Today, we are excited to announce that DeepSeek R1 distilled Llama and Qwen models are available through Amazon Bedrock Marketplace and Amazon SageMaker JumpStart. With this launch, you can now release DeepSeek AI's first-generation frontier model, DeepSeek-R1, in addition to the distilled variations varying from 1.5 to 70 billion parameters to construct, experiment, and properly scale your generative AI concepts on AWS.
In this post, we demonstrate how to begin with DeepSeek-R1 on Amazon Bedrock Marketplace and SageMaker JumpStart. You can follow similar actions to release the distilled variations of the designs too.
Overview of DeepSeek-R1
DeepSeek-R1 is a big language model (LLM) established by DeepSeek AI that utilizes support learning to enhance reasoning capabilities through a multi-stage training process from a DeepSeek-V3-Base structure. An essential differentiating feature is its reinforcement knowing (RL) step, which was utilized to improve the design's actions beyond the standard pre-training and tweak procedure. By integrating RL, DeepSeek-R1 can adapt better to user feedback and objectives, ultimately boosting both relevance and clarity. In addition, DeepSeek-R1 uses a chain-of-thought (CoT) technique, suggesting it's equipped to break down complicated inquiries and factor through them in a detailed manner. This directed reasoning process allows the design to produce more precise, transparent, and detailed answers. This design combines RL-based fine-tuning with CoT capabilities, aiming to generate structured actions while concentrating on interpretability and user interaction. With its comprehensive capabilities DeepSeek-R1 has captured the industry's attention as a versatile text-generation design that can be integrated into numerous workflows such as agents, rational thinking and information analysis tasks.
DeepSeek-R1 utilizes a Mixture of Experts (MoE) architecture and is 671 billion criteria in size. The MoE architecture enables activation of 37 billion criteria, making it possible for efficient reasoning by routing questions to the most relevant expert "clusters." This technique enables the model to concentrate on various issue domains while maintaining total efficiency. DeepSeek-R1 needs at least 800 GB of HBM memory in FP8 format for reasoning. In this post, we will use an ml.p5e.48 xlarge circumstances to release the model. ml.p5e.48 xlarge comes with 8 Nvidia H200 GPUs providing 1128 GB of GPU memory.
DeepSeek-R1 distilled models bring the reasoning capabilities of the main R1 design to more effective architectures based on popular open models like Qwen (1.5 B, 7B, 14B, and 32B) and Llama (8B and 70B). Distillation describes a procedure of training smaller, more efficient designs to imitate the habits and reasoning patterns of the larger DeepSeek-R1 design, utilizing it as a teacher design.
You can deploy DeepSeek-R1 design either through SageMaker JumpStart or Bedrock Marketplace. Because DeepSeek-R1 is an emerging model, we recommend deploying this model with guardrails in location. In this blog, we will utilize Amazon Bedrock Guardrails to introduce safeguards, prevent hazardous material, and examine models against crucial security criteria. At the time of composing this blog, for DeepSeek-R1 implementations on SageMaker JumpStart and Bedrock Marketplace, Bedrock Guardrails supports just the ApplyGuardrail API. You can produce numerous guardrails tailored to different use cases and apply them to the DeepSeek-R1 design, improving user experiences and standardizing safety controls across your generative AI applications.
Prerequisites
To deploy the DeepSeek-R1 model, you require access to an ml.p5e instance. To examine if you have quotas for P5e, open the Service Quotas console and under AWS Services, pick Amazon SageMaker, and verify you're using ml.p5e.48 xlarge for endpoint usage. Make certain that you have at least one ml.P5e.48 xlarge instance in the AWS Region you are deploying. To request a limit boost, produce a limit boost request and connect to your account team.
Because you will be releasing this model with Amazon Bedrock Guardrails, make certain you have the proper AWS Identity and Gain Access To Management (IAM) authorizations to use Amazon Bedrock Guardrails. For instructions, see Set up approvals to use guardrails for material filtering.
Implementing guardrails with the ApplyGuardrail API
Amazon Bedrock Guardrails enables you to present safeguards, prevent harmful material, and assess models against essential security criteria. You can implement security steps for the DeepSeek-R1 design utilizing the Amazon Bedrock ApplyGuardrail API. This permits you to apply guardrails to examine user inputs and model actions deployed on Amazon Bedrock Marketplace and SageMaker JumpStart. You can create a guardrail utilizing the Amazon Bedrock console or the API. For the example code to create the guardrail, see the GitHub repo.
The basic flow involves the following steps: First, the system gets an input for the design. This input is then processed through the ApplyGuardrail API. If the input passes the guardrail check, it's sent out to the design for inference. After receiving the design's output, another guardrail check is applied. If the output passes this final check, it's returned as the last outcome. However, if either the input or output is intervened by the guardrail, a message is returned indicating the nature of the intervention and whether it happened at the input or output stage. The examples showcased in the following areas demonstrate reasoning utilizing this API.
Deploy DeepSeek-R1 in Amazon Bedrock Marketplace
Amazon Bedrock Marketplace gives you access to over 100 popular, emerging, and specialized foundation models (FMs) through Amazon Bedrock. To gain access to DeepSeek-R1 in Amazon Bedrock, complete the following steps:
1. On the Amazon Bedrock console, pick Model brochure under Foundation models in the navigation pane.
At the time of composing this post, you can utilize the InvokeModel API to invoke the design. It does not support Converse APIs and other Amazon Bedrock tooling.
2. Filter for DeepSeek as a service provider and pick the DeepSeek-R1 model.
The design detail page provides important details about the design's abilities, rates structure, and application guidelines. You can find detailed use instructions, including sample API calls and code bits for integration. The model supports different text generation tasks, including material creation, code generation, and question answering, utilizing its support discovering optimization and CoT thinking capabilities.
The page also consists of implementation choices and licensing details to assist you start with DeepSeek-R1 in your applications.
3. To start utilizing DeepSeek-R1, choose Deploy.
You will be triggered to set up the deployment details for DeepSeek-R1. The design ID will be pre-populated.
4. For Endpoint name, get in an endpoint name (in between 1-50 alphanumeric characters).
5. For Variety of circumstances, get in a variety of circumstances (in between 1-100).
6. For Instance type, pick your circumstances type. For ideal efficiency with DeepSeek-R1, a GPU-based instance type like ml.p5e.48 xlarge is suggested.
Optionally, you can configure sophisticated security and infrastructure settings, including virtual personal cloud (VPC) networking, service function permissions, and file encryption settings. For a lot of utilize cases, the default settings will work well. However, for production deployments, you may wish to review these settings to line up with your organization's security and compliance requirements.
7. Choose Deploy to begin using the model.
When the release is total, you can check DeepSeek-R1's capabilities straight in the Amazon Bedrock playground.
8. Choose Open in play area to access an interactive interface where you can explore various triggers and change model criteria like temperature level and optimum length.
When using R1 with Bedrock's InvokeModel and Playground Console, use DeepSeek's chat template for optimum results. For example, material for inference.
This is an excellent method to explore the model's reasoning and text generation abilities before integrating it into your applications. The play ground supplies immediate feedback, assisting you understand how the model responds to numerous inputs and letting you tweak your triggers for ideal results.
You can rapidly test the model in the play ground through the UI. However, to conjure up the deployed model programmatically with any Amazon Bedrock APIs, you need to get the endpoint ARN.
Run reasoning using guardrails with the released DeepSeek-R1 endpoint
The following code example shows how to carry out reasoning using a deployed DeepSeek-R1 design through Amazon Bedrock utilizing the invoke_model and ApplyGuardrail API. You can create a guardrail using the Amazon Bedrock console or the API. For the example code to develop the guardrail, see the GitHub repo. After you have created the guardrail, utilize the following code to implement guardrails. The script initializes the bedrock_runtime client, configures reasoning specifications, and sends a request to produce text based on a user timely.
Deploy DeepSeek-R1 with SageMaker JumpStart
SageMaker JumpStart is an artificial intelligence (ML) center with FMs, integrated algorithms, and higgledy-piggledy.xyz prebuilt ML services that you can deploy with just a few clicks. With SageMaker JumpStart, you can tailor pre-trained models to your usage case, with your information, and deploy them into production using either the UI or SDK.
Deploying DeepSeek-R1 model through SageMaker JumpStart provides two practical methods: using the instinctive SageMaker JumpStart UI or implementing programmatically through the SageMaker Python SDK. Let's explore both techniques to help you pick the technique that best fits your needs.
Deploy DeepSeek-R1 through SageMaker JumpStart UI
Complete the following actions to release DeepSeek-R1 utilizing SageMaker JumpStart:
1. On the SageMaker console, select Studio in the navigation pane.
2. First-time users will be prompted to create a domain.
3. On the SageMaker Studio console, choose JumpStart in the navigation pane.
The design web browser shows available designs, with details like the company name and design capabilities.
4. Search for DeepSeek-R1 to see the DeepSeek-R1 model card.
Each model card reveals crucial details, consisting of:
- Model name
- Provider name
- Task category (for example, Text Generation).
Bedrock Ready badge (if suitable), indicating that this design can be registered with Amazon Bedrock, enabling you to use Amazon Bedrock APIs to invoke the model
5. Choose the design card to see the design details page.
The model page includes the following details:
- The model name and provider details. Deploy button to deploy the model. About and Notebooks tabs with detailed details
The About tab consists of crucial details, such as:
- Model description. - License details.
- Technical specifications.
- Usage guidelines
Before you release the model, it's suggested to examine the model details and license terms to verify compatibility with your usage case.
6. Choose Deploy to proceed with implementation.
7. For Endpoint name, utilize the instantly produced name or develop a custom one.
- For Instance type ¸ choose an instance type (default: ml.p5e.48 xlarge).
- For Initial circumstances count, go into the variety of instances (default: 1). Selecting suitable circumstances types and counts is vital for expense and efficiency optimization. Monitor your implementation to adjust these settings as needed.Under Inference type, Real-time reasoning is selected by default. This is optimized for sustained traffic and low latency.
- Review all setups for accuracy. For this model, we strongly suggest adhering to SageMaker JumpStart default settings and making certain that network isolation remains in place.
- Choose Deploy to deploy the design.
The deployment process can take several minutes to finish.
When implementation is total, your endpoint status will alter to InService. At this point, the design is prepared to accept reasoning requests through the endpoint. You can keep an eye on the deployment progress on the SageMaker console Endpoints page, which will show pertinent metrics and status details. When the deployment is total, you can invoke the model utilizing a SageMaker runtime client and incorporate it with your applications.
Deploy DeepSeek-R1 utilizing the SageMaker Python SDK
To get going with DeepSeek-R1 utilizing the SageMaker Python SDK, you will need to set up the SageMaker Python SDK and make certain you have the needed AWS permissions and environment setup. The following is a detailed code example that shows how to release and utilize DeepSeek-R1 for inference programmatically. The code for releasing the model is supplied in the Github here. You can clone the notebook and run from SageMaker Studio.
You can run additional requests against the predictor:
Implement guardrails and run reasoning with your SageMaker JumpStart predictor
Similar to Amazon Bedrock, you can also utilize the ApplyGuardrail API with your SageMaker JumpStart predictor. You can develop a guardrail using the Amazon Bedrock console or the API, and implement it as shown in the following code:
Tidy up
To avoid unwanted charges, complete the actions in this section to tidy up your resources.
Delete the Amazon Bedrock Marketplace release
If you released the model using Amazon Bedrock Marketplace, total the following actions:
1. On the Amazon Bedrock console, under Foundation designs in the navigation pane, choose Marketplace deployments. - In the Managed deployments area, find the endpoint you wish to delete.
- Select the endpoint, and on the Actions menu, choose Delete.
- Verify the endpoint details to make certain you're deleting the appropriate deployment: 1. Endpoint name.
- Model name.
- Endpoint status
Delete the SageMaker JumpStart predictor
The SageMaker JumpStart model you released will sustain costs if you leave it running. Use the following code to erase the endpoint if you want to stop sustaining charges. For more details, see Delete Endpoints and Resources.
Conclusion
In this post, we explored how you can access and release the DeepSeek-R1 design using Bedrock Marketplace and SageMaker JumpStart. Visit SageMaker JumpStart in SageMaker Studio or Amazon Bedrock Marketplace now to begin. For more details, refer to Use Amazon Bedrock tooling with Amazon SageMaker JumpStart designs, SageMaker JumpStart pretrained models, Amazon SageMaker JumpStart Foundation Models, Amazon Bedrock Marketplace, and Getting going with Amazon SageMaker JumpStart.
About the Authors
Vivek Gangasani is a Lead Specialist Solutions Architect for Inference at AWS. He assists emerging generative AI companies construct ingenious options utilizing AWS services and accelerated compute. Currently, he is focused on establishing strategies for fine-tuning and optimizing the reasoning efficiency of large language models. In his spare time, Vivek delights in treking, seeing films, and attempting different foods.
Niithiyn Vijeaswaran is a Generative AI Specialist Solutions Architect with the Third-Party Model Science team at AWS. His location of focus is AWS AI accelerators (AWS Neuron). He holds a Bachelor's degree in Computer Science and Bioinformatics.
Jonathan Evans is a Professional Solutions Architect working on generative AI with the Third-Party Model Science group at AWS.
Banu Nagasundaram leads item, engineering, and tactical collaborations for Amazon SageMaker JumpStart, SageMaker's artificial intelligence and generative AI hub. She is enthusiastic about developing services that assist clients accelerate their AI journey and unlock service value.