The important part is self-awareness, how a neuronal network can develop one, can we say that machine that only imitates self-consciousness are sentient?

Sentience (classically) means to experience sensations, i.e. feelings that have to do with sensory input. There's zero reason to assume a transformer trained on text can have that. There's also zero reason to assume that it has a meaningful concept of itself, which can easily be seen when it mimics talking about oneself.

this is more of a rhetorical question, not a stone in your garden