Closed gevezex closed 4 years ago
Hi @gevezex ,
training of the models (both cased and uncased) is currently in progress.
The cased model is now at 1.4M steps, uncased model at 0.5M steps.
I updated the README with links to the TensorBoards.
After evaluation (I plan to evaluate the models on PoS and NER datasets from here), I'll upload the models to the Huggin Face model hub.
I think training and evaluation of the cased model will be finished until Friday, so stay tuned!
Great! Bert Multilingual performed very bad on that dataset, so I am very curious about your trained model.
Hi @stefan-it ,
Thank you for the repo. Did you finish the training?
Hey guys,
the model is now uploaded to the Hugging Face model hub:
https://huggingface.co/dbmdz/bert-base-turkish-cased
You can directly use it in the Transformers library :)
Uncased model is coming next week (training has finished, only evaluation is missing).
Hi @stefan-it . Any updates about the uncased model?
Hi @serdar-eric ,
uncased model will be released this week. Additionally, I also trained cased and uncased models with a larger vocab size (128k instead of 32k), I'm currently doing the evaluation, but I think it will be ready until Sunday :)
Hi @serdar-eric ,
uncased model will be released this week. Additionally, I also trained cased and uncased models with a larger vocab size (128k instead of 32k), I'm currently doing the evaluation, but I think it will be ready until Sunday :)
It's great to hear that @stefan-it . Thank you very much
Hi @serdar-eric ,
the uncased model is now available from model hub: https://huggingface.co/dbmdz/bert-base-turkish-uncased.
Additionally, you can now find models with a larger vocab size (128k) on the model hub:
Thank you very much @stefan-it . Already started to download and work on it :)
Stefan, I need 128K TF model for comparing with 32K. Could u share? Thanx a lot! :)
@stefan-it Stefan, reminder. I've short time to submit the thesis. BTW, your Github repo link is on my paper references... :) Thanx a lot.
@balki7 no problem, I'm currently preparing the archives :)
Download links are now available:
# sha256: 85977347ef031c5cf5f41bd79342e1548b119a1c5b1e7c5dcc1cae4855c4d2b1
wget https://schweter.eu/cloud/bert-base-turkish-128k-cased/bert-base-turkish-128k-cased-tf.tar.gz
# sha256: b73916ee54a25b1a75df87f55dea4adbc8b61b6f82ef0983e556b1007e3be3df
wget https://schweter.eu/cloud/bert-base-turkish-128k-uncased/bert-base-turkish-128k-uncased-tf.tar.gz
Please let me know if the checkpoints are working + good luck with your thesis :)
Çok teşekkür ederim Stefan... :)
So where can we download the TR trained model?