Does we save a random parameters if no train for model

1

I try to create a simple Mlp model without train, and then use torch.save to save the model parameters （they are random values which can be checked by print(param_tensor,‘\t’,model.state_dict()[param_tensor])）, so in fact, I only save a lots of random values, and it is not a tuned parameters?

PS: I also try to open the saved data mlp-simple1.pth by vim, but it only shows a bunch of garbled characters.

z00443407@llvm:~/source/test/simple_mlp$ cat save_model1.py 
import torch
from torch import nn
import torch_mlir

name : str
class MlpModel(torch.nn.Module):
    def __init__(self):
        super(MlpModel, self).__init__()
        self.net = nn.Sequential(nn.Flatten(),
                    nn.Linear(784, 256),
                    nn.ReLU(),
                    nn.Linear(256, 10))

    def forward(self, x):
        return self.net(x)


# 保存模型
model = MlpModel()
def save_model():
    path = "./mlp-simple1.pth"
    for param_tensor in model.state_dict():
        print(param_tensor,'\t',model.state_dict()[param_tensor].size())
        # print(param_tensor,'\t',model.state_dict()[param_tensor]) # Random value
    torch.save(model.state_dict(), path)

save_model()
2

Yes, saving the state_dict directly after initializing the model will save it’s initial random values, which were not trained yet.