bedrock-inference

_{Keywords: generative AI, Amazon AWS, Bedrock, API, wrapper, LLM, inference, NLP.}

---

Bedrock Inference is a simple python package to handle 2FA and on-demand calls to AWS Bedrock foundation models.

Requirements

To invoke Amazon Bedrock models, you need to authenticate with your AWS IAM account and establish a connection through MFA. This requires to retrieve some codes

• An Access key ID (es: AKIAIOSFODNN7EXAMPLE) and its correspondent secret access key (es: wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY) for API requests

  • These credentials can be found created at Home -> Security credentials -> Access Keys.

    • If you don't have an access key yet: Access Keys -> Create access key

  • If you have AWS CLI/SDK installed and set up on your machine, chances are your credentials are already stored at ~/.aws/credentials and/or in the AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY environment variables. In this case, the package will retrieve them automatically.

    • Note: AWS CLI and AWS SDks are not required to use bedrock-inference.

• A Multi-factor authentication device ARN (es: arn:aws:iam::012345678910:mfa/example_username)

  • The ARN of your MFA device can be located at Home -> Security credentials -> Multi-factor authentication

• The 6 digits temp token generated by the MDA device (es: 366 712)

Tip

Even if you don't have any AWS service set up on your machine, you can still create the ~/.aws/credential file to store your access key credentials for later reuse. The credential file should look like this:

[bedrock_profile]
aws_access_key_id = AKIAIOSFODNN7EXAMPLE
aws_secret_access_key = wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY

The profile name you choose can be used later in aws_login_mfa in place of aws_access_key_id and aws_secret_access_key arguments.

Installation

1. Make sure you have the latest version of pip installed

   pip install --upgrade pip

2. Install araucanaxai through pip

   pip install bedrock_inference

Usage

Here is a minimal example of how the AWS Bedrock models can be invoked with bedrock-inference.

from bedrock_inference import bedrock

# Authenticates through MFA and establish a session
session = bedrock.aws_login_mfa(arn="MFA_device_arn", token="MFA_token", aws_access_key_id="your_ID", aws_secret_access_key="your_key")
#session =  bedrock.aws_login_mfa(arn="MFA_device_arn", token="MFA_token") #use this if you already have AWS credentials set up for API requests on your machine

# Instantiate youre Bedrock caller
caller = bedrock.Bedrock(session=session, region="your_aws_region")

#Check available models in your region
print(caller.list_models())

#Invoke the target AWS model and retrieve the generated answer
answer = caller.invoke_model(prompt="What's the conceptual opposite of 'Hello World'?", model_id="target_aws_model_id")
print(answer)

A more extensive example, including advanced usage, can be found in this notebook.

Contacts and Useful Links

• Repository maintainer: Tommaso M. Buonocore

• Project Link: https://github.com/detsutut/bedrock-inference

• Package Link: https://pypi.org/project/bedrock-inference/

License

Distributed under MIT License. See LICENSE for more information.