I am willing to optimize dimer in both, singlet and triplet, excited state. But I don't know whether it is possible from gaussian, and if yes, how to do.
You are asking about doing eom-ccsd? this is how to do so for two excited states and optimizing with respect to the 1st excited state, the states are singlets as am showing under the title after the 0 charge (neutral molecule) using 1, which is for singlet , for triplet states use 3 instead of 1
opt eomccsd=(nstate=2,root=1)
Title
0 1
But keep in your mind that eom-ccsd analytic gradients are only implemented in the recent Gaussian16 version, older versions will perform numerical gradients, which is extremely expensive