流式输出Streaming Output
Streaming Output
回答边生成边推送：你看到逐字出现，而不是等全文算完。Answers stream out as they are generated: you read word by word instead of waiting for the full text.
自回归模型逐 token 生成，流式模式把每个 token 即时推送给客户端。
常见实现是 SSE（服务器推送事件流）：服务器以事件流持续下发。
价值在首字延迟：长回答不必等全部算完才开始读。
中途可停：取消请求即停止生成与计费。
Autoregressive models generate token by token; streaming pushes each token to the client immediately.
A common implementation is SSE (server-sent events): the server delivers a continuous event stream.
The value is time-to-first-token: long answers don't wait for full computation before you start reading.
It can be stopped mid-way: canceling the request stops generation and billing.
聊天界面的逐字效果、为什么能中途打断、为什么按输出 token 计费——都来自这个机制。The word-by-word effect in chat UIs, why you can interrupt, why output tokens are billed — all come from this mechanism.
回到名词地图看第2站的其他 35 个词 →Back to the map: the other 35 terms in station 2 →