Unverified50% confidenceFactExact time
输入Token可以在prefill阶段并行处理，而输出Token必须在decode阶段逐个串行生成
1
Sources
50%
Confidence
Long-term
Relevance
8/24/2026
First Seen
Sources
Related Entities
Related Claims
Unverified输出Token定价高于输入Token是因为Prefill阶段可并行矩阵运算而Decode阶段必须串行自回归解码，Decode阶段GPU吞吐量约为Prefill阶段的1/6到1/1077% similarUnverifiedDecode 阶段属于访存密集型操作，每生成一个新 token 需从显存读取全部历史 token 的 KV Cache75% similarUnverified自回归解码具有串行特性，每个Token的生成依赖前一个Token，难以并行74% similarUnverifiedConstrained Decoding computes the set of valid tokens at each token step of autoregressive generation based on a finite state machine or context-free grammar and renormalizes the probabilities, mathematically guaranteeing that the output conforms to the target schema.73% similarUnverified自回归生成每个Token都需要完整的前向传播，而输入只需一次编码，这是输出比输入贵的成本结构原因71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/796472API
curl https://kongchang.com/api/v1/knowledge/claims/796472MCP
get_claim(id=796472)