load_bert_ issue

[mathiasGM]

Hi,

I am trying to use the 'bert_ner_model' and 'bert_tone_model' using the "load_bert_tone_ner" and "load_bert_tone_model" methods.

I get the following errors when loading the models:

NER:

---

RuntimeError: Error(s) in loading state_dict for BertForTokenClassification:
size mismatch for classifier.weight: copying a param with shape torch.Size([9, 768]) from checkpoint, the shape in current model is torch.Size([2, 768]).
size mismatch for classifier.bias: copying a param with shape torch.Size([9]) from checkpoint, the shape in current model is torch.Size([2]).

Tone:

RuntimeError: Error(s) in loading state_dict for BertForSequenceClassification:
size mismatch for classifier.weight: copying a param with shape torch.Size([3, 768]) from checkpoint, the shape in current model is torch.Size([2, 768]).
size mismatch for classifier.bias: copying a param with shape torch.Size([3]) from checkpoint, the shape in current model is torch.Size([2]).

Any ideas on how to fix this?

Final note:
Thanks for your work on Danish NLP! You are really advancing the field with this repo.

[AmaliePauli]

Hi mathiasGM,
What is your environment? The code is tested with danlp==0.0.9, and transformers==2.3.0 - perhaps try to install from source. Let me know if this help, or else please list your environment.
Thanks, we are glad that you like our work.
Amalie

[Lunde89]

Downgrading Transformers to 2.3.0 worked for me. I'd assume you're running into the same issue.

[mathiasGM]

@AmaliePauli, @Lunde89: Downgrading to Transformers == 2.3.0 solved it!
Thank you very much!