A legal team uses a foundation model on Amazon Bedrock to draft standardized contract clauses. They complain that identical prompts sometimes produce noticeably different wording, and they want responses to be as consistent and repeatable as possible. Which inference parameter adjustment should they make to reduce this variability?