Adjusting Gemini Verbosity via Supervised Fine-Tuning

PMLE By islever Editorial Last updated

Short answer

D Ask users to provide examples of responses with the appropriate verbosity as a list of question and answer pairs. Use this dataset to perform supervised fine-tuning (SFT) in Agent Platform. Re-evaluate the verbosity of responses with the tuned version.

Question

You deployed a conversational application that uses Gemini. You recently collected thousands of rows of user feedback about the verbosity and accuracy of the model’s responses. The feedback indicated that the responses are factually correct but users want different levels of verbosity depending on the type of question. You need to adjust the model to return responses that are more consistent with users’ expectations. You want to use a scalable solution and minimize application maintenance. What should you do?
  1. Experiment with other models in Model Garden. Perform A/B testing by setting each model as your application’s default model. Choose a model based on the results.
  2. Ask users to indicate all scenarios where they expect concise responses versus verbose responses. Modify the application’s prompt to include these scenarios and their respective verbosity levels. Re-evaluate the verbosity of responses with updated prompts.
  3. Implement a keyword-based routing layer. If the user’s input contains the words “detailed” or “description,” return a verbose response. If the user’s input contains the word “fact,” re-prompt Gemini to summarize the response and return a concise response.
  4. Ask users to provide examples of responses with the appropriate verbosity as a list of question and answer pairs. Use this dataset to perform supervised fine-tuning (SFT) in Agent Platform. Re-evaluate the verbosity of responses with the tuned version. Correct Answer

Key Takeaways

The main trap is choosing prompt engineering (B) or routing (C) for style adjustments; SFT is superior for embedding consistent behavioral patterns at scale with minimal ongoing maintenance.

Learn how to adjust Gemini response verbosity using Supervised Fine-Tuning (SFT) in Google Cloud AI. This guide explains why SFT offers a scalable, low-maintenance solution compared to prompt engineering or routing layers.

Common mistake: Candidates often select Option B (Prompt Engineering) because it seems simpler initially. However, maintaining detailed scenario-specific prompts for thousands of user feedback rows is not scalable and violates the 'minimize maintenance' requirement.

Expert Analysis

The core challenge is to adapt the Gemini model's output style (verbosity) based on question type, while prioritizing scalability and minimal maintenance. The scenario explicitly states that responses are factually correct but lack stylistic consistency.

Supervised Fine-Tuning (SFT) is the appropriate mechanism for embedding specific behavioral patterns or styles into a foundational model when prompt engineering proves insufficient or too brittle. By providing a dataset of question-answer pairs demonstrating the desired verbosity levels, the model learns to generalize this behavior across new inputs. This approach centralizes the logic within the model itself, eliminating the need for complex application-side routing rules or constant prompt updates, thus meeting the requirements for scalability and low maintenance.

Option A is incorrect because switching models in Model Garden does not guarantee improved verbosity control; different models have inherent stylistic tendencies but do not solve the specific instruction-following nuance required here. Option B relies on prompt engineering with explicit scenarios. While valid for simple cases, managing numerous edge cases via prompts becomes unscalable and increases maintenance overhead as user intent diversity grows. Option C proposes a keyword-based routing layer, which is rigid, error-prone (keyword matching fails on semantic variations), and requires significant application code maintenance, violating the 'minimize maintenance' constraint.

The community consensus supports D, noting that SFT directly teaches the model the desired style through examples, creating a scalable solution where the model handles variability internally without external rule engines.

Exam Strategy

When a question asks to improve model behavior (style, tone, format) rather than factual accuracy, look for fine-tuning options if 'scalability' and 'low maintenance' are key constraints. Prompt engineering (B/C) is usually the wrong answer for long-term, complex behavioral adaptation because it shifts complexity to the application layer. Always prioritize solutions that embed logic into the model itself when maintenance costs are a deciding factor.

Frequently Asked Questions

Why is prompt engineering not the best solution for verbosity?

While effective for simple changes, prompt engineering requires manual updates for every new scenario or edge case. As user intent diversifies, maintaining a large set of conditional prompts becomes unsustainable and violates the 'minimize maintenance' goal.

Does fine-tuning affect factual accuracy?

Fine-tuning primarily adjusts style and pattern recognition. Since the scenario states the base model is already factually correct, SFT allows you to refine the presentation (verbosity) without compromising the underlying knowledge.

More PMLE FAQ →

Community Votes

D
100%

100% of anonymous learners picked answer D. Votes are pick records left by other test-takers — they are not the verified answer.

Discussion 3 contributions

Compiled from public community threads and reader submissions, lightly edited for clarity. Handles and vote tallies are shown as recorded, and comments reflect the views of individual contributors.

From public community threads 3

  1. taylz876 👍 1 Answered D
    Repeated Q329 Why D is right • The answers are right, but the style needs to change. Users want different levels of detail depending on the question. That's a behavior the model must learn, and example pairs teach it directly. • Scalable with low maintenance: once tuned, the model handles new questions without hand-written rules. Why the others are wrong • A: Trying other models doesn't target the verbosity problem. • B: A prompt listing every scenario needs ongoing upkeep and can't cover every case. • C: A keyword router is brittle and needs ongoing upkeep.
  2. RajuMVV 👍 1 Answered D
    The correct answer is D. Ask users to provide examples of responses with the appropriate verbosity as a list of question and answer pairs. Use this dataset to perform supervised fine-tuning (SFT) in Agent Platform. Re-evaluate the verbosity of responses with the tuned version.
  3. n2183712847 👍 1 Answered D
    Correct answer: D. Supervised fine-tuning using user-provided question-and-answer examples teaches the model the desired verbosity patterns at scale, improving consistency while avoiding complex application-side routing or prompt logic.

Reader comments & corrections

No reader corrections yet — spotted an error or have a note? Share it below.

Log in to comment, report an error, or add a note about this question.

Practice All PMLE Questions

Access 65 questions with complete answers and detailed explanations.

View Full PMLE Practice Test →

← Back to PMLE Study Guide