Create a scalar tensor in LibTorch, Part 1

				
					#include <iostream>
#include <ATen/ATen.h>
#include <torch/data/datasets/tensor.h>
#include <c10/core/TensorOptions.h>
#include <ATen/core/TensorBase.h>

int main()
{
	// Scalar
	std::cout << "-------------- Scalar --------------\n";
	const at::Tensor my_scalar = torch::tensor(68.0f).cuda();
	std::cout << "Scalar: " << my_scalar << std::endl;
	std::cout << "Scalar Dimension: " << my_scalar.ndimension() << std::endl;
	std::cout << "Scalar Cuda: " << std::boolalpha << my_scalar.is_cuda() << std::endl;
	std::cout << "Scalar as tensor shape: " << _shape_as_tensor(my_scalar).cuda() << std::endl;
	std::cout << "Scalar element sizes: " << my_scalar.element_size() << std::endl;
	std::cout << "Scalar sizes: " << my_scalar.sizes() << std::endl;
}

This is a short C++ code using the libtorch library to create a scalar tensor called my_scalar with a value of 68.0f, which is stored in the GPU memory.

				
					const at::Tensor my_scalar = torch::tensor(68.0f).cuda();

We then use several methods to extract information about this scalar tensor:

my_scalar.ndimension(): This method returns the number of dimensions in the tensor, which for a scalar tensor is always zero. Here, it will output “Scalar Dimension: 0”.

				
					std::cout << "Scalar Dimension: " << my_scalar.ndimension() << std::endl;

my_scalar.is_cuda(): This method returns a true or false value indicating whether the tensor is stored in GPU memory or not. Since we created my_scalar in GPU memory using the cuda() method, the output will be “Scalar Cuda: true”.

				
					std::cout << "Scalar Cuda: " << std::boolalpha << my_scalar.is_cuda() << std::endl;

_shape_as_tensor(my_scalar).cuda(): We use a helper function _shape_as_tensor() to create a tensor that represents the shape of my_scalar, which is an empty tensor with zero dimensions. We then move this tensor to the GPU memory using cuda(). The output will be “Scalar as tensor shape: []”.

				
					std::cout << "Scalar as tensor shape: " << _shape_as_tensor(my_scalar).cuda() << std::endl;

my_scalar.element_size(): This method returns the size of each element in the tensor in bytes. Since my_scalar is a scalar tensor with a single floating-point value, its element size is 4 bytes (for a 32-bit floating-point value). The output will be “Scalar element sizes: 4”.

				
					std::cout << "Scalar element sizes: " << my_scalar.element_size() << std::endl;

my_scalar.sizes(): This method returns a vector of integers representing the size of each dimension in the tensor. Since my_scalar has no dimensions, its size vector is empty. The output will be “Scalar sizes: []”.

				
					std::cout << "Scalar sizes: " << my_scalar.sizes() << std::endl;

Finally, the code uses std::cout to print out the results of each method call. The output will be:

				
					-------------- Scalar --------------
Scalar: 68
Scalar Dimension: 0
Scalar Cuda: true
Scalar as tensor shape: []
Scalar element sizes: 4
Scalar sizes: []

I hope you find this post useful.

Good luck!.

Photo by Suzi Kim on Unsplash

Learning Playground

Create a tensor in LibTorch

April 7, 2023

LibTorch makes it easy to work with tensors in C++. In this post, we'll walk through the process of creating and using tensor in LibTorch. ...

Learning Playground

Create Matrix in LibTorch

April 3, 2023

LibTorch makes it easy to work with vectors in C++. In this post, we'll walk through the process of creating and using Matrix in LibTorch. ...

Learning Playground

Create Vector in LibTorch

April 2, 2023

LibTorch makes it easy to work with vectors in C++. In this post, we'll walk through the process of creating and using vectors in LibTorch.

Learning Playground

Create a scalar tensor in LibTorch, Part 2

March 30, 2023

The provided code initializes a scalar value in LibTorch, which is a single numeric value used in computations.

Learning Playground