Copying Scalars from Torch to Rust?

[Yash Rachamalla]

Currently, when I try to copy a scalar value from a 0 dim tensor with double_value(&[0]), it returns the following error

called `Result::unwrap()` on an `Err` value: Torch("select() cannot be applied to a 0-dim tensor.\nException raised from select_symint at /build/python-pytorch/src/pytorch-cuda/aten/src/ATen/native/TensorShape.cpp:1827 (most recent call first):\nframe #0: c10::Error::Error(c10::SourceLocation, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >) + 0xa2 (0x73af2cd9c8a2 in /usr/lib/libc10.so)\nframe #1: <unknown function> + 0x7a9854 (0x73af21da9854 in /usr/lib/libtorch_cpu.so)\nframe #2: <unknown function> + 0x23130c5 (0x73af239130c5 in /usr/lib/libtorch_cpu.so)\nframe #3: at::_ops::select_int::redispatch(c10::DispatchKeySet, at::Tensor const&, long, c10::SymInt) + 0xc5 (0x73af234e5e75 in /usr/lib/libtorch_cpu.so)\nframe #4: <unknown function> + 0x438bc40 (0x73af2598bc40 in /usr/lib/libtorch_cpu.so)\nframe #5: <unknown function> + 0x438bf6c (0x73af2598bf6c in /usr/lib/libtorch_cpu.so)\nframe #6: at::_ops::select_int::redispatch(c10::DispatchKeySet, at::Tensor const&, long, c10::SymInt) + 0xc5 (0x73af234e5e75 in /usr/lib/libtorch_cpu.so)\nframe #7: <unknown function> + 0x3d36167 (0x73af25336167 in /usr/lib/libtorch_cpu.so)\nframe #8: <unknown function> + 0x3d3688c (0x73af2533688c in /usr/lib/libtorch_cpu.so)\nframe #9: at::_ops::select_int::call(at::Tensor const&, long, c10::SymInt) + 0x1b1 (0x73af235535e1 in /usr/lib/libtorch_cpu.so)\nframe #10: at::Tensor::select(long, long) const + 0x53 (0x5fdc7c88649d in ./Milkshake)\nframe #11: <unknown function> + 0x1e11bd (0x5fdc7c8861bd in ./Milkshake)\nframe #12: <unknown function> + 0x1ef078 (0x5fdc7c894078 in ./Milkshake)\nframe #13: <unknown function> + 0x1e8e19 (0x5fdc7c88de19 in ./Milkshake)\nframe #14: <unknown function> + 0x1b099f (0x5fdc7c85599f in ./Milkshake)\nframe #15: <unknown function> + 0x1b0a75 (0x5fdc7c855a75 in ./Milkshake)\nframe #16: <unknown function> + 0x388c5 (0x5fdc7c6dd8c5 in ./Milkshake)\nframe #17: <unknown function> + 0x4a4f9 (0x5fdc7c6ef4f9 in ./Milkshake)\nframe #18: <unknown function> + 0x5d985 (0x5fdc7c702985 in ./Milkshake)\nframe #19: <unknown function> + 0x5f771 (0x5fdc7c704771 in ./Milkshake)\nframe #20: <unknown function> + 0x4f19b (0x5fdc7c6f419b in ./Milkshake)\nframe #21: <unknown function> + 0x367de (0x5fdc7c6db7de in ./Milkshake)\nframe #22: <unknown function> + 0x452a1 (0x5fdc7c6ea2a1 in ./Milkshake)\nframe #23: <unknown function> + 0x44b961 (0x5fdc7caf0961 in ./Milkshake)\nframe #24: <unknown function> + 0x4527a (0x5fdc7c6ea27a in ./Milkshake)\nframe #25: <unknown function> + 0x60e5e (0x5fdc7c705e5e in ./Milkshake)\nframe #26: <unknown function> + 0x25cd0 (0x73af21357cd0 in /usr/lib/libc.so.6)\nframe #27: __libc_start_main + 0x8a (0x73af21357d8a in /usr/lib/libc.so.6)\nframe #28: <unknown function> + 0x35eb5 (0x5fdc7c6daeb5 in ./Milkshake)\n")

What is the intended way to copy the values of tensors?

[Iago-lito]

Hi, if your tensor has no dims, maybe .double_value(&[]); ?