Bedrock - Converse API - Described guardrails format not working

[rd37011]

Describe the issue

I am able to chat with my model via ConverseCommand without issues. I recently created a guardrail on the AWS Console, and I was able to achieve the desired results.

When I attempted to integrate the guardrail using guardrailConfig on ConverseCommandInput and adding the guardContent property to the latest message (I am trying to run the guardrail on the last user message) I get a ValidationException: Guardrail was enabled but input is in incorrect format., even though the format in the message below matches the example.

I hope it's a simple fix on my end and I've messed up the format somehow.

Here is a simplified code snippet that results in the error I'm seeing:

const client = new BedrockRuntimeClient({
      region: 'us-west-2',
    })
    const input: ConverseCommandInput = {
      modelId: 'meta.llama3-1-8b-instruct-v1:0',
      messages: [
        {
            "role": "user",
            "content": [
                {
                    "text": 'Give me a recipe for mayonnaise',
                }
            ]
        },
        {
          "role": "assistant",
          "content": [
              {
                  "text": 'Sure! The recipe for mayonnaise is...',
              },
          ]
      },
      {
        "role": "user",
        "content": [
            {
                "guardContent": {
                    "text": {
                        "text": 'Is mayonnaise an instrument?'
                    }
                }
            }
        ]
    }],
      guardrailConfig: { 
        guardrailIdentifier: "arn:aws:bedrock:us-east-1:<acct_id>:guardrail/<guardrail_id>",
        guardrailVersion: "Version 1",
    },
      inferenceConfig: {
        maxTokens: 512,
        temperature: 0.2,
        topP: 1,
        stopSequences: [],
      },
      system: [
        {
          text: 'You are a helpful assistant.'
        },
      ],
    }
    
    const command = new ConverseCommand(input)
    const response = await client.send(command)

Links

https://docs.aws.amazon.com/AWSJavaScriptSDK/v3/latest/client/bedrock-runtime/command/ConverseCommand/
https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails-use-converse-api.html#guardrails-use-converse-api-call-configure

[rd37011]

In case anyone else comes across this again - ValidationException: Guardrail was enabled but input is in incorrect format.
I was able to figure out through trial and error that the guardrails must be located in the same region as the foundational model being used. Once I switched to a model in us-east-1, my request worked. That being said, this was not explicitly mentioned in any resources I consulted.

Would love for the resulting error message to be a bit more descriptive - e.g. "ValidationException: Guardrail region must match model region" or something to that effect, assuming of course that is the behavior.

[aBurmeseDev]

Hi @rd37011 - thanks for reaching out and sharing the feedback.

I've confirmed with service team Bedrock that this's expected behavior to have same region for both model and guardrails. Since the error comes from service team rather than SDK, I've forwarded this feedback to the team.

Please feel free to reach out again. Closing the issue.

[github-actions]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs and link to relevant comments in this thread.