-
Retrieve questions from the test set via
retrieve_questions.py
script (qald_10.json
should be in the same directory as the script). -
Translate questions from other languages to English via
t5_translation.ipynb
. -
Run
KEQA_inference.ipynb
notebook to generate predictions using KEQA framework. Rename the file with predictions in the following manner:en_prediction.txt
for EN questions,ru_questions.txt
for RU questions,de_questions.txt
for DE questions andze_questions.txt
for ZH questions. -
Run the following shell pipe:
tail -n +N_START questions.txt | head -n CHUNK_SIZE | python3 -m deeppavlov interact kbqa_сq >> answers.txt
,
where N_START
is the number of the first line (question) to process and CHUNK_SIZE
is the number of questions to process.
- Finally, generate submissions for all languages via
generate_submission.ipynb
.