Unverified50% confidenceFactExact time
自回归推理时batch size通常为1,每个token生成只做一次矩阵-向量乘法,算术强度极低,导致memory-bound问题,单用户场景下GPU利用率往往不到10%
1
Sources
50%
Confidence
Long-term
Relevance
8/27/2026
First Seen
Sources
Related Entities
Cite This Claim
Stable URI
https://kongchang.com/claim/810211API
curl https://kongchang.com/api/v1/knowledge/claims/810211MCP
get_claim(id=810211)