Using the nn.Transformer during the eval

I have daily data for n locations to predict some weather values for each location daily based on those daily data. My custom model is using the transformer model for it. In the training time I have the ground truth value of that weather variable which I can pass on to the tgt attribute required by the transformer model.Hiowever, during the evaluation/inference phase I will not hace the ground truth value to be sent to the tgt attribute.

Shouldn’t transformer generate the value based on my input features only during the inference without sending the tgt attribute or should I send the placeholder like torch.zeros() for that case? I am confused , please help me with that issue