Question

You are building a generative application in Azure AI Foundry that produces structured JSON summaries of legal contracts. Testers report that identical input contracts sometimes produce differently worded summaries across runs, which fails downstream schema validation and complicates auditing. You must make the output as consistent and reproducible as possible for the same input while keeping the same deployed model. Which combination of generation parameter adjustments should you apply?

Accepted Answer

Set temperature close to 0 and top_p close to 1, keeping other sampling controls at defaults

Answer

Increase temperature toward 1.0 and lower top_p toward 0.1 to focus the distribution

Answer

Raise the frequency_penalty and presence_penalty to their maximum values

Answer

Increase max_tokens and enable multiple choices (n>1) so the app can pick the best response