Simple tutorial of pytorch avoid use of inner function

1

I look for full simple tutorial/example of pytorch avoid use of inner function except autograd such as…

Linear.

I have shown tutorials which are most works are automated as below.

class Net(nn.Module):
    def __init__(self):
        super.__init__()
        fc1 = linear(10,20)
   
    def forward(x):
        return fc1(x)

net =Net()

...
loss.backward()
opim.step()

I want to learn manual code(hard code) as below.

class Net(nn.Module):
    def __init__(self):
        super.__init__()
        w1 = torch.rand(10)
        b1 = torch.zeros(10)
   
    def forward(x):
        return torch.mm(x, w1) + b1

net =Net()
...
loss.backward()
opim.step()

etc…

2

Hi,

Sorry I’m not sure to understand your question.
Could you rephrase it please?

3

sorry for that
additionally described

4

So first, you never want to use .data.

If you want to do your weight update manually, you can do:

loss.backward()
with torch.no_grad():
  w1 -=  learning_rate * w1.grad