How to assign a tensor with another tensor?

1 
import torch
from LeNet5 import LeNet5

model = torch.load('LeNet5.pth')  # 加载模型
layerWeight = model.state_dict()['conv1.0.bias'].clone()
print("----layerWeight is ",layerWeight)
print("----model is ", model.state_dict()['conv1.0.bias'])

model.state_dict()['conv1.0.bias'][0] = 1

print("++++layerWeight is", layerWeight)
print("++++model is", model.state_dict()['conv1.0.bias'])

print("layerWeight type is ",type(layerWeight))
print("model type is",type(model.state_dict()['conv1.0.bias']))

model.state_dict['conv1.0.bias'] = layerWeight


print("****layerWeight is",layerWeight)
print("****model is", model.state_dict()['conv1.0.bias'])

Above the code, I try model.state_dict[‘conv1.0.bias’] = layerWeight,(model.state_dict[‘conv1.0.bias’] and layerWeight are both the type of tensor)but get the error in the picture below.

image
image1380×351 52.2 KB

Please hope you could tell me how to solve it.

2

The error is raised, since you are not calling the model.state_dict() method.
That being said, a direct assignment will most likely not work and you could use:

sd = model.state_dict()
sd[key] = layerWeight
model.load_state_dict(sd)

instead.

2 Likes
3

Why are you so experienced! It’s an amazing thing for me because I could find nothing about that in pytorch document.

1 Like