Set 128K eager baseline and organize deployment documentation
This commit is contained in:
@@ -1,5 +1,7 @@
|
||||
# CUDA Graph 性能剖析:图已生效,主要收益受 GPU 工作占比限制
|
||||
|
||||
> 历史实验记录:保留当时的配置和恢复状态。当前部署已改为 128K eager 基线,见 [当前状态](README.md)。
|
||||
|
||||
2026-09-18,Asia/Shanghai。方案与脚本见 [profile128k](../experiments/profile128k/plan.md)。
|
||||
|
||||
CUDA Graph 确实被重放,并非参数失效。本轮短回答生成速度中位数由 **29.954 → 31.316 tokens/s(约 +4.5%)**,
|
||||
|
||||
Reference in New Issue
Block a user