I tried to:
exp(G)/(STEADY_STATE(G))=((exp(G(-1))/STEADY_STATE(G))^(gammaG))((exp(B)exp(PI)(STEADY_STATE(Y))/(exp(Y(-1))(STEADY_STATE(B)))))^((1-gammaG)*phiG)*eps_G;

I understand that the steady state of G is not log-linearized form, but could not figured out how to change it. Could you please give some suggestions.

Here is something inconsistent. G is a log-level and \exp(G) is the corresponding level.

You are dividing the level by the steady state of the log-level.

If you have a steady_state_model-block you could define a variable GG=exp(G); if you already have the steady state of the log-level. Then you could use GG in your steady_state operator.