I want to extract values (e.g. some lab values) from medical narratives. I can tell GPT 4 to show me just the value, e.g. 15 ml/l, or 80% and i get the answer without problem. If i try with llama-2 or Med42 (i tried 70B models for each), it is nearly impossible to just get the number. It will ramble and even hallucinate. Any tips ?