New Amazon SageMaker AI Innovations Reimagine How Customers Build and Scale Generative AI and Machine Learning Models

Three new Amazon SageMaker HyperPod capabilities, and the addition of popular AI applications from AWS Partners directly in SageMaker, help customers remove undifferentiated heavy lifting across the AI development lifecycle, making it faster and easier to build, train, and deploy models

At AWS re:Invent, Amazon Web Services, Inc. (AWS), an Amazon.com, Inc. company (NASDAQ: AMZN), today announced four new innovations for Amazon SageMaker AI to help customers get started faster with popular publicly available models, maximize training efficiency, lower costs, and use their preferred tools to accelerate generative artificial intelligence (AI) model development. Amazon SageMaker AI is an end-to-end service used by hundreds of thousands of customers to help build, train, and deploy AI models for any use case with fully managed infrastructure, tools, and workflows.

This press release features multimedia. View the full release here: https://www.businesswire.com/news/home/20241204660610/en/

HyperPod AI Partner Apps in SageMaker (Graphic: Business Wire)

• Three powerful new additions to Amazon SageMaker HyperPod make it easier for customers to quickly get started with training some of today’s most popular publicly available models, save weeks of model training time with flexible training plans, and maximize compute resource utilization to reduce costs by up to 40%.
• SageMaker customers can now easily and securely discover, deploy, and use fully managed generative AI and machine learning (ML) development applications from AWS partners, such as Comet, Deepchecks, Fiddler AI, and Lakera, directly in SageMaker, giving them the flexibility to choose the tools that work best for them.
• Articul8, Commonwealth Bank of Australia, Fidelity, Hippocratic AI, Luma AI, NatWest, NinjaTech AI, OpenBabylon, Perplexity, Ping Identity, Salesforce, and Thomson Reuters are among the customers using new SageMaker capabilities to accelerate generative AI model development.

“AWS launched Amazon SageMaker seven years ago to simplify the process of building, training, and deploying AI models, so organizations of all sizes could access and scale their use of AI and ML,” said Dr. Baskar Sridharan, vice president of AI/ML Services and Infrastructure at AWS. “With the rise of generative AI, SageMaker continues to innovate at a rapid pace and has already launched more than 140 capabilities since 2023 to help customers like Intuit, Perplexity, and Rocket Mortgage build foundation models faster. With today’s announcements, we’re offering customers the most performant and cost-efficient model development infrastructure possible to help them accelerate the pace at which they deploy generative AI workloads into production.”

SageMaker HyperPod: The infrastructure of choice to train generative AI models

With the advent of generative AI, the process of building, training, and deploying ML models has become significantly more difficult, requiring deep AI expertise, access to massive amounts of data, and the creation and management of large clusters of compute. Additionally, customers need to develop specialized code to distribute training across the clusters, continuously inspect and optimize their model, and manually fix hardware issues, all while trying to manage timelines and costs. This is why AWS created SageMaker HyperPod, which helps customers efficiently scale generative AI model development across thousands of AI accelerators, reducing time to train foundation models by up to 40%. Leading startups such as Writer, Luma AI, and Perplexity, and large enterprises such as Thomson Reuters and Salesforce, are accelerating model development thanks to SageMaker HyperPod. Amazon also used SageMaker HyperPod to train the new Amazon Nova models, reducing their training costs, improving the performance of their training infrastructure, and saving them months of manual work that would have been spent setting up their cluster and managing the end-to-end process.

Now, even more organizations want to fine-tune popular publicly available models or train their own specialized models to transform their businesses and applications with generative AI. That is why SageMaker HyperPod continues to innovate to make it easier, faster, and more cost-efficient for customers to build, train, and deploy these models at scale with new innovations, including:

• New recipes help customers get started faster: Many customers want to take advantage of popular publicly available models, like Llama and Mistral, that can be customized to a specific use case using their organization’s data. However, it can take weeks of iterative testing to optimize training performance, including experimenting with different algorithms, carefully refining parameters, observing the impact on training, debugging issues, and benchmarking performance. To help customers get started in minutes, SageMaker HyperPod now provides access to more than 30 curated model training recipes for some of today’s most popular publicly available models, including Llama 3.2 90B, Llama 3.1 405B, and Mistral 8x22B. These recipes greatly simplify the process of getting started for customers, automatically loading training datasets, applying distributed training techniques, and configuring the system for efficient checkpointing and recovery from infrastructure failures. This empowers customers of all skill levels to achieve improved price performance for model training on AWS infrastructure from the start, eliminating weeks of iterative evaluation and testing. Customers can browse available training recipes via the SageMaker GitHub repository, adjust parameters to suit their customization needs, and deploy within minutes. Additionally, with a simple one-line edit, customers can seamlessly switch between GPU- or Trainium-based instances to further optimize price performance. Researchers at Salesforce were looking for ways to quickly get started with foundation model training and fine-tuning, without having to worry about the infrastructure, or spending weeks optimizing their training stack for each new model. With Amazon SageMaker HyperPod recipes, they can conduct rapid prototyping when customizing foundation models. Now, Salesforce’s AI Research teams are able to get started in minutes with a variety of pre-training and fine-tuning recipes, and can operationalize foundation models with high performance.

• Flexible training plans make it easy to meet training timelines and budgets: While infrastructure innovations help drive down costs and allow customers to train models more efficiently, customers must still plan and manage the compute capacity required to complete their training tasks on time and within budget. That is why AWS is launching flexible training plans for SageMaker HyperPod. In a few clicks, customers can specify their budget, desired completion date, and maximum amount of compute resources they need. SageMaker HyperPod then automatically reserves capacity, sets up clusters, and creates model training jobs, saving teams weeks of model training time. This reduces the uncertainty customers face when trying to acquire large clusters of compute to complete model development tasks. In cases where the proposed training plan does not meet the specified time, budget, or compute requirements, SageMaker HyperPod suggests alternate plans, like extending the date range, adding more compute, or conducting the training in a different AWS Region, as the next best option. Once the plan is approved, SageMaker automatically provisions the infrastructure and runs the training jobs. SageMaker uses Amazon Elastic Compute Cloud (EC2) Capacity Blocks to reserve the right amount of accelerated compute instances needed to complete the training job in time. By efficiently pausing and resuming training jobs based on when those capacity blocks are available, SageMaker HyperPod helps make sure customers have access to the compute resources they need to complete the job on time, all without manual intervention. Hippocratic AI develops safety-focused large language models (LLMs) for healthcare. To train several of their models, Hippocratic AI used SageMaker HyperPod flexible training plans to gain access to accelerated compute resources they needed to complete their training tasks on time. This helped them accelerate their model training speed by 4x and more efficiently scale their solution to accommodate hundreds of use cases. Developers and data scientists at OpenBabylon, an AI company that customizes LLMs for underrepresented languages, have has been using SageMaker HyperPod flexible training plans to streamline their access to GPU resources to run large scale experiments. Using SageMaker HyperPod, they conducted 100 large scale model training experiments that allowed them to build a model that achieved state-of-the-art results in English-to-Ukrainian translation. Thanks to SageMaker HyperPod, OpenBabylon was able to achieve this breakthrough on time while effectively managing costs.

• Task governance maximizes accelerator utilization: Increasingly, organizations are provisioning large amounts of accelerated compute capacity for model training. These compute resources involved are expensive and limited, so customers need a way to govern usage to ensure their compute resources are prioritized for the most critical model development tasks, including avoiding any wastage or underutilization. Without proper controls over task prioritization and resource allocation, some projects end up stalling due to lack of resources, while others leave resources underutilized. This creates a significant burden for administrators, who must constantly re-plan resource allocation, while data scientists struggle to make progress. This prevents organizations from bringing AI innovations to market quickly and leads to cost overruns. With SageMaker HyperPod task governance, customers can maximize accelerator utilization for model training, fine-tuning, and inference, reducing model development costs by up to 40%. With a few clicks, customers can easily define priorities for different tasks and set up limits for how many compute resources each team or project can use. Once customers set limits across different teams and projects, SageMaker HyperPod will allocate the relevant resources, automatically managing the task queue to ensure the most critical work is prioritized. For example, if a customer urgently needs more compute for an inference task powering a customer-facing service, but all compute resources are in use, SageMaker HyperPod will automatically free up underutilized compute resources, or those assigned to non-urgent tasks, to make sure the urgent inference task gets the resources it needs. When this happens, SageMaker HyperPod automatically pauses the non-urgent tasks, saves the checkpoint so that all completed work is intact, and automatically resumes the task from the last-saved checkpoint once more resources are available, ensuring customers make the most of their compute. As a fast-growing startup that helps enterprises build their own generative AI applications, Articul8 AI needs to constantly optimize its compute environment to allocate its resources as efficiently as possible. Using the new task governance capability in SageMaker HyperPod, the company has seen a significant improvement in GPU utilization, resulting in reduced idle time and accelerated end-to-end model development. The ability to automatically shift resources to high-priority tasks has increased the team's productivity, allowing them to bring new generative AI innovations to market faster.

Accelerate model development and deployment using popular AI apps from AWS Partners within SageMaker

Many customers use best-in-class generative AI and ML model development tools alongside SageMaker AI to conduct specialized tasks, like tracking and managing experiments, evaluating model quality, monitoring performance, and securing an AI application. However, integrating popular AI applications into a team’s workflow is a time-consuming, multi-step process. This includes searching for the right solution, performing security and compliance evaluations, monitoring data access across multiple tools, provisioning and managing the necessary infrastructure, building data integrations, and verifying adherence to governance requirements. Now, AWS is making it easier for customers to combine the power of specialized AI apps with the managed capabilities and security of Amazon SageMaker. This new capability removes the friction and heavy lifting for customers by making it easy to discover, deploy, and use best-in-class generative AI and ML development applications from leading partners, including Comet, Deepchecks, Fiddler, and Lakera Guard, directly within SageMaker.

SageMaker is the first service to offer a curated set of fully managed and secure partner applications for a range of generative AI and ML development tasks. This gives customers even greater flexibility and control when building, training, and deploying models, while reducing the time to onboard AI apps from months to weeks. Each partner app is fully managed by SageMaker, so customers do not have to worry about setting up the application or continuously monitoring to ensure there is enough capacity. By making these applications accessible directly within SageMaker, customers no longer need to move data out of their secure AWS environment, and they can reduce the time spent toggling between interfaces. To get started, customers simply browse the Amazon SageMaker Partner AI apps catalog, learning about the features, user experience, and pricing of the apps they want to use. They can then easily select and deploy the applications, managing access for the entire team using AWS Identity and Access Management (IAM).

Amazon SageMaker also plays a pivotal role in the development and operation of Ping Identity’s homegrown AI and ML infrastructure. With partner AI apps in SageMaker, Ping Identity will be able to deliver faster, more effective ML-powered functionality to their customers as a private, fully managed service, supporting their strict security and privacy requirements while reducing operational overhead.

All of the new SageMaker innovations are generally available to customers today.

To learn more, visit:

• The AWS Blog for details on today’s announcements: HyperPod flexible training plans, HyperPod task governance, and AI apps from partners in SageMaker.
• The Amazon SageMaker AI page to learn more about the capabilities.
• The Amazon SageMaker customer page to learn how companies are using Amazon Bedrock.
• The AWS re:Invent page for more details on everything happening at AWS re:Invent.

About Amazon Web Services

Since 2006, Amazon Web Services has been the world’s most comprehensive and broadly adopted cloud. AWS has been continually expanding its services to support virtually any workload, and it now has more than 240 fully featured services for compute, storage, databases, networking, analytics, machine learning and artificial intelligence (AI), Internet of Things (IoT), mobile, security, hybrid, media, and application development, deployment, and management from 108 Availability Zones within 34 geographic regions, with announced plans for 18 more Availability Zones and six more AWS Regions in Mexico, New Zealand, the Kingdom of Saudi Arabia, Taiwan, Thailand, and the AWS European Sovereign Cloud. Millions of customers—including the fastest-growing startups, largest enterprises, and leading government agencies—trust AWS to power their infrastructure, become more agile, and lower costs. To learn more about AWS, visit aws.amazon.com.

About Amazon

Amazon is guided by four principles: customer obsession rather than competitor focus, passion for invention, commitment to operational excellence, and long-term thinking. Amazon strives to be Earth’s Most Customer-Centric Company, Earth’s Best Employer, and Earth’s Safest Place to Work. Customer reviews, 1-Click shopping, personalized recommendations, Prime, Fulfillment by Amazon, AWS, Kindle Direct Publishing, Kindle, Career Choice, Fire tablets, Fire TV, Amazon Echo, Alexa, Just Walk Out technology, Amazon Studios, and The Climate Pledge are some of the things pioneered by Amazon. For more information, visit amazon.com/about and follow @AmazonNews.

View source version on businesswire.com: https://www.businesswire.com/news/home/20241204660610/en/

Amazon.com, Inc. Media Hotline [e-mail] www.amazon.com/pr