Model-New DeepSeek AI on AWS: When to Use Bedrock or SageMaker
Amazon has embraced controversial new DeepSeek AI tech from a Chinese language startup regardless of considerations about information safety, privateness, compliance, and nationwide safety dangers which have induced many organizations to limit its utilization within the workforce.
This week, Amazon introduced that it has built-in DeepSeek AI into its SageMaker and Bedrock platforms in its AWS cloud. They’re utilized by information scientists and builders to construct, prepare, and deploy machine studying fashions. The corporate just isn’t alone on this, as all three cloud giants, together with Microsoft and Google, have additionally built-in DeepSeek AI into their platforms. Judging from bulletins and associated documentation, nonetheless, Google’s DeepSeek initiatives are extra restricted than these of Amazon and Microsoft (see “Cloud Giants Supply DeepSeek AI, Restricted by Many Orgs, to Devs“).
Amazon first introduced its DeepSeek AI integration in a LinkedIn put up by AWS CEO Matt Garman:
DeepSeek R1 is the most recent basis mannequin to seize the creativeness of the trade. We have at all times been targeted on making it simple to get began with rising and fashionable fashions straight away, and we’re giving clients loads of methods to check out DeepSeek AI.
At this time, clients can run the distilled Llama and Qwen DeepSeek fashions on Amazon SageMaker AI, use the distilled Llama fashions on Amazon Bedrock with Customized Mannequin Import, or prepare DeepSeek fashions with SageMaker by way of Hugging Face.
We have at all times believed that no single mannequin is correct for each use case, and clients can count on every kind of latest choices to emerge sooner or later.
That is why Amazon Bedrock and Amazon SageMaker AI make it seamless for patrons to check and consider the most recent fashions as they emerge and choose the perfect ones primarily based on their distinctive wants.
That was adopted up by a Jan. 30 put up, “DeepSeek-R1 fashions now obtainable on AWS,” together with “DeepSeek-R1 mannequin now obtainable in Amazon Bedrock Market and Amazon SageMaker JumpStart.”
“DeepSeek launched DeepSeek-V3 on December 2024 and subsequently launched DeepSeek-R1, DeepSeek-R1-Zero with 671 billion parameters, and DeepSeek-R1-Distill fashions starting from 1.5-70 billion parameters on Jan. 20, 2025,” mentioned Channy Yun, a principal developer advocate, in that first put up. “They added their vision-based Janus-Professional-7B mannequin on Jan. 27, 2025. The fashions are publicly obtainable and are reportedly 90-95% extra inexpensive and cost-effective than comparable fashions. Per DeepSeek, their mannequin stands out for its reasoning capabilities, achieved by way of progressive coaching methods comparable to reinforcement studying.”
The second put up notes that “As a result of DeepSeek-R1 is an rising mannequin, we advocate deploying this mannequin with guardrails in place,” and explains the right way to use Amazon Bedrock Guardrails to introduce safeguards, forestall dangerous content material, and consider fashions towards key security standards.
Together with emphasizing security, AWS suggested as to which providing to make use of for which use instances.
First, to summarize every:
-
Amazon Bedrock is a completely managed service that simplifies the method of constructing and scaling generative AI functions. It supplies entry to a wide range of high-performing basis fashions (FMs) from main AI corporations by way of a single API1. Customers can experiment with and consider these fashions, customise them with their very own information utilizing methods like fine-tuning and Retrieval Augmented Technology (RAG), and construct brokers that carry out duties utilizing enterprise programs and information sources. Because it’s serverless, there is not any have to handle infrastructure, making it simple to combine and deploy generative AI capabilities into functions utilizing acquainted AWS providers.
[Click on image for larger view.] DeepSeek in Bedrock (supply: AWS). -
Amazon SageMaker is a completely managed service that helps information scientists and builders construct, prepare, and deploy machine studying fashions at scale. It simplifies the machine studying course of by offering instruments for each step of the workflow, from information preparation to mannequin deployment. SageMaker contains built-in Jupyter notebooks for information exploration and preprocessing, built-in algorithms optimized for pace and accuracy, automated mannequin tuning to enhance efficiency, and one-click deployment for seamless scaling. It additionally helps each supervised and unsupervised studying and supplies a safe, managed surroundings for coaching and inferencing fashions. Basically, SageMaker streamlines the complete machine studying lifecycle, making it simpler and sooner to develop high-quality fashions.
[Click on image for larger view.] DeepSeek in SageMaker (supply: AWS).
Bedrock or SageMaker?
“Amazon Bedrock is finest for groups looking for to shortly combine pre-trained basis fashions by way of APIs,” Yun mentioned. “Amazon SageMaker AI is right for organizations that need superior customization, coaching, and deployment, with entry to the underlying infrastructure.”
Yun detailed all of the ins and outs of utilizing every service, once more emphasizing the utility of Amazon Bedrock Guardrails, which builders can use to independently consider consumer inputs and mannequin outputs, having the ability to management the interplay between customers and DeepSeek-R1 with an outlined set of insurance policies by way of filtering undesirable and dangerous content material in GenAI functions.
Nevertheless, he famous, “The DeepSeek-R1 mannequin in Amazon Bedrock Market can solely be used with Bedrock’s ApplyGuardrail API to judge consumer inputs and mannequin responses for customized and third-party FMs obtainable outdoors of Amazon Bedrock.”
He additionally defined utilizing DeepSeek-R1 with Amazon SageMaker JumpStart, described as a ML hub that includes foundational fashions (FMs), built-in algorithms, and prebuilt ML options that may be deployed shortly.
And once more, these security guardrails come into play.
“As like Bedrock Market, you need to use the ApplyGuardrail API within the SageMaker JumpStart to decouple safeguards on your generative AI functions from the DeepSeek-R1 mannequin,” he mentioned. “Now you can use guardrails with out invoking FMs, which opens the door to extra integration of standardized and totally examined enterprise safeguards to your utility move whatever the fashions used.”
Different Choices
Together with the Bedrock and SageMaker steerage, Yun famous alternate options.
- Amazon Bedrock Customized Mannequin Import for the DeepSeek-R1-Distill fashions: This helps customers herald and make the most of personalized fashions alongside current FMs by way of a single serverless API, eliminating the necessity for infrastructure administration. Customers can import DeepSeek-R1-Distill Llama fashions starting from 1.5 to 70 billion parameters. The distillation course of entails coaching smaller, extra environment friendly fashions to duplicate the habits and reasoning patterns of the bigger DeepSeek-R1 mannequin with 671 billion parameters, utilizing it as a trainer mannequin.
- Amazon EC2 Trn1 situations for the DeepSeek-R1-Distill fashions: AWS Deep Studying AMIs (DLAMI) supply tailor-made machine photos for deep studying throughout varied Amazon EC2 situations, from small CPU-only setups to highly effective multi-GPU configurations. Deploying DeepSeek-R1-Distill fashions on AWS Trainium or AWS Inferentia situations supplies optimum price-performance.
Learn the posts for additional info on pricing, information safety and extra.
Concerning the Writer
David Ramel is an editor and author at Converge 360.
