|
|
During this past, AWS Re: Inventory, CEO of Amazon Andy Jassy, shared a valuable lesson in Amazon’s own experience with the development of almost 1,000 generative AI applications throughout the company. From this extensive range of deployment AI, JASSY offered three key observations that formed the implementation of AI Aicon.
The first is that as you get to the scale in generative AI applications, the cost of calculation really matters. People are very hungry for better price performance. The second is actually one who is difficult to create a really good generative AI application. The third is the diversity of used models that we gave our builders freedom to choose what they want to do. We are not surprised, because we are constantly learning the same lesson over and over again and again, which is that you are going to one instrument to rule the world.
As Andy stressed, the wide and deep series of Amazon models provided allows customers to choose pre -scraper skills that best serve their unique needs. By carefully monitoring both customers’ needs and technological progress, the AWS regularly expands our curatorial selection of models to include promising new models along with established popular industrial regulations. This nagoing expansion of high -performance and differentiated model offers help customers remain in the foreground of the AI innovation.
This leads us to the Chinese startup AI Deepseek. Deepseek launched Deepseek-V3 in December 2024 and subsequently released Deepseek-R1, Deepseek-R1-Vero with 671 billion parameters and Deepseek-R1-Distill Models in the range of 1.5 to 70 billion parameters of January 2025. -7b 27 January 2025. Models are publicly available and reportedly are 90-95% more affordable and cost -effective than comparable models. On Deepseek, their model excels in its ability to think through innovative technical training such as learning.
You can now deploy Deepseek-R1 models in Amazon Bedrock and Amazon Sagemaker AI. Amazon Bedrock is best for teams that try to quickly integrate pre -trained Models of the API Foundation. Amazon Sagemaker AI is ideal for organizations that want advanced adaptation, training and deployment, with access to the underlying infrastructure. In addition, you can also use the AWS Trainium and AWS Infentia to deploy DeepSeek-R1-Distill frequently efficiently via the Amazon Elastic Compute Cloud (AMAZON EC2) or Amazon Sagemaker AI.
With AWS you can use Deepseek-R1 models to create, experiment and responsible scale of your generative AI ideas using this powerful, cost-effective model with minimal investment in infrastructure. You can also confidently manage generative innovations AI by building AWS Services, which are only for security. We strongly recommend the integration of your Deepseek-R1 with Amazon Bedrock Guardrails to add a layer of protection for your AI generative applications that Amazon Bedrock and Amazon Sagemaker AI can use.
Today you can choose how to deploy Deepseek-R1 models on AWS today: 1/ Amazon Bedrock Marketplace for Deepseek-R1,, 2/ Amazon Sagemaker Jumstart for DeepSeek-R1,, 3/ Amazon Bedrock Custom Model Import for DeepSeek-R1-Distilland 4/ Amazon EC2 TRN1 Instance for DeepSeek-R1-Distill.
Let me go through different ways to start with Deepseek-R1 on AWS. Whether you create your first AI APO or scaling of existing solutions, these methods provide flexible starting points based on your team’s expertise and requirements.
1. Model Deepseek-R1 on the Amazon Bedrock Marketplace market
Amazon Bedrock Marketplace offers more than 100 popular, developing and specialized FMSs along with the current selection of models leading in Amazon Bedrock. You can easily discover the models in one catalog, subscribe to the model, and then put the model on the right endpoints.
To access the Deepseek-R1 market on the Amazon Bedrock Marketplace market, go to the Amazon Bedrock Console console and select Catalog model under Endowment models section. Deepseek can be quickly found by searching or filtering by model providers.
After logging out on the model page, the details included the model and implementation manual capabilities, you can directly deploy the model by providing the end point name, selecting the number of instances, and selecting the instance type.
You can also configure advanced options that allow you to customize the Deepseek-R1 security and infrastructure settings, including the VPC network, the service role and encryption settings. As for production deployment, you should check these settings to comply with your organization’s security and compliance.
With Amazon Bedrock Guardrails, you can independently evaluate user inputs and model outputs. You can control the interaction between users and Deepseek-R1 with a defined set of filtering of undesirable and harmful content in generative AI applications. The Deepseek-R1 model on the Amazon Bedrock Marketplace market can only be used with the Bedrock Appormuardrail API to evaluate user inputs and models for their own and third FMS outside Amazon Bedrock. To learn more, read security measures implementing security measures with Amazon Bedrock Guardrails.
Amazon Bedrock railing can also be integrated with other subsoil tools include Amazon agents and Amazon Bedrock Nowledge Bases for building more safer and safer AI generation applications in accordance with AI principles. If you want to know more, visit the AWS responsible AI.
When using DeepSeek-R1 with Bedrock’s InvokeModel Please use the Deepseek Cat template and the template template for optimal results. For the exam, <｜begin▁of▁sentence｜><｜User｜>content for inference<｜Assistant｜>.
Get this step-by-bed wizard about how to deploy Deepseek-R1 on the Amazon Bedrock market. If you want to learn more, visit the deployment of models on Amazon Bedrock Marketplace.
2. Model Deepseek-R1 in Amazon Sagemaker Jumpstart
Amazon Sagemaker Jumstart is a machine learning hub (ML) with FMS, built -in algorithms and pre -created ml solutions that you can use just a few clicks. To deploy Deepseek-R1 in Sagemaker Jumstart, you can discover Deepseek-R1 in Sagemaker Unified Studio, Sagemaker Studio, SageMaker AI Console or programmaker through Sagemaker Python SDK.
In the Amazon Sagemaker AI, Open Sagemaker Unified Studio or Sagemaker Studio. In case of Sagemaker studio choose Jumstart and look for “DeepSeek-R1“IN All public models page.
You can select the model and select the deployment and create an endpoint with the default settings. When the end point comes InsertYou can perform infections by sending the requirements for its endpoint.
You can deduce performance performance and controls of ML operations with Amazon Sagemaker AI features such as Amazon Sagemaker Pipelines, Amazon Sagemaker Debugger or container protocols. The model is deployed in the AWS Secure environment and under the control of the virtual private cloud (VPC) that helps support data security.
As well as the basin market, you can use ApplyGuardrail API in the Sagemaker Jumsstart to clean the warranty for your generative AI from Deepseek-R1. Now you can use the handrail without induction of FMS that open the door to greater integration of standardized and thoroughly tested business guarantees on your application flow of the appearance of the models used.
Let’s take a step for this guide to Smal on how to deploy Deepseek-R1 in Amazon Sagemaker Jumstart. You want to learn more, visit the Discover Sagemaker Jumstart models in the Sagemaker Unified Studio or deploy models Sagemaker Jumstart in Sagemaker Studio.
3. Models DeepSeek-R1-Destill using your own Amazon Bedrock model
Amazon Bedrock Custom Model Import provides the ability to import and use your adapted models along with FMS exvision through a single without server united API without having to manage basic infrastructure. With your own Amazon Bedrock Import model, you can import DeepSeek-R1-Distill LLAMA models ranging from 1.5 to 70 billion parameters. As I emphasized in my blog post about distillation of the Amazon Bedrock, the process of distillation includes training smaller and more efficient models that mimic the behavior and reasoning of the larger Deepseek-R1 with 671 billion parameters using a teacher model.
After storing these publicly available models in Amazon Simple Storage (Amazon S3) or in the Amazon Sagemaker register, go to Imported models under Endowment models In the Amazon Bedrock Console and import and put them in fully managed and without server via Amazon Bedrock. This approach eliminates the need for infrastructure management while providing security and scalabibility at the company level.
Take this step-to-make guide to deploy Deepseek-R1 models using the Amazon Bedrock’s own model. If you want to learn more, visit the import adapted to the Amazon Bedrock.
4. Models Deepseek-R1-Distill using AWS Trainium and Aws Invorentia
The AWS Deep Learning Friends (PLIMI) provides adapted images that you can use for deep learning in various Amazon EC2 instances, from a small example for CPUs to the latest high-performance instances of more-GPU. You can deploy Deepseek-R1-Distill on AWS Trainuim1 or AWS Inferrentia2 instances to get the best price performance.
You want to start, go to the Amazon EC2 and run and trn1.32xlarge Installation of the EC2 with Multi Framework Neuron called the Dlími Deep Learning Ami Neuron (Ubuntu 22.04).
When you connect to the EC2 instance, install the VLM, an open source tool that serves large language models (LLMS) and download the DeepSeek-R1-Distill from hugging your face. You can use the model with VLLM and trigger a model server.
To learn more, register for this guide step by step on how to deploy the Deepseek-R1-Distill LLAMA models on the AWS INVORENTIA and Trainium.
You can also visit the DeepSeek-R1-Distill-Lalma-8B or Deepseek-Ai/Deepseek-R1-Distill-Lalma-70B for hugging face. Choose Deployment and then Amazon Sagemaker. FROM AWS INFLOENTIA and Trainium Tab, copy an example of the code for deploying Lama Deepseek-R1-Disclusion Lama.
Sale of Deepseek-R1, various wizards of its deployment for Amazon EC2 and Amazon Elastic Kubernetes Service (Amazon Eks). Here is another material for you to see:
What to know
Here are some important things.
- Prices -For publicly available models such as Deepseek-R1, you only have the price of infrastructure based on the inference clock you choose for Amazon Bedrock Mareplace, Amazon Sagemaker Jumstart and Amazon EC2. To import your own model, you are only charged for the model’s inference, based on the number of copies of your own model is active, charged in 5 -minute windows. If you want to learn more, check out Amazon Bedrock prices, Amazon Sagemaker AI Pricking and Amazon EC2 Pricking Pages.
- Data security -Amazon Bedrock and Amazon Sagemaker You can use Enterprise security features to help you provide your data and applications secure and private. This means that your data is not shared with model providers and is not used to improve models. This application on all models -Proprietary and publicly available DeepSeek -R1 models on Amazon Bedrock and Amazon Sagemaker. You want to learn more, visit Amazon Bedrock Security and protection and security in Amazon Sagemaker AI.
Now available
Deepseek-R1 is generally available on Amazon Bedrock Marketplace and Amazon Sagemaker Jumstart. You can also use Deepseek-R1-Distill models using your own Amazon Bedrock Import and Amazon EC2 instance with AWS Traid and Inferrentia Chips.
Try DeepSeek-R1 models in Amazon Bedrock Console, Amazon Sagemaker AI Console and Amazon EC2 Console and send feedback to AWS Re: Post for Amazon Bedrock and AWS Re: Post for Sagemaker AI or through your usual AWS contacts contacts.
– Channels