thaki-AI's picture
paper: 2026-07-28 KV Cache Tiering Meets Prefill-Decode Disaggregation: Mapping the Cost-Latency F
a834836 verified
Raw
History Blame Contribute Delete
996 Bytes
% This file was created with matplot2tikz v0.5.4.
\begin{tikzpicture}
\begin{axis}[legend entries={{G*}},
height=6.8cm,
legend cell align={left},
legend style={
fill opacity=1.0,
draw opacity=1,
text opacity=1,
at={(0.5,-0.18)},
anchor=north,
draw=none
},
tick pos=both,
width=12cm,
xmajorgrids,
xmin=-0.35, xmax=7.35,
xtick style={color=black},
xtick={0,1,2,3,4,5,6,7},
xticklabels={0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9},
ylabel={log10 optimal capacity G* (GB)},
ymajorgrids,
ymin=-0.34085, ymax=6.67385,
ytick style={color=black},
ytick={-1,0,1,2,3,4,5,6,7},
yticklabels={
\(\displaystyle {-1}\),
\(\displaystyle {0}\),
\(\displaystyle {1}\),
\(\displaystyle {2}\),
\(\displaystyle {3}\),
\(\displaystyle {4}\),
\(\displaystyle {5}\),
\(\displaystyle {6}\),
\(\displaystyle {7}\)
}
]
\addplot [black, mark=o, mark size=1.5, mark options={solid,fill opacity=0}]
table {%
0 -0.022
1 0.226
2 0.472
3 0.76
4 1.148
5 1.754
6 2.922
7 6.355
};
\end{axis}
\end{tikzpicture}