Can Pytorch run all in double precision (float64)?

(Son Nam Nguyen) #1

Hello all,

I’m from Tensorflow background, and I’m considering switch to Pytorch.
Tensorflow is designed to work mostly with single precision (float32) while what I want is double (float64).
Can someone confirm that Pytorch can run entire training on double precision or not ?

(Simon Wang) #2

Yes, PyTorch can use double tensors throughout entire training.