Ikibazo gihuza ibibazo · Hagati Urwego
Max Tokens hanyuma uhagarike Ikibazo
Gerageza itandukaniro riri hagati yikimenyetso kinini nikirangantego, uko uhagarika urutonde rukora, nuburyo bwo gusoma kurangiza impamvu zo gufata troncation.
Ikibazo 1 Bya 8
Ikimenyetso ntarengwa kigarukira iki?
Reba buri kibazo8 ibibazo
Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.
- Ikimenyetso ntarengwa kigarukira iki?
- The total tokens of input plus output in the context window
- The number of tokens the model may generate in a single response
- The number of requests per minute
- Ingano yamagambo yicyitegererezo
- Ihamagarwa ryawe risubiza impamvu yanyuma yuburebure muri API ya OpenAI. Byagenze bite?
- The model finished its answer naturally
- A stop sequence was matched
- The output hit the maximum token limit and was cut off
- The model requested a tool call
- Ninde Anthropic guhagarara_impamvu yerekana icyitegererezo cyemeje igisubizo cyacyo cyuzuye?
- end_turn
- max_tokens
- stop_sequence
- tool_use
- Ni ukubera iki gushiraho ibimenyetso bike byerekana agaciro bidatuma ibisubizo bigufi?
- The API ignores low values
- The model writes as usual and is simply cut off when it reaches the limit
- It makes the model skip the first half of its answer
- It only affects input tokens
- Ese guhuza bihuye mubisanzwe bikubiye mubyanditswe byagarutse?
- Yes, always at the end
- Yes, but only when streaming
- No, it is normally left out of the output
- Only if it is a single character
- A reasoning model returns an empty visible answer. What is a likely cause described in the guide?
- The stop sequence was too long
- Hidden reasoning tokens used up the output limit
- The context window was too large
- The prompt was written in lowercase
- Which stop sequence is the best choice for ending after a structured answer block?
- i
- a single space
- </answer>
- .
- When streaming a response, when should your code decide whether the output was truncated?
- After the first chunk arrives
- When the finish reason arrives at the end of the stream
- Before sending the request
- Streaming responses cannot be truncated