Assume that two transformer modes$($eg$.$ BERT$)$ with varied hyper parameter tuning followed by training each resulted in below contextual embedding for the$-$word "bank$.$Which of the models IS more informative? Justify your answer with an appropriate numerical metric.

Model A

A$1$bank$=[0\mathrm{.}1,0\mathrm{.}5,2]$

A$2$bank$=[0\mathrm{.}5,2,0\mathrm{.}1]$

Model B:

B$1$bank$=[0\mathrm{.}2,0\mathrm{.}45,-0\mathrm{.}5]$

B$2$bank$=[0\mathrm{.}5,0\mathrm{.}2,-0\mathrm{.}45]$