We should add structured output support to AmazonBedrockChatGenerator.
Here are the AWS docs explaining the feature https://docs.aws.amazon.com/bedrock/latest/userguide/structured-output.html. It will require adding support for outputConfig.textFormat in our response to the AWS Converse API. We currently support other configs like inferenceConfig, toolConfig, guardrailConfig, etc.
This came up in our call with @ju-gu and @bogdankostic where this feature is needed and currently being accomplished with a custom parsing component to handle the cases when the output of the LLM is not JSON compliant.
We should add structured output support to
AmazonBedrockChatGenerator.Here are the AWS docs explaining the feature https://docs.aws.amazon.com/bedrock/latest/userguide/structured-output.html. It will require adding support for
outputConfig.textFormatin our response to the AWS Converse API. We currently support other configs like inferenceConfig, toolConfig, guardrailConfig, etc.This came up in our call with @ju-gu and @bogdankostic where this feature is needed and currently being accomplished with a custom parsing component to handle the cases when the output of the LLM is not JSON compliant.