Pytorch does not update weight after optimizer.step()

I'm coding Neural Network with Pytorch but i have a prolem. Pytorch does not update weight after optimizer.step().
This is my code:
Neuralnetwork.py

import torchclass NeuralNetwork(torch.nn.Module):    def __init__(self,                 size_of_hidden_layer:tuple=(5, 5)                 ) -> None:        super().__init__()        self.number_of_hidden_layers = size_of_hidden_layer[0]        self.number_of_node_each_hidden_layers = size_of_hidden_layer[1]        self.hidden_layers = None        self.setActFunc('relu')    def fit(self, X, y):        if self.hidden_layers != None:            return        self.hidden_layers = torch.nn.ModuleList()        self.hidden_layers.append(torch.nn.Linear(in_features=X.shape[1],                                                   out_features=self.number_of_node_each_hidden_layers))        for _ in range(self.number_of_hidden_layers-1):            self.hidden_layers.append(torch.nn.Linear(in_features=self.number_of_node_each_hidden_layers,                                                out_features=self.number_of_node_each_hidden_layers))        self.output = torch.nn.Linear(in_features=self.number_of_node_each_hidden_layers,                                          out_features=y.shape[1])    def setActFunc(self, act_func:str):        if act_func == 'relu':            self.activation = torch.nn.ReLU()        if act_func == 'leaky':            self.activation = torch.nn.LeakyReLU()        if act_func == 'sigmoid':            self.activation = torch.nn.Sigmoid()    def forward(self, X):        A = X        for layer in self.hidden_layers:            Z = layer(A)            A = self.activation(Z)        Z = self.output(A)        y = torch.nn.Softmax(dim=1)(Z)        return (y >= 0.5) * 1.0

Trainer.py

import torchfrom torch.nn import Module, MSELossfrom torch.optim import SGD, Adamclass Trainer():    def __init__(self) -> None:        pass    def train(self, model:Module, X:torch.Tensor, y:torch.Tensor, epochs=10, lr=0.01, act_func='relu'):        train_model = model        train_model.setActFunc(act_func)        train_model.fit(X, y)        optimizer = Adam(params=train_model.parameters(), lr=lr)        loss_func = MSELoss()        for epoch in range(epochs):            train_model.train()            optimizer.zero_grad()            pred = train_model(X)            loss = loss_func(pred.squeeze(), y.squeeze())            loss.requires_grad = True            loss.backward()            optimizer.step()

main.py

import torchimport NeuralNetwork as NNimport Trainerimport numpy as npif __name__ == '__main__':    torch.random.manual_seed(42)    # Create data    sample = 10    feature = 100    X = torch.rand(size=(sample, feature), requires_grad=True) * 10    y = (torch.rand(size=(sample, 1), requires_grad=True) >= 0.5).type(torch.float)    # Check if model work    model = NN.NeuralNetwork(size_of_hidden_layer=(1, 15))    model.fit(X, y)    # Check weight before train    print(model.state_dict())    # Train model    trainer = Trainer.Trainer()    trainer.train(model, X=X, y=y, epochs=1000)    # Check weight after train    print(model.state_dict())

When i run the file main.py, then the result is:

Before train:OrderedDict([('hidden_layers.0.weight', tensor([[ 0.0260, -0.0809, -0.0605, ..., 0.0265, -0.0153, 0.0411],[-0.0417, -0.0943, -0.0384, ..., -0.0058, 0.0766, 0.0625],[ 0.0919, -0.0732, 0.0643, ..., -0.0694, 0.0010, 0.0003],...,[ 0.0034, 0.0239, -0.0124, ..., -0.0813, -0.0271, 0.0883],[ 0.0391, -0.0119, 0.0206, ..., -0.0621, -0.0815, 0.0218],[-0.0551, 0.0284, 0.0847, ..., -0.0059, 0.0009, -0.0260]])), ('hidden_layers.0.bias', tensor([-0.0683, 0.0907, -0.0331, 0.0477, 0.0696, -0.0657, -0.0125, 0.0578,0.0969, -0.0126, 0.0725, -0.0785, -0.0650, 0.0822, -0.0095])), ('output.weight', tensor([[ 0.2551, 0.0188, -0.1071, 0.2452, 0.2258, 0.0240, -0.1565, -0.0112,-0.1242, 0.1881, 0.0120, 0.0544, 0.1566, 0.0855, -0.1612]])), ('output.bias', tensor([0.2145]))])

After train:OrderedDict([('hidden_layers.0.weight', tensor([[ 0.0260, -0.0809, -0.0605, ..., 0.0265, -0.0153, 0.0411],[-0.0417, -0.0943, -0.0384, ..., -0.0058, 0.0766, 0.0625],[ 0.0919, -0.0732, 0.0643, ..., -0.0694, 0.0010, 0.0003],...,[ 0.0034, 0.0239, -0.0124, ..., -0.0813, -0.0271, 0.0883],[ 0.0391, -0.0119, 0.0206, ..., -0.0621, -0.0815, 0.0218],[-0.0551, 0.0284, 0.0847, ..., -0.0059, 0.0009, -0.0260]])), ('hidden_layers.0.bias', tensor([-0.0683, 0.0907, -0.0331, 0.0477, 0.0696, -0.0657, -0.0125, 0.0578,0.0969, -0.0126, 0.0725, -0.0785, -0.0650, 0.0822, -0.0095])), ('output.weight', tensor([[ 0.2551, 0.0188, -0.1071, 0.2452, 0.2258, 0.0240, -0.1565, -0.0112,-0.1242, 0.1881, 0.0120, 0.0544, 0.1566, 0.0855, -0.1612]])), ('output.bias', tensor([0.2145]))])