# Method 2: Automatic gradient computation (more common)
loss.backward()  # Compute all gradients automatically
print(f"Weight gradient: {w1.grad.item():.4f}")
print(f"Bias gradient: {b.grad.item():.4f}")

# Manual parameter update
learning_rate = 0.1
with torch.no_grad():  # Disable gradient tracking for updates
    w1 -= learning_rate * w1.grad
    b -= learning_rate * b.grad

# Clear gradients for next iteration
w1.grad.zero_()
b.grad.zero_()

        __
        __