Getting error "Input too long" for GAAB solution

[monali-2210]

Describe the bug
I am getting error in a text based use case while using the RAG flow with Knowledge Base, where my kb has multimedia storage S3 bucket enabled. While testing out with normal bedrock knowledge base test playground mode works fine, GAAB solution provides error.

{ "level": "ERROR", "location": "handle_event:103", "message": "An exception occurred in the processing of chat: Error occurred while invoking Bedrock model family 'anthropic' model 'apac.anthropic.claude-3-5-sonnet-20241022-v2:0'. An error occurred (ValidationException) when calling the Converse operation: The model returned the following errors: Input is too long for requested model.", "timestamp": "2025-08-21 10:23:24,337+0000", "service": "BEDROCK_CHAT", "xray_trace_id": "1-68a6f390-ff0b18e031e0b7a792d9255d" }

To Reproduce

1. Create a bedrock Knowledge base, LLM as a parser(used Claude 3 Haiku, semantic chunking, embedding model titan text v2, enable multimedia storage to a s3 bucket).
2. Test the bedrock knowledge base
3. Deploy a GAAB usecase (usecase type - text usecase, model - Anthropic Claude 3.5 sonnet v2, RAG based point to the same Knowledge Base, prompt can be default)
4. Test it out

Expected behavior
A clear and concise description of what you expected to happen.

Please complete the following information about the solution:

• Version: v3.0.4

To get the version of the solution, you can look at the description of the created CloudFormation stack. For example, "(SO0276) - Generative AI Application Builder on AWS Solution. Version v1.0.0".

• Region: ap-south-1
• Was the solution modified from the version published on this repository?
• If the answer to the previous question was yes, are the changes available on GitHub?
• Have you checked your service quotas for the sevices this solution uses?
• Were there any errors in the CloudWatch Logs?: Attached above

Screenshots
If applicable, add screenshots to help explain your problem (please DO NOT include sensitive information).

Additional context
Add any other context about the problem here.

[supreetkt]

Hi @monali - thanks for the report. It seems like you're seeing a difference in response in Bedrock console and the solution. Some things to check here:

1. Are you using the same model for querying the LLM in both the cases?
2. Are prompts and other model settings the same across both?
3. Are the chunks that are being pulled from the knowledge base too long in the solution vs Bedrock console due to differences in knowledge base settings? Are you retrieving the same number documents in Bedrock console as well as the solution?

One way to also debug this is to turn on "Model invocation logging" in the Bedrock console to see the differences in your requests in both the cases. This can be seen inside the "Settings" in Bedrock console. You can create a log group where your exact invocations to Bedrock will be logged and then see what is exactly being sent into the model. It seems like when you're using the solution, your input is larger than when you're trying the Bedrock console.

[monali-2210]

Hey,

1. Yes
2. Yes
3. Yes, same number of documents is being retrieved. Documents/ source chunks are image, in the solution it's being passed as image byte array which is huge/ too long. Byt in bedrock knowledge console it handles those source chunks automatically.

[supreetkt]

Hi @monali-2210 - it seems like your settings are different in Bedrock KB and the solution. Please provide all details for both to proceed.