What is Amazon Bedrock? You may simply call it AWS bedrock, which is a fully managed service by using it you can scale your generative AI application with FM (foundation models).
These FM are from leading AI start-ups and Amazon. You can use /access it from the AWS management console and GUI, also you can use it via unified API to integrate it into your applications.
Based on your use case you get to choose from a wide range of FMs on the AWS platform, by visiting the FM development company portal you can find more about the model that is best suited for your use case/need.
Now you use these (FM) foundational models and the (LLMs)large language models built by other companies, this becomes the foundation for your new application.
When using Amazon bedrock, you don’t need to manage any infrastructure, it is a completely managed service, and you will experience the power of serverless technology. You can start quickly and integrate it into your application using AWS single API.
Differentiating Factors – Amazon Bedrock
As of now, Amazon has not announced its own consumer-facing generative AI applications. So, the core focus is to provide the platform.
Amazon on the other had provided access to its own AI model, which is called Titan. The developers who are interested in building applications using AWS bedrock models, need to consume AWS Bedrock service.
Amazon also invested in developing chips designed for training and inferences.
The training chip is AWS Trainium and the one used for inferences is AWS Inferentia accelerators, both are used in DL and ML workloads. This saves money for the company required for large computing resources.
So, this could be less expensive for many developers to use AWS cloud than its competition. The other AWS bedrock competitors, like Microsoft and Google, also have their own chip designs for AI training.
Video Credit Amazon Web Services
What Amazon Bedrock Developers Get
Get to choose from different FMs:
As a developer, you get access to Amazon Titan models and other access to leading models including AI21 Labs’ Jurassic, Cohere’s Command, and more. Amazon will keep adding new partners in this space.
Get to experiment with different FMs
By using interactive playgrounds and other options it is easy and quick to experiment with different FMs.
Get To Evaluate FMs and select the best one for your needs
As you can do actual model evaluation using Bedrock it helps to do evaluations and select the best suitable FMs for a specific use case.
Get To customize FMs with your data
You can use your customized FM for your use case very easily, as a developer there is no much of coding required to do so.
Get to use Unified/Single API
You as a developer understand one single API and use the same API to interact with different models, this requires minimal code changes.
AWS Bedrock Architecture
Let’s look at simple a AWS bedrock architecture diagram, this will help you to visualize what are the different components and how they interact.
How to Access Amazon Bedrock?
To begin, log in to the AWS Management Console and select Northern Virginia (us-east-1 Region) from the Region dropdown.
Using the search bar, search for “Bedrock” and select “Amazon Bedrock” from the search results displayed.
Now, you will see the Amazon Bedrock service page. On this page select “Get started”.
Now, to use Bedrock Service, you will have to request access to specific Bedrock’s foundation models (FMs) you want to use. On the Bedrock overpage select “Request model access” and you will get the page to edit and select required models.
So, what you need to do is select a particular FM or you can select all and request for access.
Once access is granted, the models will be available for use, generally, it is immediate and you will also receive an email notification on your AWS account email ID.
Note: You will also need the correct IAM Permissions to request access.
If you do not have permission to request access to models, you can still browse the Bedrock page.
List of Bedrock’s foundation models (FMs)
| Amazon | AI21 Labs | Anthropic | Cohere | Meta |
|---|---|---|---|---|
| Titan Embeddings G1 – Text | Jurassic-2 Ultra | Claude | Command | Llama 2 Chat 13B |
| Titan Text G1 – Lite | Jurassic-2 Mid | Claude Instant | Command Light | Llama 2 Chat 70B |
| Titan Text G1 – Express | Embed English | Llama 2 13B | ||
| Titan Image Generator G1 | Embed Multilingual | Llama 2 70B | ||
| Titan Multimodal Embeddings G1 | Stability AI | |||
| SDXL 0.8 | ||||
| SDXL 1.0 |
List Of Supported Regions
The below list of AWS Bedrock region availability will help you to understand where is the service available and what options you will get.
| Region | Model evaluation | Knowledge base | Agents | Fine-tuning (custom models) | Continued pre-training (custom models) | Provisioned Throughput |
|---|---|---|---|---|---|---|
| US East (N. Virginia) | Yes | Yes | Yes | Yes | Yes | Yes |
| US West (Oregon) | Yes | Yes | Yes | Yes | Yes | Yes |
| Asia Pacific (Singapore) | No | No | No | No | No | No |
| Asia Pacific (Tokyo) | No | No | No | No | No | No |
| Europe (Frankfurt) | No | No | No | No | No | No |
For an updated list pls. visit your AWS Management Console
Core Benefits of Using Amazon Bedrock Service
- Accelerate the goto market timeline, develop your generative AI applications using available FMs, and use AWS bedrock API for easy integration, without the need to manage infrastructure.
- Choose FMs from AI21 Labs, Stability AI, and Amazon more to find the best fit FM for your needs.
- Leverage AWS capabilities and tools to deploy scalable, reliable, and secure generative AI applications quickly.
These are my top 3 core AWS Bedrock benefits, there are many more.
AWS Bedrock Use Cases
Text generation
You can create new and original content, such as short stories, essays, and social media posts.
Chatbots/Virtual assistants
You can build conversational interfaces such as chatbots/virtual assistants to enhance the customer experience.
Text summarization
Get your summary of textual content quickly without going through the entire information, such as articles, blog posts, books, and documents.
Image generation
This is my favorite, You can create realistic and artistic images of various subjects, environments, and scenes by just giving language prompts.
AWS Bedrock Features And Functions
Take advantage of AWS bedrock functions and features available to speed-up your
generative AI application development journey. Explore the following capabilities of foundation models to get started.
Create applications that help customers on how to questions:
You can build an AWS bedrock chatbot /agent that uses foundation models, makes API calls, and (optionally) queries knowledge bases. Based on the information it can reason through and carry out tasks for your customers.
Use appropriate models for specific tasks with training data:
You can customize an Amazon Bedrock foundation model (FM) by giving access to your training data for fine-tuning to adjust the parameters and improve its performance.
Determine the best model for your needs:
You can evaluate different available FM and then compare the outputs to determine the best model that can be used for your needs.
Improve your FM-based application’s efficiency:
There is an option, where you can provision the throughput for a foundation model to run inference more efficiently. This will reduce the overall time.
Prevent inappropriate content:
Use Guardrails for Amazon Bedrock, and configure rules that can provide governance to safeguard your generative AI applications and AWS Bedrock data privacy.
AWS Bedrock Pricing
How much does Amazon Bedrock cost? With Amazon Bedrock, as of now, you have only two consumption plans.
Option I – AWS Bedrock on demand
With this mode, you only pay for what you use. There are no time-based term commitments in this potion.
Text generation models:
For every input token processed and every output token generated, you are charged.
Embeddings models:
You will be charged for every input token processed. A token is the basic unit that a model uses to understand user input and prompts to generate results.
Image generation models:
It is simple to understand, you will be charged for every image generated.
Option II – AWS Bedrock Provisioned Throughput
In this option, you can purchase model units for a specific base or custom model.
A model unit provides a certain throughput. This is measured by the maximum number of input tokens or output tokens processed per minute.
- Provisioned throughput pricing is charged per hour of usage.
- You can choose between a month or six-month time period.
- These are upfront commitment terms.
This option is for large, consistent inference workloads that need to guarantee throughput. So, AWS Bedrock cost can be on-demand usage or provisioned throughput, look at your needs and select the best option that suits your needs.
Note: Currently AWS bedrock custom models are only available through the Provisioned Throughput option.
Example of AWS bedrock generative AI
For learning purposes, I have used the AWS bedrock service, and below is the output. I have followed the same steps given above to access and request access to Bedrock’s foundation models (FMs).
AWS Bedrock examples of titan image generator:
I have used “Image Play Ground” and Titan Image Generator G1v1 Using on-demand throughput (ODT).
The Negative prompt given was:
“Draw two blue and gold macaws sitting on a tree branch with a green and colorful background. Birds are in a forest. So the background should have trees, and birds facing each other on the tree top branch. We can see the bird’s beautiful colors. The picture should have a water panting look.”
Generated Image of blue and gold Macaws
The rest of the setting was at default. To produce the output (3 Images) it did not take much time. A reference image was given from the internet.
Conclusion:
An ever-growing demand for generative AI applications in every industry has given a boost to the need for new technology platforms. Cloud providers have an opportunity to help businesses fulfill this need to develop and scale generative AI applications.
Amazon Bedrock helps you go from generic models to ones that are customized for your needs in just a few clicks. Also, AWS Bedrock data security ensures that your data is not shared, and not used to train its models.
Many use cases are awaiting to have access to these technologies and I am super excited to see them in action. Let us know your views of Amazon Bedrock in the comment section below.