Jeffrey Wan

> Add python bindings for torch.Tensor._set_sizes_and_strides(). Note: Do we want to make this available from python? Should we name it "unsafe" or something? Call self._set_sizes_and_strides() on NT during __init__, which...

> Alternatively, I could make this just take _override_storage_size; is there a preference? hmm numel != storage_size, so technically override_storage_size is more accurate

Could there also be cases where a more fine-grained approach is preferred? For example if I have a subclass wrapping a backend, I only want to decompose when the computation...

@pytorchbot merge

@pytorchbot merge

@pytorchbot merge