[Question]Do Graphformer Encoder need to train for get the new representation of graph?

[feyhong1112]

Sorry to post this, I have a big gap knowledge in deep learning. Hope, you will not mind me.

May I move on to my point?

• My work is to get the representation of graphomer model for using in downsteam

• Every time, I run the input graph into model to get the last hidden state of model, I got the different number in last hidden state every time.

• I have also encode the Euclidian in shortest path

  Does it normal or not to have different last hidden state different number everytime? or I have to train the model first?

Thank you for visit my post and clarify my problem if I misunderstand please, forgive for me.