One model, input the same data twice, different outputs

1

I have a model and an input. In the training stage, when I input the same data twice one by one:
…
output1 = model(inputData)
output2 = model(inputData)
…

output1 and output2 are completely different. Why?

2

Hard to say without linking the model code. Dropout maybe?

3 
seed = 8
torch.manual_seed(seed)
# add your code below...
```
4

  1. In the training stage, do you sure that the model parameters are unchanged?

  2. dropout may have an influence on outputs in the training stage.

5

Thanks! Yes, there is a dropout layer in my model. Besides, I use an RNN, I guess it maybe another reason.

6

Thanks! dropout is the reason.

7

have you try

model.eval()
output1 = model(inputData)
output2 = model(inputData)

dropout only used in training mode, so by changing the model mode to eval, dropout will not be used.