Why is the output of a linear layer different when the batch size is 1?

8

Thank you!
Just to make it clear: there is no option to disable such optimization related to singleton batch dimension, correct?