引導式測驗 · 中等 等級
如何使用 LLM API 建立聊天機器人測驗
圍繞 API 請求循環、狀態和提供者差異、串流、工具執行、令牌限制和安全錯誤處理來建立聊天機器人。
問題 1 的 8
在客戶端管理的聊天完成流程中,每個新回合通常會發送什麼上下文?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- 在客戶端管理的聊天完成流程中,每個新回合通常會發送什麼上下文?
- Only the newest user message
- The full array of prior messages plus the new one
- A compressed binary session token
- Just the system prompt
- 在客戶端管理的無狀態聊天流程中,為什麼機器人可以引用先前的回合?
- The model stores a session on the provider's server automatically
- The client resends the full message history on every call
- The API caches embeddings of prior turns
- The model has a persistent memory across all users
- 根據該指南,聊天機器人中串流回應的主要實際好處是什麼?
- It reduces the total token cost of the request
- It allows the model to skip the system prompt
- It improves perceived speed by rendering text as it arrives
- It removes the need for error handling
- 根據指南,什麼應該觸發指數退避而不是立即重試?
- A successful 200 response
- Rate limit errors (429) and transient server errors (5xx)
- Any assistant message longer than expected
- A user typing a new message
- 工具呼叫如何融入基本的聊天機器人循環?
- The model executes the function itself on the provider's servers
- The application runs the requested function and appends its result as a message before calling the model again
- Tool calls replace the system prompt entirely
- The client must switch to a different API endpoint
- For a new OpenAI chatbot, what API and conversation-state approach does current documentation recommend?
- Chat Completions is no longer supported
- Responses is recommended for new projects; Chat Completions remains supported and commonly uses client-supplied message history
- Responses can only accept one user turn
- Both APIs always store the same server-side conversation
- Why is counting characters or words insufficient for managing a chatbot's context window?
- Context limits are enforced in tokens, which don't map directly to characters or words
- Characters are always more expensive than tokens
- The API rejects any request with punctuation
- Word counts are only relevant for streaming
- According to the guide, what is a recommended strategy when a conversation's message array grows too large for the context window?
- Delete the system prompt to save space
- Trim or summarize older turns before the limit is reached
- Switch every message to the assistant role
- Stop streaming responses