Unexpected key(s) in state_dict: "transformer.text_encoder.embeddings.position_ids".

[vdorbala]

When running the demo script, I get -
Unexpected key(s) in state_dict: "transformer.text_encoder.embeddings.position_ids".

while trying to load the model using -
model, postprocessor = torch.hub.load('ashkamath/mdetr:main', 'mdetr_efficientnetB5', pretrained=True, return_postprocessor=True)

Does the model not work anymore?

[iremeyiokur]

It might be related with the transformers version. I solved the problem by adding strict=False into load_state_dict as proposed in the following link which basically ignores non-matching keys. I hope it will work for you.

https://discuss.pytorch.org/t/missing-keys-unexpected-keys-in-state-dict-when-loading-self-trained-model/22379/6

[Buchimi]

How can this be done in the colab?

[SKALGO22]

It might be related with the transformers version. I solved the problem by adding strict=False into load_state_dict as proposed in the following link which basically ignores non-matching keys. I hope it will work for you.

https://discuss.pytorch.org/t/missing-keys-unexpected-keys-in-state-dict-when-loading-self-trained-model/22379/6

@iremeyiokur , your proposed fix, loads the model without errors, but then I get no logits while making inferences...it's just a NaN tensor. I also tried deleting that particular key from the sate_dict yet the issue persists. Was it the same for you?