pytorch

b84e0996 - don't call native namespace directly in autograd formulas (#79946)

don't call native namespace directly in autograd formulas (#79946) Fixes regression introduced by https://github.com/pytorch/pytorch/pull/79872, which also triggered internal build failure: ``` ld.lld: error: undefined symbol: at::native::NestedTensor_unbind(at::Tensor const&, long) >>> referenced by Functions.cpp:9749 (buck-out/gen/ee8c4ff0/xplat/caffe2/gen_aten_libtorch/autograd/generated/Functions.cpp:9749) >>> buck-out/gen/ee8c4ff0/xplat/caffe2/torch_mobile_trainAndroid#android-arm64,compile-pic-Functions.cpp.o8a165b98/autograd/generated/Functions.cpp.o:(torch::autograd::generated::NestedTensorBackward0::apply(std::__ndk1::vector<at::Tensor, std::__ndk1::allocator<at::Tensor> >&&)) clang: error: linker command failed with exit code 1 (use -v to see invocation) ``` Pull Request resolved: https://github.com/pytorch/pytorch/pull/79946 Approved by: https://github.com/malfet, https://github.com/soulitzer, https://github.com/drisspg