Ricky Chen
Ricky Chen
I likely won't have time to support this, but will gladly take a PR for it!
You should upgrade your PyTorch. The more recent versions (I think versions >= 1.6?) have .real implemented for non-complex tensor types. In the meantime, I'll work on a fix for...
Does the conversation from #166 help? The quantity you're seeing is for computing dL/dt0, which intuitively, has the opposite gradient direction compared to t1, because t0 *shortens* the integration time...