Example of Bedrock Converse API with Managed Prompt Variant

Question

From https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_Converse.html

> If you use a prompt created through Prompt management, specify the ARN of the prompt version. For more information, see Test a prompt using Prompt management.

From referenced https://docs.aws.amazon.com/bedrock/latest/userguide/prompt-management-test.html
> You can test your prompt in the following ways:
>    To run inference on the prompt, send an InvokeModel InvokeModelWithResponseStream, Converse, or ConverseStream request with an Amazon Bedrock runtime endpoint and specify the ARN of the prompt in the modelId parameter.
> Note
> The following restrictions apply when you use a Prompt management prompt with Converse or ConverseStream:
>  -  You can't include the additionalModelRequestFields, inferenceConfig, system, or toolConfig fields.
>  -  If you include the messages field, the messages are appended after the messages defined in the prompt.
> -  If you include the guardrailConfig field, the guardrail is applied to the entire prompt. If you include guardContent blocks in the ContentBlock field, the guardrail will only be applied to those blocks.

There is mention of utilizing a Prompt created in Prompt Management with the Converse API, but there is no mention of how to utilize a **particular variant** of the specified Managed Prompt.
How is a variant selected when a Managed Prompt has multiple variants?

re:Post Agent · Answer

When using a managed prompt with multiple variants in the Amazon Bedrock Converse API, you need to specify the exact ARN of the prompt version you want to use, which includes the variant information.

The ARN structure for a prompt version includes the variant identifier. When you create a prompt in Prompt Management, each variant gets its own unique ARN. To use a specific variant, you would use that variant's complete ARN as the `modelId` parameter in your Converse API call.

For example, in Python using boto3:

```python
import boto3

bedrock = boto3.client('bedrock-runtime')

response = bedrock.converse(
modelId='arn:aws:bedrock:region:account-id:prompt/prompt-name/version/variant-id',
promptVariables={
"variable_name": {"text": "variable_value"},
"another_variable": {"text": "another_value"}
}
)
```

Or using the AWS CLI:

```
aws bedrock-runtime converse \
--model-id arn:aws:bedrock:region:account-id:prompt/prompt-name/version/variant-id \
--prompt-variables '{"variable_name":{"text":"variable_value"},"another_variable":{"text":"another_value"}}'
```

The key is that you're not selecting a variant separately - the variant is already embedded in the ARN you use as the modelId. Each variant has its own distinct ARN that you would use to invoke that specific variant.  
**Sources**  
[Amazon Bedrock Prompt Management is now available in GA | AWS Machine Learning Blog](https://aws.amazon.com/blogs/machine-learning/amazon-bedrock-prompt-management-is-now-available-in-ga/)  
[Using Bedrock invokeModel API with prompt created in prompt management | AWS re:Post](https://repost.aws/questions/QUXSyHftdhTF-fwWTQwFZCcw/using-bedrock-invokemodel-api-with-prompt-created-in-prompt-management)

Example of Bedrock Converse API with Managed Prompt Variant

Add your answer

Relevant content