Question

You are building a customer-facing summarization feature on Databricks that condenses long support email threads. Business stakeholders require that each summary never exceed roughly three sentences, and they've noticed the underlying foundation model sometimes returns verbose multi-paragraph outputs that break the UI layout. You want the most reliable way to enforce concise output while keeping a single foundation model call. What should you do?

Accepted Answer

Add explicit length constraints in the prompt (e.g., 'Respond in no more than 3 sentences') and set a conservative max_tokens value on the completion request.

Answer

Increase the temperature parameter so the model generates more focused, shorter responses.

Answer

Truncate the model's raw output string to a fixed character count after it is returned.

Answer

Switch to a provisioned throughput endpoint so the model has enough capacity to follow instructions.