Ikibazo gihuza ibibazo · Hagati Urwego
Nigute Wubaka Chatbot hamwe na LLM API Ikibazo
Kubaka ikiganiro hafi ya API isaba loop, leta hamwe nabatanga itandukaniro, gutembera, gukora ibikoresho, imipaka yikimenyetso, no gukemura amakosa neza.
Ikibazo 1 Bya 8
In a client-managed Chat Completions flow, what context is commonly sent on each new turn?
Reba buri kibazo8 ibibazo
Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.
- In a client-managed Chat Completions flow, what context is commonly sent on each new turn?
- Only the newest user message
- The full array of prior messages plus the new one
- A compressed binary session token
- Just the system prompt
- In a client-managed, stateless chat flow, why can the bot refer to earlier turns?
- 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
- According to the guide, what is the main practical benefit of streaming responses in a chatbot?
- 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
- According to the guide, what should trigger exponential backoff rather than an immediate retry?
- 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
- How does tool calling get incorporated into the basic chatbot loop?
- 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