Sign in Get started

Interview Q&A · 22 questions · 4 answered publicly

Fine-tuning Interview Questions

Ordered by how often they're asked. Sign in free to see every answer, rate your confidence and run timed mock interviews.

  1. Most asked Fine-tuning leaf: Fine-tune vs RAG vs prompts-decision framework?

    Summary

    Prompts for quick behavior; RAG for changing knowledge; fine-tune for stable style/format/tool patterns.

    Longer Answer

    Prompts for quick behavior; RAG for changing knowledge; fine-tune for stable style/format/tool patterns. Combine when needed.

    Key Points

    • Prompts for quick behavior; RAG for changing knowledge; fine-tune for stable style/format/tool patterns
    • Combine when needed
    • Classic decision question
    • Use the answer to connect the concept back to a realistic production or interview scenario for 'Fine-tuning leaf: Fine-tune vs RAG vs prompts-decision framework?'

    Comparison / Tradeoffs

    Aspect Notes
    Core idea Prompts for quick behavior; RAG for changing knowledge; fine-tune for stable style/format/tool patterns.
    Practical tradeoff Classic decision question.

    Permalink

  2. Most asked Fine-tuning leaf: What is LoRA?

    Summary

    Train low-rank adapters while freezing base weights-cheaper, swappable domain adapters, easier rollback.

    Longer Answer

    Train low-rank adapters while freezing base weights-cheaper, swappable domain adapters, easier rollback. PEFT core.

    Key Points

    • Train low-rank adapters while freezing base weights-cheaper, swappable domain adapters, easier rollback
    • PEFT core
    • Relevant areas: Ai, Fine Tuning, Lora
    • Use the answer to connect the concept back to a realistic production or interview scenario for 'Fine-tuning leaf: What is LoRA?'

    Comparison / Tradeoffs

    Aspect Notes
    Core idea Train low-rank adapters while freezing base weights-cheaper, swappable domain adapters, easier rollback.
    Practical tradeoff PEFT core.

    Permalink

  3. Most asked Fine-tuning leaf: When does fine-tuning fail to add knowledge?

    Summary

    Weights memorize poorly vs retrieval for fresh facts; knowledge goes stale; RAG is better for updatable corpora.

    Longer Answer

    Weights memorize poorly vs retrieval for fresh facts; knowledge goes stale; RAG is better for updatable corpora. Knowledge limits.

    Key Points

    • Weights memorize poorly vs retrieval for fresh facts; knowledge goes stale; RAG is better for updatable corpora
    • Knowledge limits
    • Relevant areas: Ai, Fine Tuning
    • Use the answer to connect the concept back to a realistic production or interview scenario for 'Fine-tuning leaf: When does fine-tuning fail to add knowledge?'

    Comparison / Tradeoffs

    Aspect Notes
    Core idea Weights memorize poorly vs retrieval for fresh facts; knowledge goes stale; RAG is better for updatable corpora.
    Practical tradeoff Weights memorize poorly vs retrieval for fresh facts; knowledge goes stale; RAG is better for updatable corpora.

    Permalink

  4. Most asked Fine-tuning leaf: SFT vs DPO/RLHF briefly?

    Summary

    SFT imitates demonstrations; DPO/RLHF align preferences/rewards.

    Longer Answer

    SFT imitates demonstrations; DPO/RLHF align preferences/rewards. Use preference data when style/safety matters.

    Key Points

    • SFT imitates demonstrations; DPO/RLHF align preferences/rewards
    • Use preference data when style/safety matters
    • Alignment methods
    • Use the answer to connect the concept back to a realistic production or interview scenario for 'Fine-tuning leaf: SFT vs DPO/RLHF briefly?'

    Comparison / Tradeoffs

    Aspect Notes
    Core idea SFT imitates demonstrations; DPO/RLHF align preferences/rewards.
    Practical tradeoff Alignment methods.

    Permalink

  5. Fine-tuning leaf: How do you build a fine-tune dataset?
  6. Fine-tuning leaf: How do you evaluate a fine-tuned model?
  7. Fine-tuning leaf: How do you prevent training data leakage of secrets?
  8. Fine-tuning leaf: What is QLoRA?
  9. Fine-tuning leaf: Catastrophic forgetting-how to mitigate?
  10. Fine-tuning leaf: Serving multiple LoRA adapters?
  11. Fine-tuning leaf: Full fine-tune vs PEFT cost tradeoffs?
  12. Fine-tuning leaf: Domain adaptation for code assistants?
  13. Fine-tuning leaf: What is overfitting in instruction tuning?
  14. Fine-tuning leaf: License/compliance checks before FT?
  15. Fine-tuning leaf: How do you roll back a bad adapter?
  16. Fine-tuning leaf: Distillation use case?
  17. Fine-tuning leaf: Hyperparameters that matter most?
  18. Fine-tuning leaf: Continual fine-tuning strategy?
  19. Fine-tuning leaf: Hardware considerations?
  20. Fine-tuning leaf: Synthetic data-pros/cons?
  21. When should you fine-tune a model versus using RAG or prompt engineering?
  22. What is LoRA/PEFT and why is it popular for adapting LLMs?