RuntimeError: Error(s) in loading state_dict for BertForSequenceClassification

[Lunde89]

Full error message:
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]).

Trying to run the following code:
from danlp.models import load_bert_tone_model
classifier = load_bert_tone_model()

Using:
Torch 1.6+cpu
Python 3.7.4
daNLP 0.0.4

[AmaliePauli]

Hi Lunde89,
The DaNLP version 0.0.4 do not support the BertTone model, please update the package to version 0.0.9 .
I hope this will fix your issue.
Amalie

[Lunde89]

Hi Amalie,

I might have mis-typed in the original comment. Sorry about that. At least, i'm using 0.0.9 now - and it produces the same error.

[AmaliePauli]

Alright, what about the transformes version? You also need this package for it to run. It is tested with transformers==2.3.0. ?

[Lunde89]

Hi Amalie,

I figured out the issue The problem is that the version you get from PyPi isn't the same as the one on this repository. After i installed from source, everything works like a charm! 👍

[KennethEnevoldsen]

Just to +1 this comment I have had the exact same problem. Seems like an update of the PyPi is needed.

A simple fix:

pip install tqdm==4.42.1
pip install transformers==2.3.0
pip install danlp==0.0.9

Notice that: flair 0.6.1 requires transformers>=3.0.0, but you'll have transformers 2.3.0 which is incompatible.

(update)
It does however seem that the PyPi isn't up to date with the repo as the following give me this error:

from danlp.models import load_bert_tone_model
classifier = load_bert_tone_model()
classifier.predict('Jeg tror alligvel, det bliver godt')  # works
classifier.predict_proba('Analysen viser, at økonomien bliver forfærdelig dårlig')

AttributeError: 'BertTone' object has no attribute 'predict_proba'

Looking into the class using dir(classifier), we get:

['__class__',
 '__delattr__',
 '__dict__',
 '__dir__',
 '__doc__',
 '__eq__',
 '__format__',
 '__ge__',
 '__getattribute__',
 '__gt__',
 '__hash__',
 '__init__',
 '__init_subclass__',
 '__le__',
 '__lt__',
 '__module__',
 '__ne__',
 '__new__',
 '__reduce__',
 '__reduce_ex__',
 '__repr__',
 '__setattr__',
 '__sizeof__',
 '__str__',
 '__subclasshook__',
 '__weakref__',
 'model_pol',
 'model_sub',
 'predict',
 'tokenizer_pol',
 'tokenizer_sub']

notably not way to extract probability.

[kirajen]

Hi all,

I am now facing the same issue. I have ensured all my versions are as in the comment above.

[KennethEnevoldsen]

So @kirajen please try updating to the latest version. You can do this using:

pip install git+https://github.com/alexandrainst/danlp.git