Access AWS services + Simple Bedrock Tutorial

This repo demonstrates how to get started with AWS services and how to make a basic call to Bedrock in Python

• Access AWS services + Simple Bedrock Tutorial
  • 1) AWS Management Console
    • 1a) Logging into AWS
    • 1b) Creating your user
    • 1c) Creating an Access Key
    • 1d) Enabling AWS Bedrock models
  • 2) Setting up your Python environment
    • 2a) Installing the required packages
    • 2b) Putting the AWS credentials in the right place
  • 3) Using the Python files

1) AWS Management Console

Before jumping into the code, we first need to make a new user. AWS typically has a root user that gives out permissions, with other users that run specific services. Because the hackathon is so short, we're only going to create 1 user and give that one all the permissions you need.

1a) Logging into AWS

• Navigate to your provided workshop account url
  • Should look something like https://catalog.us-east-1.prod.workshops.aws/<...>
• Click the Email one-time password (OTP) option in orange
• Enter in your SCU email address
• Check your email, copy-paste the code into the box, and click Sign In
• Scoll to the bottom of the Terms and Conditions, check the box to agree, then click Join event
• On the Event dashboard page it directs you to, go to the sidebar and click Open AWS console in the AWS account access section

1b) Creating your user

• Type IAM in the search bar at the top, and click on the top result

• Click on the Users table under Access management tab on the left sidebar
• You should see your root user, which will have a bunch of Access denied text over it. Ignore that and click the orange Create user button in the top right
• Give it a name, whatever you think is best
  • You do NOT need to give it access to the AWS console, as the root user can already do that
• In the permission options, click the Attach policies directly box
  • If you create more users, you can copy permissions, but do NOT copy from the root
• There are many permission policies, but I would recommend searching the service you want to use, then FullAccess after it
  • Ex: AmazonBedrockFullAccess, click the checkbox to add it
  • You can add policies later if you find out you missed one
  • I would recommend Bedrock, SageMaker, EC2, and S3 Full Access to start
• Once you've selected your services, click the Next button at the bottom
• If everything is correct, click Create user

1c) Creating an Access Key

This will allow us to use AWS services in our code

• After creating your user, the Users IAM section will now show the root and your new account's name. Click on your new account
• This page is where you can add permissions later if necessary, but we need to create an Access Key. In the Summary box, click the Create access key link on the right side

• It will ask you for a use case, select the Local code option. Ignore the suggestion it gives you and check the conformation box and click Next
• The description is optional, so leave it blank and click Create access key
• This gives you an access key and a secret. Copy both of these in a place you won't lose them
  • THE SECRET CAN NEVER BE VIEWED AGAIN ONCE YOU CLOSE THIS SCREEN
  • If you lose it, just create another key
• NOTE: These are extremely sensitive keys! Do NOT push these codes to GitHub and share them with your team via Discord DM, iMessage, etc.

1d) Enabling AWS Bedrock models

If you plan to use AWS Bedrock models in your code (or plan on following this README past this point), you need to get the model ID in 1 of 2 ways.

• In the search bar, type Bedrock and click on that service, it has a green icon
• Feel free to click on Model catalog under the Discover section on the left sidebar and explore the different models, determining what would be best for your use case
  • You can use the playgrounds in the Test section on the left sidebar to explore models further
  • Keep note of values such as temperature, top P, etc. as you tune your model responses to your liking
• Once you have one you'd like to use in your code, click on the model in the Model catalog section and look for the Inference type field
  • If your model does NOT have one (ex: OpenAI models), you can just copy the Model ID

• If this field is defined (ex: Claude models), you need a different parameter: inference ID
  • Under the Infer section on the left sidebar, click Cross-region inference
  • Search for your model and copy the Inference profile ID field
    • Use the US option if possible, although global profiles work as well

2) Setting up your Python environment

First, we need to create a virtual environment for Python, which essentially installs your packages to one project instead of globally

2a) Installing the required packages

• Create a virtual environment by running the following in the terminal directory of your project
  • Windows: py -m venv venv
  • Mac/Linux: python3 -m venv venv
• Then, you need to actually use the environment, so type the following
  • Windows: venv\Scripts\Activate.ps1
  • Mac/Linux: source ./venv/bin/activate
• Your terminal should have (venv) at the beginning of it now
  • I would recommend setting your VSCode interpreter to the venv path in the bottom right when a Python file is open
• I have made a requirements.txt to install all of the packages you will need, so run the following to install them
  • pip install -r requirements.txt

2b) Putting the AWS credentials in the right place

In order to not push your keys to GitHub, we need to put them in a .env file

• Open that file and replace YOUR_ACCESS_KEY and YOUR_SECRET_KEY with the ones you copied earlier
• Go to the .gitignore file and uncomment (remove the #) from line 4 that says .env
  • A .gitignore file tells git to not push certain files and keep them local to computer
  • Remember, your AWS credentials are EXTREMELY CONFIDENTIAL, hence why we do this
• Now you are ready to call AWS services!

3) Using the Python files

There are 2 files I have made for you: bedrock.py and converse.py. Both call Amazon Bedrock, using the model you enabled and specify, to generate a response to a given prompt.

• Change model_id to the model you selected on the AWS console
  • This changes depending on if you just need the Model ID or a model that requires an Inference ID
• Change user_message to the prompt you want to use
• Change the inferenceConfig if you want to specifiy temperature, topP, etc.
• Run the file you changed with the following, with the terminal showing (venv)
  • python bedrock.py or python converse.py
• You should see the response in the console!
  • bedrock.py may take a long time depending on the prompt, as it waits for the entire response to complete
  • converse.py acts more like ChatGPT, where it generates the response in real-time