أساسيات PyTorch

# CPU version (fine for learning)
pip install torch torchvision

# With an NVIDIA GPU: get the exact command from pytorch.org
# Easiest: use Google Colab -> Runtime -> Change runtime type -> GPU

import torch

t = torch.tensor([[1.0, 2.0, 3.0],
                  [4.0, 5.0, 6.0]])
print(t.shape, t.dtype)

print(t * 10)                 # element-wise, like NumPy
print(t.sum(dim=0))           # dim = axis in NumPy
print(t @ t.T)                # matrix multiplication

zeros = torch.zeros(2, 3)
rand = torch.rand(2, 3)       # uniform random 0..1
print(rand.shape)

import torch

device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using:", device)          # on Colab with a GPU: cuda

x = torch.rand(1000, 1000).to(device)   # move data to the GPU (if available)
y = x @ x                               # runs on that device
print(y.device)

import torch

x = torch.tensor(3.0, requires_grad=True)
y = x ** 2
y.backward()                 # compute dy/dx
print("dy/dx at x=3:", x.grad.item())

# Works through any chain of operations
w = torch.tensor(2.0, requires_grad=True)
loss = (w * 4 - 10) ** 2     # prediction w*4, target 10
loss.backward()
print("gradient of loss w.r.t. w:", w.grad.item())