get RuntimeError: Error(s) in loading state_dict for E2E: when doing test on model from training

[icang1694]

here's my config for training, i use FUNSD dataset
%run 'main.py' --add-geom --add-embs --add-hist --add-visual --add-eweights --src-data 'FUNSD' --gpu 0 --edge-type 'fully' --node-granularity 'gt' --model 'e2e' --weights *.pt

then i run the best model using this:
%run 'main.py' -addG -addT -addE -addV --gpu 0 --test --weights e2e-20230213-0530.pt

then i got the Error:
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1497, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for E2E:
Unexpected key(s) in state_dict: "projector.modalities.3.0.weight", "projector.modalities.3.0.bias", "projector.modalities.3.1.weight", "projector.modalities.3.1.bias".
size mismatch for projector.modalities.1.0.weight: copying a param with shape torch.Size([300, 4]) from checkpoint, the shape in current model is torch.Size([300, 300]).
size mismatch for projector.modalities.2.0.weight: copying a param with shape torch.Size([300, 300]) from checkpoint, the shape in current model is torch.Size([300, 1448]).
size mismatch for message_passing.linear.weight: copying a param with shape torch.Size([1200, 2400]) from checkpoint, the shape in current model is torch.Size([900, 1800]).
size mismatch for message_passing.linear.bias: copying a param with shape torch.Size([1200]) from checkpoint, the shape in current model is torch.Size([900]).
size mismatch for message_passing.lynorm.weight: copying a param with shape torch.Size([1200]) from checkpoint, the shape in current model is torch.Size([900]).
size mismatch for message_passing.lynorm.bias: copying a param with shape torch.Size([1200]) from checkpoint, the shape in current model is torch.Size([900]).
size mismatch for edge_pred.W1.weight: copying a param with shape torch.Size([300, 2414]) from checkpoint, the shape in current model is torch.Size([300, 1814]).
size mismatch for node_pred.0.weight: copying a param with shape torch.Size([4, 1200]) from checkpoint, the shape in current model is torch.Size([4, 900]).

[andreagemelli]

Hi @icang1694 ,
at test time it seems like you are not using histogram textual embeddings --add-hist.
Try again and if does not work post here the two models, during training and testing time (so we can get a look at them).
A.

[icang1694]

thankyou @andreagemelli i miss that option