フリー問題

問題文

A developer needs a tensor that has the same values as an existing tensor but is excluded from gradient tracking, and they must not modify the original. Which expression is appropriate?

選択肢

1. torch.no_grad(x), which marks the tensor as excluded from every graph built after that point.
2. x.item(), which strips the graph information while keeping the tensor usable in later arithmetic.
3. x.requires_grad_(False), which is the documented way to obtain a detached copy while leaving the original tensor untouched for the rest of the program.
4. x.detach(), which returns a tensor sharing the data but detached from the graph.