Викторина, связанная с руководством · Средний Уровень
Точная настройка, RAG и наводящая викторина
Тесты на то, как диагностировать пробелы в знаниях, формате, стиле и навыках, когда RAG превосходит тонкую настройку и почему наборы оценок определяют решение.
Вопрос 1 из 8
A model answers questions about your company's pricing incorrectly because prices change every quarter. Which approach best fits this gap?
Предварительный просмотр каждого вопроса8 вопросы
Ответы и пояснения появляются по мере прохождения теста.
- A model answers questions about your company's pricing incorrectly because prices change every quarter. Which approach best fits this gap?
- Fine-tune the model on last quarter's price list
- Use RAG to retrieve the current price documents at answer time
- Lower the model's temperature
- Add more examples of polite answers to the prompt
- According to the guide, what is usually the first lever to try for format and style problems?
- Full fine-tuning
- Building a vector database
- Prompting with clear instructions and examples
- Switching to a larger model
- Why is fine-tuning described as an unreliable way to add factual knowledge?
- It cannot change model weights
- Работает только с моделями изображений.
- It is hard to update and can increase confident errors about rarely seen material
- It always deletes the model's existing knowledge
- Which situation most clearly favors fine-tuning?
- You need a narrow task done at high volume and want a smaller, cheaper model to match a larger one
- Your documents change every day
- You want answers to cite source passages
- You have not yet built any evaluation examples
- In a RAG system giving wrong answers, what should you check before blaming the language model?
- Дата обучения модели
- Whether the correct passage appears in the retrieved results
- The number of GPU cores
- The length of the system prompt only
- What does LoRA do in the context of fine-tuning?
- Retrieves documents from a search index
- Compresses prompts to save tokens
- Trains small adapter matrices instead of updating all model weights
- Removes the need for training data
- Which statement about combining the three methods is accurate according to the guide?
- They are mutually exclusive and cannot be combined
- Production systems often use a fine-tuned model reading retrieved context, driven by a good prompt
- RAG replaces the need for any prompt
- Fine-tuning must always come before prompting
- What should drive the choice between prompting, RAG and fine-tuning?
- Whichever method is newest
- A representative evaluation set measuring each option
- The vendor's default setting
- The size of the training budget alone