I am training a CNN+LSTM model for binary classification with dropout layer (0.3 dropping ratio). And I am using nn.CrossEntropyLoss along with optimizer as Adam. I can see my output values keep changing during my model.train() process, but during my model.eval(), the output value keeps repeating for different data, like exactly the same value. For example tensor([0.9600, -1.2629]), no better what data I feed into. How to solve this issue? Also, I tried to disable the dropout layer, but still the outputs are the same throughout. Thanks in advance.