Please see imbd2023 for more information.
Install pipenv
, then run pipenv sync
to install all packages specified in Pipfile.lock
tree -L 2
# ├─data
# │ ├─B_traing1 # pre-training data
# │ │ ├─not reworkable
# │ │ └─reworkable
# │ └─B_traing2 # training data
# │ ├─not reworkable
# │ └─reworkable
# ├─FTP # pretrained model dir
# ├─model # finetuned model dir
# └─src
- check
src/configs.py
- run
create/augmentation_data.py
if augmentation is needed. - for training stage, run
src/create_ref_table.py
to prepare data split, then runsrc/train.py
with specific arguments. - for testing stage, run
src/validate.py
orsrc/predict.py
based on whether labeled answers are available.
The source of the data is provided by the competition's official orgaizers. This project only open-sources the related code.