첫 torch.tensor 만들기

A tensor is PyTorch's core data container, a grid of numbers much like a NumPy array but ready to run on a GPU. 🧮

The simplest way to make a tensor is torch.tensor from a Python list. PyTorch copies your numbers into a fast numerical grid.

Printing a tensor shows its values wrapped in tensor(...), so you always know you are looking at PyTorch data, not a plain list.

The shape tells you the size along each dimension. A flat list of three values has a shape of three.

Nest lists to make a grid. This matrix has two rows and three columns, so its shape reads as two by three.

Every tensor has a dtype, the kind of number it stores. Whole numbers become int64, decimals become float32 by default.

Networks learn with decimals, so most model data is float32. Add a decimal point and PyTorch picks the float type for you.

You can request a type directly with the dtype argument, which is handy when you need floats from integer input.

Need a blank tensor of a given size? torch.zeros fills the shape you ask for with zeros, perfect as a starting buffer.

Model weights often start random. torch.rand gives a tensor of the shape you choose, filled with values between zero and one.

The number of dimensions is the tensor's rank. Read it with .ndim: a vector is rank one, a matrix is rank two.

You created tensors from lists, read their shape and dtype, and made zeros and random grids. This is the data deep learning runs on. ✅