NVIDIA-Certified Professional: Generative AI LLMs 試験対策問題集

模擬試験モードの設定

上記は本サイトの模擬試験モードの設定値です。実際の試験の出題数・制限時間・合格基準は、各認定プログラムの公式サイトでご確認ください。

出題分野

• LLM Architecture（LLM のアーキテクチャ）（出題比率 6%）
• Safety, Ethics, and Compliance（安全性・倫理・法令遵守）（出題比率 5%）
• Prompt Engineering（プロンプト設計）（出題比率 12%）
• Data Preparation（データ準備）（出題比率 9%）
• Model Optimization（モデル最適化）（出題比率 17%）
• Fine-Tuning（ファインチューニング）（出題比率 12%）
• Evaluation（評価と計測）（出題比率 8%）
• GPU Acceleration and Optimization（GPU 高速化と最適化）（出題比率 14%）
• Model Deployment（モデルの配備）（出題比率 9%）
• Production Monitoring and Reliability（本番の監視と信頼性）（出題比率 8%）

フリー問題

1. One deployment must serve three features from the same weights: a data extractio…
2. An assistant answers from an internal knowledge base and can call a function tha…
3. A classification prompt scores 84 percent with 8 examples in the request and 85 …
4. A team must be able to change the instruction text weekly without risking a regr…
5. Instruction data will arrive continuously from three teams that use different fi…
6. A fine-tuning corpus assembled from support transcripts contains customer names,…
7. A model must fit into a smaller memory budget and also run faster on hardware th…
8. A service must raise its supported context from 4,000 to 32,000 tokens on the sa…
9. A team must both distill and quantize to reach a memory target. Which ordering a…
10. Twelve business units each need their own adapted behavior, and the platform mus…
11. A requirement states that the assistant must not state anything the retrieved pa…
12. The same model must be benchmarked on an on-premises cluster and on rented cloud…
13. A summarization feature is scored by comparing each generated summary against on…
14. A cluster has four nodes of eight devices each. Devices inside a node are joined…
15. A team must distribute the matrix multiplications of multi-head attention across…
16. A team enables mixed-precision training. Which pair of measures keeps the numeri…
17. A team must decide what goes inside the container image and what stays outside, …
18. A service handling 300,000 requests a day must adopt a new model version. The re…
19. The team wants every future version to be comparable against its predecessor wit…
20. The success-rate panel for a generation endpoint sits at 99.9 percent all week, …