Just FYI — I'm not an ML person. I was trying out Bedrock and wanted a reliable way to get JSON responses from the Llama 3.3 model on AWS Bedrock. Is there a better way than just prompting?

I have previously tried running Llama on Ollama locally and it works great since it's supported out of the box.

I have looked into outlines + vllm, but that requires a powerful machine + GPU to produce structured outputs, and that's just wasteful, because I would be better off hosting my own Llama and using ollama.