Forem: Nikola 🇦🇺 🇸🇰

Build Your Own Generative AI Applications with AWS: Introduction to Amazon Bedrock

Nikola 🇦🇺 🇸🇰 — Thu, 23 Nov 2023 08:04:50 +0000

In today’s tech world, AI is the superstar that’s changing the game. It’s not just about sci-fi movies anymore — it’s all around us, shaping our lives. One of the big players in this AI revolution is ChatGPT. You might have heard of it — it’s the cool tech that can chat with you almost like a human. However, ChatGPT is not the only player in the game. AWS recently announced general availability of Amazon Bedrock — a fully managed service that will help you build and scale your own generative AI applications.

AWS is making continuous effort to democratize AI and enable broader access to its capabilities. With the launch of their new service, this exciting new technology is now attainable not just for research institutions or well-funded start-ups, but for organizations of all sizes and sectors. Amazon Bedrock enables you to quickly and easily access high-performing AI models and seamlessly integrate them into your applications in a cost-effective way and without managing any infrastructure. The service offers multiple foundational models to start with, but you can add your own custom data and further train the model for more specific tasks. This can be done without worrying about data security and without hiring an army of machine learning engineers. Regardless of the size of your business, you can take advantage of these models and go from an idea to a market-ready product in no time even if you’re on a budget.

What are foundational models (FMs)?

Foundation models are large deep learning neural networks trained on massive datasets. Instead of developing the traditional machine learning (ML) models that typically perform specific tasks from scratch, you can use foundational models as base models for developing more specialized applications, speeding up and optimizing the development of new ML models. FMs are trained on a broad spectrum of generalized and unlabeled data which gives them the ability to perform a wide range of general tasks out of the box, including text and image generation, natural language conversation and language understanding.

Foundational models are a form of generative AI so if you’ve used ChatGPT you already know how to use them. You provide one or more inputs (prompts) in the form of human language instructions, and the FM will generate output.

What can foundation models do?

The models can answer natural language questions. You can leverage this ability to engage in a dialogue and easily build a virtual assistant or a chatbot. Another handy capability is summarising, synthesizing or searching large amounts of text. These are generative models so of course, they are capable of creating content like articles or social media posts. The FMs can even generate or debug computer code in various programming languages. The ability to understand language also allows FMs to perform tasks such as transcribing a video or translating languages. Another capability is the generation of images from input text, as well as photo and video editing. FMs can also be used to identify images and physical objects which can be used in robotics.

If an off-the-shelf solution doesn’t suffice, you can use your own data to customize the model in order to improve the its performance on specific tasks. When you provide some labeled examples related to a specific task that needs to be carried out, you help the model learn. As a result, you can create a new model that is better capable of completing the tasks you need than the base model.

Now let’s take a closer look at the foundational models available in Amazon Bedrock.

Titan

A family of powerful, general-purpose models created by Amazon. Titan models were pretrained by AWS on large datasets and built to support a variety of use cases. Amazon Titan foundational models can help you generate or summarize text or hold open-ended conversations, but you can also use them for search and personalization.

There are currently three versions of the models. Titan Text Express, Titan Text Lite (both in preview) — these models can create and summarize text. The third one is Titan Embeddings (generally available, you can read the announcement here). Titan Embeddings is optimized for tasks such as search, personalization, and clustering based on semantic similarity.

Jurassic

Versatile foundational models by AI21 Labs, capable of generating nuanced text across a wide variety of industry sectors and which allow users to interact with the data in natural language and gain insights. Do you need to summarize a financial report, generate a product description or craft a bespoke marketing campaign? You got it. The models support English, as well as several other languages including Spanish, French, German, Portuguese, Italian and Dutch.

Bedrock offers Jurassic-2’s Ultra and Mid models. J2-Ultra is the largest and most powerful model in the Jurassic series. It is a great option for the very complex language processing tasks and generative text applications. J2-Mid is well-suited to complex language tasks and its fine-tuning options enable quality optimization without compromising affordability and high efficiency.

Claude

These foundational models by Anthropic excel at thoughtful dialogue, content creation, complex reasoning, creativity, and coding. Clause can process extensive amount of information such as legal documents or transcripts and it’s coding skills in a variety of programming languages are constantly improving.

You can opt for Claude Instant, which is a faster, more economical, yet very capable model that can manage various tasks, including casual conversation, text analysis, summarization, and document comprehension. Amazon Bedrock also offers Claude 2 and Claude 1.3. Claude 1.3 is an earlier version of Anthropic’s general-purpose large language model. Claude 2 is the newest and the most advanced version of the model which offers higher performance in math, reasoning, and coding tasks than its predecessor. The model was designed to be helpful, harmless, and honest within conversations and I have to say I really enjoyed interacting with it. If you want to try it for yourself, you can talk to Claude here.

Command

Command by Cohere is another text generation model but this time the focus is on business use cases. The text generation models can be used to automate creating content such as social media posts or product descriptions. Employee productivity can be enhanced with the use of virtual assistants which can perform routine tasks like scheduling meetings. Something that deserves to be highlighted here is the integrity — models are trained from known, purchased, or public data sources, and subjected to adversarial testing and bias mitigation.

There are two types of models available. Command — the text generation model which can summarize and generate text. You can either use Command or Command Light which is the smaller version of Command. Then there is the Embed model, which can be used for search, clustering, or classification tasks. There are two options available — an English only version and a multilingual version which supports over a hundred languages.

Llama 2

These models by Meta AI are optimized for dialogue use cases. When the second version of the model was being created, a lot of effort was put into preventing harmful, offensive or inappropriate responses which makes it ideal for customer service applications. Llama 2 can generate, translate and summarize text and even correct your grammar.

You can pick from two versions — Llama-2–13b-chat or Llama-2–70b-chat (coming soon). Llama-2–70b-chat will allow you to build a chatbot with the best accuracy. If your priority is to create a chatbot quickly and cost-effectively, even if that means sacrificing some precision, then Llama-2–13b-chat is a great option.

Stable Diffusion XL

This text-to-image model by Stability AI is capable of generating images of high quality in virtually any art style and is the best open model for photorealism. If you are in marketing, advertising, media, entertainment or gaming, this model might get your attention. You can use it to help you generate compelling ad campaigns and other assets.

Bedrock offers Stable Diffusion XL 0.8 and Stable Diffusion XL 1.0, both currently in preview. Stable Diffusion XL 1.0 is the most advanced text-to-image model from Stability AI, capable of producing images at a resolution of up to 1024×1024 pixels — much higher than the previous models. This upgrade allows Stable Diffusion XL to produce images with finer details like textures, patterns and facial features with much higher fidelity. You can read more and see some examples of the generated images here.

Model access

You can access the Amazon Bedrock API in all the familiar ways — using the AWS Command Line Interface, an AWS SDK, or a SageMaker Notebook. Bedrock is currently only supported in five regions, you can see the list here.

To use any of these foundational models, you need to request access (select the Model access link in the left side navigation panel in the Amazon Bedrock console). You will need to have the correct IAM permissions and for some of the models, you may need to submit use case details before you are able to request access.

On the Examples tab in the console, you will be able to find example prompts and API requests for each of the supported models. You can run these examples by choosing Open in playground.

Pricing

When you use Amazon Bedrock, you will be charged for model inference and customization. For inference, you can choose On Demand or Provisioned Throughput plan.

With On Demand mode, you will only pay for what you use. It will allow you to use FMs on a pay-as-you-go basis without the need to make fixed time commitments. What you are charged for will depend on whether you are using a text generation, image generation or embeddings models. You can read the details here.

Provisioned Throughput mode might be a better option if your application’s performance requirements are high and you need guaranteed throughput. You will be charged by the hour and you can choose between 1-month or 6-month commitment terms. Custom models can only be accessed using Provisioned Throughput.

If you decide to customize your model, you will also be charged for model training. To see the details of how you’re going to be charged for training an inference on custom models and also the breakdown of pricing for every provider and model, click here.

Getting Started with Amazon Bedrock

AWS SkillBuilder offers a 60-minute course that can help you understand Amazon Bedrock a bit better. You can access the documentation for the service here.

Summary

Foundational models are unlocking new AI capabilities for enterprises across sectors. These robust, versatile models can provide the core technical building blocks companies need to bring innovative AI systems to market faster. For enterprises looking to harness AI’s benefits, foundational models provide proven, production-ready springboards to develop tailored solutions on budget and on time.

Are new to the world of cloud computing and want to learn more but don’t know where to start? I can help you out! I created an online course for people who are preparing for their first AWS Certification exam. It contains all the tips and tricks you’ll need to ace your exam. You can find more information here.

AWS Certification Exam - in Person or Online?

Nikola 🇦🇺 🇸🇰 — Tue, 28 Feb 2023 08:27:46 +0000

If you’re thinking about obtaining an AWS Certification, you may already know that you have the option to take your exam at a testing centre or from the comfort of your own home or office. Each option has some advantages and disadvantages that you should consider before booking your exam. Having experienced both, I’ve compiled some tips to give you a better understanding of the differences between taking the exam at a testing centre versus taking it at home.

Taking a test online from your home or your office

Your location doesn’t matter

One of the most significant benefits of online exams is their accessibility. If you live in an area where there are no testing centres available, or if the nearest testing centre is too far away or hard to get to, online exams provide a convenient alternative. No matter where you are in the world, as long as you have a reliable internet connection, you can take your online exam.

However, you might choose an online exam even if you do live near a testing centre. Commuting on the exam day can be a source of additional stress, and unforeseen circumstances such as road closures or traffic accidents can make the journey even more challenging. With online exams, you can eliminate the need for planning and commuting, allowing you to focus on the exam and perform at your best.

Familiar environment

Taking an exam from the comfort of your home and in a familiar environment can help you feel calm and relaxed. However, it’s important to keep in mind that online exams have specific guidelines that must be followed. These guidelines may require you to rearrange your furniture and clear the space around you. Additionally, you will need to take photos of your room or office before the exam, so it’s important to remove any personal items that you don’t want your proctor (the person who is going to monitor your behaviour during the online exam to help prevent cheating and keep you on track to successfully complete the exam) to see. If you follow these guidelines, you will be able to pass your exam check-in smoothly and create an environment where you can take your exam without any distractions.

Beware of interruptions

When taking your exam from home, it is essential to carefully manage your environment and do whatever you can to prevent any interruptions. Even innocent sounds or movements can be perceived as an attempt to cheat. Remember to turn off computer notifications and ensure that no one enters the room during your test. If you are expecting any deliveries, it is best to reschedule them, as leaving your room during the test is not permitted. Animals could also be a problem — if there is a risk that your pet could enter the room or make a lot of noise, it might be better to consider taking the test at a testing centre instead. This also applies if you live in a loud area with a lot of outside noise, such as from a busy road or construction site.

Pro tip: Before your exam, turn on the lights in your room, regardless of the time of day. Weather conditions can change rapidly, and if your room becomes too dim, the proctor may have difficulty seeing you clearly. Keep in mind that you won’t be able to leave your seat during the exam to adjust the lighting, so it’s best to do this beforehand.

Avoid suspicious behaviour

During the exam, your camera and microphone will be in use, and any unusual behaviour will be detected. To avoid any misunderstandings, try to avoid talking to yourself while thinking, as this could be interpreted as an attempt to communicate with others. Additionally, it’s crucial to keep your focus on the screen throughout the exam, even though this can be challenging and tiring for longer tests. Looking away from the screen will be identified by the proctor and may result in a warning. If something suspicious happens multiple times, the proctor will end your exam and you will have to reschedule (you will not get a refund for your exam fee in this case).

Technical requirements

To take the exam using your personal equipment, you will need a computer with a functional camera and microphone, as well as a stable and sufficiently fast internet connection. Prior to the exam, you will be able to run a system check to ensure all necessary functions are working correctly. It’s best to complete the system check as soon as you can (you will get a link when you register) using the same computer and internet connection you plan on using during the exam so that you have ample time to address any potential issues. Additionally, it’s recommended to use your personal computer for the exam, as some company computers might be running processes in the background that could interfere with the testing software.

Check in early

When you book your exam, you will get to pick a time you want your exam to start but it’s best to check in about 30 minutes early. That should give you enough time to upload the photos of your ID, your desk, and your room. The online proctor may ask you some additional questions to make sure that the testing guidelines are being followed. You can’t skip the check in and if you show up more than 15 minutes late, you will not be able to start your exam.

Taking a test at a testing centre

Everything is prepared and set up for you

Testing centres have several advantages over taking exams at home. Firstly, you don’t need to worry about unexpected software issues or unreliable internet connection — everything is set up and prepared. Additionally, the environment is controlled to avoid any sudden disturbances that could interrupt your exam. However, keep in mind that other people may be taking their exams at the same time, which can result in noises such as chair squeaks, coughing or sighs. Although AWS Certification exams don’t require typing, others might, so you should also expect some keyboard clicking sounds in the background. Noise-cancelling headphones are typically available, but it’s best to confirm availability with your testing centre. Lastly, you will be provided with a sheet of paper and a pencil to write on (this is not allowed on the online exam), which should be left behind when you leave.

Commute

One downside of testing centres is the need to plan your commute. If the testing centre is in an unfamiliar area, consider driving there a day in advance and checking for any planned roadwork (or trackwork and other interruptions if you are using public transport). Keep in mind that commuting during morning or afternoon peak hours may take longer. Allocate an extra 10–15 minutes for unexpected changes. When you arrive, allow an additional 10–15 minutes for check-in, including presenting your ID and storing your belongings in a locker. Using the restroom is recommended, particularly for longer exams.

Pro tip: before selecting a testing centre, be sure to read the reviews. Previous test-takers may have experienced issues such as malfunctioning equipment. To avoid similar problems, it’s important to review the ratings and feedback from previous test-takers. Leaving a review after your own testing experience can help others in the future.

Conclusion

Which of the two available options is going to work better for you depends on you personal preferences and individual needs. Personally, having done both multiple times, I prefer the testing centre because planning a commute is less stressful for me than setting up a room to meet exam requirements and hoping my laptop and internet connection won’t fail.

I hope this article has been helpful and wish you the best of luck on your exam!

Did you like this article? There’s more! I created an online course specifically for people who are about to start preparing for their first AWS Certification exam. It contains all the tips and tricks you’ll need to ace your exam. You can find more information here.