Arguments: (AttributeError("'TransformerWordEmbeddings' object has no attribute 'tokenize'"),)

[Student-xyf]

Hi urchade,

Thanks a lot for this great work. 😃
I'm trying to reproduce your results on CONLL04. But I found that the code reported an error:
Arguments: (AttributeError("'TransformerWordEmbeddings' object has no attribute 'tokenize'"),)
Through debugging, it was found that the flair library was referenced:
from flair.embeddings import TransformerWordEmbeddings
Then pass this code:
self.bert_layer = TransformerWordEmbeddings(model_name, fine_tune=fine_tune, subtoken_pooling=subtoken_pooling)
Among them model_name="bert-base-cased", fine_tune=True, subtoken_pooling="first"

Execute to the next code:
embedder = self.bert_layer
tokenized_sentences, all_token_subtoken_lengths, subtoken_lengths = embedder._gather_tokenized_strings(sentences)

The reason for the error is that the _gather_tokenized_strings method under the embedder cannot be found. I would like to ask what the reason is. It is because my flair version is wrong.

My flair version is: flair==0.13.1

Looking forward to your reply, thank you

Sincerely

[urchade]

I am not with my computer, but by memory ut should be 0.11.x

[urchade]

Do not hesitate if you have additional help

Did not have much time to polish the code 🙏

[Student-xyf]

Thank you for your reply. I used the flair=0.11.0 version and the same error still occurred.

[urchade]

[Student-xyf]

thank you

[Student-xyf]

Hi Urchade,

Thanks a lot for this great work. 😃
I would also like to ask what graphics card (how much memory) is your model trained on?

Sincerely,

Thanks

[urchade]

I have used A100 80G or V100 32 G, depending on available in our cluster

[edzq]

@xyf-495800804 Did you successfully run this code for reproduction? Can I add your WeChat to have a further discussion?

[YaxinCountry]

@edzq Did you successfully run this code for reproduction? Could I add your WeChat to have a further discussion?

[edzq]

@edzq Did you successfully run this code for reproduction? Could I add your WeChat to have a further discussion?

Yes. My id is Maxwell0115

But I am not interested this work too much. HGERE is the current state-of-the-art of this task. But this generative style model may be still interesting.