Help with running Examples

[GanbaruTobi]

When I run the Examples locally, most of them fail, because the LLM will not stick to the output format. I am unsure if there is even a format requested in the code...
So my problem is I am unsure how to debug why that is. What would be the way to display the actual used prompts generated?
Because I wonder if my doc-annotations are being parsed even.
The only way I was able to fix it to work was giving it an explicit instruction to the builder like:

const QA_INSTRUCTION: &str = "\
You are a strict data-processing engine, not a conversational assistant. \
You must output exactly the requested fields with zero conversational filler, \
no pleasantries, and no markdown wrapping. Do not say 'Here is the answer' \
or 'How can I assist you'. Just provide the raw text.\
";
#[builder(default = Predict::<QA>::builder().instruction(QA_INSTRUCTION).build())]

Thanks for any tips!