moving a model from the latest version of PyG to an old version

[davidireland3]

Similar to my question here, is it possible to take a trained model from the latest version and use it in the old version? If it helps, the PyG layers I use are SAGEConv, TopKPooling and global_mean/max_pool. When I tried to load the model in directly I got the error 'no module named torch_geometric.nn.dense.linear' so I am guessing it might not be possible.

Perhaps it is easier for me to install the older version of PyG on the machine I am training on? however I was unable to find the pip/conda command to install a previous version.

[rusty1s]

Does it work when only saving and loading the state_dict?

[davidireland3]

I believe that has worked, thanks a lot!