Is it possible to replicate the results on XNLI dataset which are present on "Multilingual README" page ?

google-research / bert

TensorFlow code and pre-trained models for BERT

https://arxiv.org/abs/1810.04805

Apache License 2.0

37.7k stars 9.55k forks source link

Is it possible to replicate the results on XNLI dataset which are present on "Multilingual README" page ? #491

Open gourango01 opened 5 years ago

gourango01 commented 5 years ago

I am not able to replicate the results for "BERT - Translate Train Cased" system on English. Can anybody know the set hyperparameters which were used for fine-tuning of BERT-Base, Multilingual Cased (New, recommended) on English ?

hlums commented 5 years ago

Actually I was able to replicate the results on Arabic in the multilingual readme file. Just used the parameters they provided for fine-tuning Chinese model and their run_classifier.py. I suppose for English it would be the same?