Hello Hossein,
seeding with existing documents was a part of gensim some versions
back, in the code that was directly ported from LDA-C. It was meant to
improve convergence (although the final, converged result was the
same).
In recent versions, that code was replaced by a more efficient
algorithm which doesn't use seeding anymore. It is an online
algorithm, so you could say it does "seeding" automatically, in a more
principled manner.
HTH,
Radim