torch：存储网络结构，load网络结构

存储torch模型到path：

torch.save(NN_model.state_dict(), 'path')

加载模型并且设置为eval()

NN_model = MyModel() # load the structure of the NN model 
state_dict = torch.load('model_state_dict.pth')
NN_model.load_state_dict(state_dict)
NN_model.eval()

以PINN为例，假设模型的输入是$(x,y)$, 输出是$u_{\theta }(x,y)$, 如果想要求$\frac{\partial u}{\partial x}$ 以及$\frac{\partial u}{\partial y}$：

u= self.NN_model(inputs)
U = torch.sum(u)
U.backward(retain_graph=True)

u_x = torch.autograd.grad(U, x, grad_outputs=torch.ones_like(U), retain_graph=True, create_graph=True,allow_unused=True )[0]
u_y = torch.autograd.grad(U, y, grad_outputs=torch.ones_like(U), retain_graph=True, create_graph=True,allow_unused=True )[0]