Missing key(s) in state_dict: "module.backbone.layers.0.stage_1.layers.0.weight",

pytorch加载模型时报错

RuntimeError: Error(s) in loading state_dict for DataParallel:

Missing key(s) in state_dict: "module.backbone.layers.0.stage_1.layers.0.weight",

这是因为加载的预训练模型之前使用了torch.nn.DataParallel()，而此时没有使用，所以可以加上该模块或者去掉。

1，加上torch.nn.DataParallel()模块

model = torch.nn.DataParallel(model)

cudnn.benchmark = True

2、创建一个没有module.的新字典，即将原来字典中module.删除掉

model.load_state_dict = torch.load('model/path')

# create new OrderedDict that does not contain `module.`

from collections import OrderedDict

new_state_dict = OrderedDict()

for k, v in state_dict.items():

    # remove `module.`

    name = k[7:] 

    new_state_dict[name] = v

# load params

model.load_state_dict(torch.load(new_state_dict))

3、思想同2

model.load_state_dict({k.replace('module.',''):v for k,v in torch.load('model/path').items()})

更多讨论请参考：

https://discuss.pytorch.org/t/solved-keyerror-unexpected-key-module-encoder-embedding-weight-in-state-dict/1686/3

---

记录pytorch .pth模型转onnx遇到的问题：

1、Missing key(s) in state_dict:

解决方法：

# original saved file with DataParallel

state_dict = torch.load('myfile.pth.tar')

# create new OrderedDict that does not contain `module.`

from collections import OrderedDict

new_state_dict = OrderedDict()

for k, v in state_dict.items():

    name = k[7:] # remove `module.`

    new_state_dict[name] = v

# load params

model.load_state_dict(new_state_dict)

2、RuntimeError: Expected object of backend CPU but got backend CUDA for sequence element 2 in sequence argument at position #1 'tensors'

解决方法：

在GPU上训练的模型在CPU上运行时会出现的问题

加上

model = torch.load('model.pth',map_location='cpu')

3、AttributeError: 'xxxx' object has no attribute 'seek'. You can only torch.load from a file that is seekable. Please pre-load the data into a buffer like io.BytesIO and try to load from it instead.