BERT fine-tuning for QA

all information about bert you can find here

What to do

Run jupyter notebook on colab to fine-tune BERT for QA on russian dataset or to test your own examples

Files description

dev.json, train.json, dev-mini.json & train-mini.json

That's the train and test (80% train, 20% test or dev) parts of sberbank's russian QA dataset.

To simply train model on your dataset use structure, similar to train.json

Train + dev = 50333 exapmles

Train-mini + dev-mini = 12000 examples

model.ckpt-*

This is fine-tuned BERT model for QA