeek-R1 model:

“`python
import boto3
from botocore.exceptions import ClientError

Create a Bedrock Runtime client in the AWS Region you want to use.

client = boto3.client("bedrock-runtime", region_name="us-west-2")

Set the model ID.

model_id = "us.deepseek.r1-v1:0"

Start a conversation with the user message.

user_message = "Describe the purpose of a ‘hello world’ program in one line."
conversation = [
{
"role": "user",
"content": [{"text": user_message}],
}
]

try:

Send the message to the model, using a basic inference configuration.

response = client.converse(<br />
    modelId=model_id,<br />
    messages=conversation,<br />
    inferenceConfig={"maxTokens": 2000, "temperature": 0.6, "topP": 0.9},<br />
)<br />

Extract and print the response text.

response_text = response["output"]["message"]["content"][0]["text"]<br />
print(response_text)<br />

except (ClientError, Exception) as e:
print(f"ERROR: Can’t invoke ‘{model_id}’. Reason: {e}")
exit(1)
“`

Implementing Amazon Bedrock Guardrails

To enable Amazon Bedrock Guardrails on the DeepSeek-R1 model, navigate to "Guardrails" under "Safeguards" in the left navigation pane. Users can create a guardrail by configuring various filters as needed. For example, if a "politics" word filter is set, the guardrails will recognize this word in prompts and block the message.

Testing the guardrails with different inputs helps assess their performance. Users can refine guardrails by setting denied topics, word filters, sensitive information filters, and blocked messaging until they meet specific requirements.

For more information about Amazon Bedrock Guardrails, visit "Stop harmful content in models using Amazon Bedrock Guardrails" in the AWS documentation or explore other blog posts on the AWS Machine Learning Blog channel.

Availability and Next Steps

DeepSeek-R1 is now fully managed in Amazon Bedrock across several AWS Regions, including US East (N. Virginia), US East (Ohio), and US West (Oregon), through cross-Region inference. Users can check the full Region list for future updates. To explore more, visit the DeepSeek in Amazon Bedrock product page and the Amazon Bedrock pricing page.

Try the DeepSeek-R1 model in the Amazon Bedrock console today and share feedback via AWS re:Post for Amazon Bedrock or through usual AWS Support contacts.

For further details, you can explore the official AWS blog post on this topic: DeepSeek-R1 Models Now Available on AWS.

For more Information, Refer to this article.