why implement class XLATensor from scratch rather than using torch::lazy::LazyTensor

Open yxd886 opened this issue 3 years ago • 1 comments

❓ Questions and Help

Hello, I am curious about the reason why implement class XLATensor from scratch rather than using torch::lazy::LazyTensor?

Jul 27 '22 05:07 yxd886

well.. it is the other way around. We implemented the XLATensor ~3 years ago and the whole Lazy Tensor system got upstreamed to Pytorch and becomes the LTC(LazyTensor). We are migrating to upstream LTC and will likely have XLATensor inherits from torch::lazy::LazyTensor.

Jul 27 '22 05:07 JackCaoG