how to generate this file bert_arb_conll12_features.hdf5 ?

[aymen-souid]

No description provided.

[juntaoy]

You will need to use the extract_bert_features.sh, and for doing this you will need provide the arabert 0.1 in tensorflow format

[aymen-souid]

i think that I've run it successfully , after that I used python evaluate.py arabic_cleaned_bert, and I got this error,
do you have any idea about this issue ?

[juntaoy]

So for the checkpoint_path you need to remove ".index"

[aymen-souid]

it doesn't seem to be resolved even after removing ".index"

[juntaoy]

There is no error for your new run, it seems you terminated the system manually

[juntaoy]

It might be a problem with your predict.py, as my code do not print anything liked killed:)

[aymen-souid]

the "killed" issue was resolved when i restart my PC
but you can have a look at this one ,I can't understand the mismatch between the graph or the problem between the tensors ?

[juntaoy]

This is the character embedding, if you use my pre-trained model you need to use the same char_vocab.arabic.txt provided

[aymen-souid]

i have checked that, but still exactly same error,
is it possible that because you've used another version of arabert ?
I'm using the version uploaded in hugging face "bert-base-arabertv01"

[juntaoy]

From what I can tell this should be a problem for the character embedding:)

[aymen-souid]

I made some modifications to the paths in experiment.conf

and I got this error ,any idea how to solve it ?

[juntaoy]

I haven't seen this type of error before.

[aymen-souid]

this error occured when i modified the char_vocab_file in model hyperparameter (in experment.conf), do I need to put the path of the vocab.txt of the arabert or the path of your pretrained model? (this error occurred when I've put the the path of your model) ?