Twitter | Pretraživanje | |
Stefan
Turkish-: Anyone interested in a Turkish BERT and wants to evaluate it on downstream tasks? I did evaluation only for UD PoS tagging - any help is really appreciated! Would really like to have a proper evaluation before adding it to the Transformers hub🤗
Reply Retweet Označi sa "sviđa mi se" More
Stefan 2. velj
Odgovor korisniku/ci @huggingface
Thanks for your positive reactions 🤗 I'll try to answer all questions now 😅 The repo for Turkish BERT model can be found here:
Reply Retweet Označi sa "sviđa mi se"
Bram Vanroy 1. velj
Odgovor korisniku/ci @_stefan_munich @huggingface
Great work, as usual! Can you share the results of the PoS tagging? I'm very curious about these highly morphological languages and how LM perform on them.
Reply Retweet Označi sa "sviđa mi se"
Stefan 2. velj
Odgovor korisniku/ci @BramVanroy @huggingface
Repo is here, I've already added the PoS tagging results 😄
Reply Retweet Označi sa "sviđa mi se"
ibrahim e. demirci 1. velj
Odgovor korisniku/ci @_stefan_munich @huggingface
What is your training corpus?
Reply Retweet Označi sa "sviđa mi se"
Stefan 2. velj
Odgovor korisniku/ci @iedmrc @huggingface
Currently I'm using the Turkish part of the OSCAR corpus:
Reply Retweet Označi sa "sviđa mi se"
Gözde Gül Sahin 1. velj
Stefan, we had created small linguistic tests (probing tasks) exactly for this purpose. We also have many for Turkish, and a web service for Transformer models:
Reply Retweet Označi sa "sviđa mi se"
Stefan 2. velj
Odgovor korisniku/ci @GzdeGlSahin1 @Thom_Wolf @huggingface
This looks pretty awesome 😍 I just uploaded an archive (with pytorch_model.bin, vocab.txt and config.json) but the interface still shows "uploading" - I'll wait a bit now. Should I try the GitHub repo 🤔
Reply Retweet Označi sa "sviđa mi se"
Gokhan Tur 1. velj
Odgovor korisniku/ci @_stefan_munich @huggingface i 2 ostali
Reply Retweet Označi sa "sviđa mi se"
Kemal Oflazer 1. velj
Odgovor korisniku/ci @tur_gokhan @_stefan_munich i 2 ostali
Stefan I have asked some colleagues to contact you for this. If several billions of words of additional Turkish text would improve your model, let me know and I can dig up what I have.
Reply Retweet Označi sa "sviđa mi se"
Zehra 1. velj
BERTAN 👍
Reply Retweet Označi sa "sviđa mi se"
Evin Pinar Ornek 1. velj
ahah it is a weird name that I checked if it really existed. there are not any normal names in Turkish that has "bert" in it. But I would suggest using Turkish translation of the name "Bert" from the Sesame Street: "Büdü" ☺️ Elmo is "Edi".
Reply Retweet Označi sa "sviđa mi se"