Bedrock vs Sagemaker

Amazon Bedrock is a fully managed service that makes foundation models (FMs) from Amazon and leading AI startups available through an API so you can choose from various FMs to find the model that's best suited for your use case. With the Amazon Bedrock serverless experience, you can quickly get started, easily experiment with FMs, privately customize FMs with your own data, and seamlessly integrate and deploy them into your applications using AWS tools and capabilities. Agents for Amazon Bedrock is a fully managed capability that makes it easier for developers to create generative-AI applications that can deliver up-to-date answers based on proprietary knowledge sources and complete tasks for a wide range of use cases.

Amazon SageMaker is a fully managed service that helps data scientists and developers build, train, and deploy machine learning models at scale. It provides a range of features and tools to simplify the machine learning workflow, from data preprocessing and model training to model deployment and monitoring.

So in a nutshell, Bedrock is the easiest way to build and scale gen​erative AI applications with foundation models (FMs); whereas SageMake is a managed machine learning service in general.

If you speed up the development of generative AI applications using FMs through an API, without managing infrastructure. and if you choose from FMs from AI21 Labs, Anthropic, Stability AI, and Amazon to find the right FM for your use case. Choose Bedrock.

In Generally, you want to build, train, and deploy machine learning models as scale including Jumpstart which get pretrained vision, text, and tabular model, exmple Jupyter notebook, and ene-to-end solutions for common use cases launched with one click. You have to choose SageMaker.

Sagemaker Foundation models allows you more flexibility and choice than bedrock. In Sagemaker infrastrucuture is being provisioned on your behalf and you can train from scratch a new model, pick for a large list of models supported by Jumpstart (including Hugginface), and if the model can be fine-tuned you can do this on sagemaker. Note many foundational models cannot be finetuned and others are only available for Research and cannot be used in commercial applications. Sagemaker will give you the most flexibility but involves more work in setting up and you are charged for endpoints when they are running.

Bedrock is focused on offering an API driven and serverless experience. It offers a curated list of foundational models. You are only charged for what you use (there is no infrastructure costs involved). Only a subset of model on Bedrock will allow fine-tuning but again this will be a very simple api driven process.

Bedrock and Sagemaker offer different characteristics and what is the right choice will depend on your use case, your foundational model choice (or even train your own), the need for fine-tuning for the model, do you have a data science team or are you more developer oriented.

Thank you both . I understand that both of them have an Ability to access and fine tune FMs. What i am trying to understand is what is that i can not do in sagemaker which is not possible in bedrock.

In otherwords what are additional steps i need to do in sagemaker which will be simplified in bedrock

If you could provide those details it will help to create a business case in my company

Some of your questions are answered above, but here is a short summary of differences between Bedrock and Sagemaker Jumpstart.

Bedrock: Provides the easiest way to build and scale generative AI applications with foundation models (FMs)

• API Based: can access models from 4 providers (Amazon, Anthropic, AI21, Stability). Test out in Playground screen for 4 models, which model works well and start using it directly (if you do not need fine-tuning)
• Serverless: It is like Serverless, so you do not need to manage infrastructure, scaling, availability etc. You just subscribe to models from various providers and start using it
• Agility: Time to quickly build a working application is quicker using Bedrock as it is API based
• Cost: Since it is API based, you only pay when you make an API call, so it can be cheaper if your traffic for Inference requests is intermittent and not that high

SageMaker Jumpstart:

• Choose the right model: You have to choose right model for your use case Machine learning (ML) hub with foundation models (public and proprietary), there are SageMaker Jumpstart notebooks available. Since there is a lot of choice of models- open source, proprietary, it can take some time to find the right model
• Deploy FM as SageMaker Endpoint (hosting) end-point: You need to have to right instance for deployment of the chosen model. There is guidance available on instance for a model. If the instance is not in your quota, it has to be requested by raising a service
• Fine-tuning model: If you need to fine-tune FM, then it leverages SageMaker Training jobs
• You have ability to choose SageMaker managed accelerated computing instance
• Manage scaling of compute instance: Based on inference request volume

I hope above gives you some pointers to choose the right service for your use case.