RunTimeError: Error(s) in loading state_dict for RobertaForSequenceClassification even after transformer==2.9.1

[Marvy101]

I get the following error after trying to run

pip install -r requirements.txt
python -m detector.server detector-base.pt

Error:

RuntimeError: Error(s) in loading state_dict for RobertaForSequenceClassification:
        Missing key(s) in state_dict: "roberta.embeddings.position_ids".
        Unexpected key(s) in state_dict: "roberta.pooler.dense.weight", "roberta.pooler.dense.bias"

I have tried transformers==2.9.1 in requirements.txt, still receiving the same error.

Thanks for your help.

[hugoroussel]

Having the same error too :)

[Marvy101]

Having the same error too :)

Hey, still waiting on a solution and trying different things.

You wanna have a discussion about AI, and the future?

[ps3-app]

Having the same issues,
RuntimeError: Error(s) in loading state_dict for RobertaForSequenceClassification: Missing key(s) in state_dict: "roberta.embeddings.position_ids". Unexpected key(s) in state_dict: "roberta.pooler.dense.weight", "roberta.pooler.dense.bias"

do you have the solution for it yet?

[AmirFone]

I am having the exact same issue.

[AmirFone]

was anyone able to resolve this issue yet?

[hugoroussel]

No luck on my side :( I think it might be related to #36 (running on macs M1). Also tried running it via docker without luck too.

[AryanGold]

I confirm that set transformers==2.9.1 resolve issue. I try it in isolated Docker container and rebuilding container from scratch, for to be sure that all packages reinstalled.

[hugoroussel]

@AryanGold can you give informations about you current setup (OS etc)?

[AryanGold]

My Host: Ubuntu 20.04, Docker version 20.10.22

I think important to run this repository in isolated Docker container or Python Virtual Environment. Also if you change some version in requirements.txt then need reinstall all pip packages without any cache etc.

I also attache sample of my Docker file.
Docker_sample.zip

[varenc]

I also ran into the same error and using transformers==2.9.1 fixed it. If you're not using a proper environment, like we all should be, then you can probably fix the problem just by doing pip uninstall transformers and then pip install transformers==2.9.1 but WMMV.

[manueltarouca]

Make sure that your python version is 3.7.3 and the following package is transformers==2.9.1.

[SachinGanesh]

replace

self.model.load_state_dict(data['model_state_dict'])

with

self.model.load_state_dict(data['model_state_dict'], strict=False)