yxd886

Support overlapping NCCL collective communication with compute on GPU

6

Hello, I am conducting an experiment using a server with 4 GPUs. I just run the script spmd_mnist_classifier_fromscratch.py under the example folder. Then I used Nvidia system to profile the...

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

1

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

Can torchscript dump backward graph?

1

Can torchscript dump backward graph? When using TorchScript, can the backward graph be dumped and compiled using MLIR pass ? Or only forward graph is supported?