paper: correct CI numbers from audit
2 changed files +1 −1
modified
paper/main.pdf
+0 −0
Binary file not shown.
modified
paper/sections/results.tex
+1 −1
@@ -164,7 +164,7 @@ the laptop (\code{benchmarks/raw/gateG/energy\_laptop/}). All-resident decoding | ||
| 164 | 164 | machine waits on the disk most of the time; the probe prefetcher brings this to 0.44~J at 18~W. Net of the 13.2~W idle power the |
| 165 | 165 | ordering reverses between the two cold configurations (0.086 vs 0.119~J), which is the usual ambiguity of attributing idle power. |
| 166 | 166 | The same run also gives the first \emph{positive} H1 evidence: over 256-token generations the probe reaches a hit rate of 0.84 and |
| 167 | −42.0~\tps{} against 31.9 for demand loading ($+10.1$ [$+7.5$, $+12.4$]~\tps{}, Table~\ref{tab:audit}), whereas the 64-token runs | |
| 167 | +42.0~\tps{} against 31.9 for demand loading ($+10.1$ [$+5.2$, $+16.9$]~\tps{}, Table~\ref{tab:audit}), whereas the 64-token runs | |
| 168 | 168 | of Section~\ref{sec:results-prefetch} showed a tie: the per-token install overhead is constant while the benefit of a warm, well-predicted |
| 169 | 169 | table grows with sequence length. H1 is therefore \emph{length-dependent} on this hardware, and the cluster-level Pareto study remains |
| 170 | 170 | pending. |
| 171 | 171 | |