oneflow.Tensor.retain_grad¶ Tensor.retain_grad()¶ Enables this Tensor to have their grad populated during backward(). This is a no-op for leaf tensors. Compatible with PyTorch.