turkish-bert icon indicating copy to clipboard operation
turkish-bert copied to clipboard

bert uncased tf checkpoints

Open hazalturkmen opened this issue 4 years ago • 4 comments

Hi! @stefan-it i need to bert-base-32k-uncased tf chekpoints for further pre-training on Cloud TPU. I found cased version from this link

wget https://schweter.eu/cloud/bert-base-turkish-cased/bert-base-turkish-cased-tf.tar.gz

is it possible to get 32k uncased version of Turkish Bert model?

Thanks for all sharing,

hazalturkmen avatar Dec 12 '21 17:12 hazalturkmen

Hi @hazalturkmen ,

in this archive you can find the last 5 checkpoints from the uncased model (I think I've chosen the 2M checkpoint for the final model):

wget wget https://schweter.eu/cloud/bert-base-turkish-uncased/bert-base-turkish-uncased.tar.gz

Hope this helps :)

stefan-it avatar Dec 12 '21 19:12 stefan-it

thanks for sharing model Stefan! this is what i was looking for :)

how long did it take to train model from scratch? do you remember? its okay if you don't :)

hazalturkmen avatar Dec 13 '21 05:12 hazalturkmen

Hey @hazalturkmen ,

sure, here you can the TensorBoard for the complete training:

8.2, 21:40 to 15.2, 08:25. So the training took ~6 days and 13,5 hours for 2M steps on a v3-8 TPU :hugs:

stefan-it avatar Dec 13 '21 09:12 stefan-it

Hi @stefan-it ,

Finally, can I learn the tensorboard configuration for google cloud tpu? i have this error and i am using firefox browser

Couldn't connect to a server on port 8080

Thank you so much :)

hazalturkmen avatar Dec 13 '21 11:12 hazalturkmen

Hi @stefan-it , I want to dowload Turkish BERT uncased tf checkpoints from previous mention codes: wget wget https://schweter.eu/cloud/bert-base-turkish-uncased/bert-base-turkish-uncased.tar.gz

but I get an error in downloading. I would be very grateful if you help me :) error:

ERROR: cannot verify schweter.eu's certificate, issued by ‘CN=R3,O=Let's Encrypt,C=US’: Issued certificate has expired. To connect to schweter.eu insecurely, use --no-check-certificate'.`

hazalturkmen avatar Feb 20 '24 11:02 hazalturkmen

@hazalturkmen maybe from here: https://huggingface.co/dbmdz/bert-base-turkish-uncased

julien-c avatar Feb 20 '24 11:02 julien-c

Hi @hazalturkmen ,

I've finally uploaded the original checkpoints to Model Hub.

For the uncased model, they are are prefixed with model.ckpt-*, and can be found here.

I hope this helps :)

stefan-it avatar Feb 27 '24 13:02 stefan-it

Thanks! @stefan-it

hazalturkmen avatar Feb 28 '24 07:02 hazalturkmen