Loss decreasing rapidly but accuracy is constant

#1

I am training a model in pytorch-geometric, for graph classification. Even with a small model, I am seeing my loss go to very low values, but the accuracy is fixed. How can this be?

Epoch: 001, Loss: -170941725.82055, Train Acc: 0.82073, Test Acc: 0.84599
Epoch: 002, Loss: -765036750.91172, Train Acc: 0.82073, Test Acc: 0.84599
Epoch: 003, Loss: -1879149609.86397, Train Acc: 0.82073, Test Acc: 0.84599
Epoch: 004, Loss: -3669840125.40666, Train Acc: 0.82073, Test Acc: 0.84599
Epoch: 005, Loss: -6205604898.08394, Train Acc: 0.82073, Test Acc: 0.84599
#2

I think we’ll need more of your code to be able to help. Would you be able to show all of it ?