引导式测验 · 中等 等级

提示缓存测验

检查提供商管理的提示前缀重用、缓存资格、稳定上下文的放置、更改策略和数据保留检查。

相关引导路径提示缓存
问题 1 的 8

What can a provider-managed prompt cache let a later request avoid?

预览每个问题8 问题

当您参加测验时，就会出现答案和解释。

  1. What can a provider-managed prompt cache let a later request avoid?
    • Recomputing some work for an eligible repeated prompt prefix
    • Running the model on any new user content
    • Checking provider-specific cache eligibility
    • Applying attention to the uncached part of a request
  2. Why can a small prompt-prefix change affect cache eligibility?
    • Every provider guarantees cache reuse regardless of prompt changes
    • Provider-specific matching rules or breakpoints may make the changed prefix ineligible for reuse
    • Causal attention prevents any prefix from being reused
    • Changing later content always invalidates every earlier prefix
  3. 哪种场景最能从提示缓存中获益？
    • 带有简短提示的一次性问题
    • 每个单词每次都会改变的提示
    • 许多请求都共享一个长的、相同的系统提示或文档
    • 无需输入生成随机数
  4. How should a team estimate the price effect of a prompt-cache hit?
    • Assume every provider discounts cached tokens by 90%
    • Assume cache hits are always free
    • Use the number of prompt characters alone
    • Use the current provider and model pricing for cached versus uncached tokens
  5. How long does a prompt cache remain available?
    • It depends on the provider, model, and cache-retention settings
    • Exactly one hour for every provider
    • Forever unless a user deletes it
    • Exactly five minutes in every API
  6. How can an OpenAI API client identify how many input tokens were served from cache?
    • The number of characters in the system message
    • The response usage field input_tokens_details.cached_tokens
    • The cache key length
    • 模型参数个数
  7. In OpenAI implicit prompt-caching mode, where is a breakpoint placed by default?
    • It caches every prompt regardless of length
    • It disables prefix matching
    • It places a breakpoint at the end of the latest eligible message
    • It stores all user prompts permanently
  8. What should a team check before caching sensitive prompt content?
    • Only the color of the API dashboard
    • Whether the prompt contains an odd number of tokens
    • Whether another provider offers the same discount
    • The provider’s retention, data-use, and cache-control terms for the selected model