Changing transformation applied to data during training

11

Could you check the data type of your input?
We have a similar issue recently here, where the inputs were FloatTensors and thus an overflow occurred.