Unverified50% confidenceFactExact time
推理延迟通常分解为首token延迟（TTFT）和后续token生成吞吐量两个核心指标
1
Sources
50%
Confidence
Long-term
Relevance
7/13/2026
First Seen
Sources
扩散语言模型崛起：DiffusionGemma实测速度碾压自回归模型
redditr/GoogleGeminiAI7/11/2026
Related Claims
Unverified对流式响应场景，应重点关注首Token延迟（TTFT）而非整体完成时间81% similarUnverified传统混合部署中Prefill阶段的大批量计算会阻塞Decode阶段的token生成，导致TTFT波动剧烈76% similarUnverifiedtokens/s指标可细分为首token延迟（TTFT）和生成速率（Generation Throughput）两个子维度71% similarUnverified推理延迟公式为：延迟 = (起草时间 + 验证时间) ÷ 每轮被接受的token数66% similarUnverified理论分析表明，在时间步长为T的轨迹上，行为克隆的累积误差以O(T²)的速度增长64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/497411API
curl https://kongchang.com/api/v1/knowledge/claims/497411MCP
get_claim(id=497411)