Closed Gy-Lu closed 1 year ago
Our previous chunk code is based on inference, which means that no gradient and inplace is just ok.
However, in training, inplace is forbidden by Pytorch. And I made some modifications on chunk=None cases to avoid inplace.
Our previous chunk code is based on inference, which means that no gradient and inplace is just ok.
However, in training, inplace is forbidden by Pytorch. And I made some modifications on chunk=None cases to avoid inplace.