Feature: AWS bedrock generator

[fabriziorocco]

Adding support for the Bedrock Generator. Adding this because customer are requesting to test LLM on AWS (focus on Amazon Nova models).
Bedrock use boto3 SDK and Converse API to make requests. Credentials are managed using AWS Credentials or new Bedrock API Key (recently introduced).

Verification

List the steps needed to make sure this thing works

• python -m garak --model_type bedrock --model_name nova-pro

If you are opening a PR for a new plugin that targets a specific piece of hardware or requires a complex or hard-to-find testing environment, we recommend that you send us as much detail as possible.

Complex Software Examples:

• Deployed AWS Bedrock AI service

No specific hardware requirements

[github-actions]

DCO Assistant Lite bot All contributors have signed the DCO ✍️ ✅

[fabriziorocco]

I have read the DCO Document and I hereby sign the DCO

[fabriziorocco]

recheck

[fabriziorocco]

Claude 3 sonnet is not supported so test is failing. I'm open to change the model in pytest

[fabriziorocco]

Hi @leondz the test errors should be fixed now. I didn't include the .rst file

[jmartin-tech]

Thank you, this is a great addition.

Comments are based on static code review. Full integration testing is still in queue.

[fabriziorocco]

added Value Error logic. It was deleted from previous commit

[jmartin-tech]

Some minor additional thoughts on UX and supported message roles. I suspect these may result in less confusion if this generator is utilized with complex configurations.

[fabriziorocco]

I tried to use the base class's _conversation_to_list method which formatted message content as a plain string, but AWS Bedrock's Converse API requires content to be a list of content blocks. The result was the test failing

[jmartin-tech]

This looks complete, thank you!

[jmartin-tech]

Sorry for the misdirection, I see this is a slightly different format than the what is used in the base class.

[leondz]

Thank you Fabrizio!