Save the model and load it as model.pth including the model state

Hi all,

I am trying to save the model in PyTorch by using the below code:

model=utils.get_model(self.model){#‘model_state_dict’: model,
#added new
‘model_state_dict’: model.state_dict(),
}, os.path.join(self.checkpoint, ‘model_{}.pth’.format(task_id)))

I am able to load the model successfully with no issues in my app. The model is been saved in to a pth file.

My second step is to take the saved model model.pth and load it via the code below into another application:


It is giving me the below error:

RuntimeError Traceback (most recent call last)
----> 1 model.load_state_dict(torch.load("/home/jovyan/.cache/torch/checkpoints/resnext50_32x4d-7cdf4587.pth"))
2 model = model.eval()

/srv/conda/envs/notebook/lib/python3.7/site-packages/torch/nn/modules/ in load_state_dict(self, state_dict, strict)
828 if len(error_msgs) > 0:
829 raise RuntimeError(‘Error(s) in loading state_dict for {}:\n\t{}’.format(
→ 830, “\n\t”.join(error_msgs)))
831 return _IncompatibleKeys(missing_keys, unexpected_keys)

RuntimeError: Error(s) in loading state_dict for Target:
Missing key(s) in state_dict: “conv_layer.0.weight”, “conv_layer.0.bias”, “conv_layer.1.weight”, “conv_layer.1.bias”, “conv_layer.4.weight”, “conv_layer.4.bias”, “conv_layer.5.weight”, “conv_layer.5.bias”, “conv_layer.7.weight”, “conv_layer.7.bias”, “conv_layer.8.weight”, “conv_layer.8.bias”, “conv_layer.11.weight”, “conv_layer.11.bias”, “conv_layer.12.weight”, “conv_layer.12.bias”, “conv_layer.14.weight”, “conv_layer.14.bias”, “conv_layer.15.weight”, “conv_layer.15.bias”, “conv_layer.18.weight”, “conv_layer.18.bias”, “conv_layer.19.weight”, “conv_layer.19.bias”, “conv_layer.21.weight”, “conv_layer.21.bias”, “conv_layer.22.weight”, “conv_layer.22.bias”, “conv_layer.24.weight”, “conv_layer.24.bias”

Therefore in my code I start to explore additional options to add the model_state too,
My question is, isn’t supposed once I use the below code save all my model including the model state?


Apparently not, that’s why I added the below to my code:, os.path.join(self.checkpoint, ‘model_state_{}.pth’.format(task_id)))

However, I am getting this error :frowning:

in save_all_models
‘model_state_dict’: model.state_dict(),
AttributeError: ‘collections.OrderedDict’ object has no attribute ‘state_dict’

Thank you for your help in advance.

For you first trial{
‘model_state_dict’: model.state_dict(),
}, os.path.join(self.checkpoint, ‘model_{}.pth’.format(task_id)))

If you load the checkpoint file and print its keys, you would see

checkpoint = torch.load('...<the path>...')
# output
>>> ['model_state_dict']

So the solution is:

  1. directly saving the model and directly load
# save, '...<path>...')
# load
  1. save the model states with a key and load it with THAT key
# save{
    'mymodel': model.state_dict()
    }, '...<path>...')
# load

And for the error AttributeError: ‘collections.OrderedDict’ object has no attribute ‘state_dict’, just use type(model) to check the type of model.

1 Like

Hi David,

Thank you heaps for your reply,
I updated my code as suggested instead of:

‘model_state_dict’: model,

to{‘model_state_dict’: model.state_dict(),
}, os.path.join(self.checkpoint, ‘model_{}.pth’.format(task_id)))

however, I am still getting the same error:{'model_state_dict': model.state_dict(),
AttributeError: 'collections.OrderedDict' object has no attribute 'state_dict'

I was reading the issues also discussed here:

and I tested what was suggested after the model saved and checkpoint loaded to net , I also saved another version of the modelFull by simply adding:, os.path.join(self.checkpoint, ‘modelFULL_{}.pth’.format(task_id)))
After loading the model:


I am getting an error:
AttributeError: Net object has no attribute Copy

thank you.

Sorry for the late reply.

For the first error, can you check the type(model) is torch.nn.Module instead of collections.OrderedDict ?

As for the second error, there are commonly two ways of saving/loading models in PyTorch.
One is save/load state_dict and the other is save/load the entire model. In your codes, you’re saving the entire model with, ...), however load it in state_dict way:

# WRONG. load state_dict, not match with entirely saving model. 
# CORRECT. Instead, you should load entire model
model = torch.load(“checkpoint/modelFULL_0.pth”)

Here for more details Saving and Loading Models — PyTorch Tutorials 1.9.0+cu102 documentation

1 Like