Model.eval() but layer have different weight for each trial

I load pytorch model, and use model.load_state_dict(). And use model.eval()
And I print weight of linear layer in order to check the weight of model
However, the output of weight is continuously if I execute python file again.
Thus the output of ‘forward’ is different for same input

How is this possible?

In addition, I have two pytorch model (pt file). One is made from my local computer. And the other is made from google colab. Interestingly, pt file from colab has constant weight if I execute file again. But pt file from my local computer has different weight if I execute file again.
Thus the outputs are always same if I use model from colab. But the outpust are always different if I use model from local computer

Could you check if your script is saving the (updated) state_dict in each run?