• This repo contains code for the paper Mandar Joshi, Eunsol Choi, Daniel Weld, Luke Zettlemoyer.

TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension In Association for Computational Linguistics (ACL) 2017, Vancouver, Canada.

• The data can be downloaded from the TriviaQA website. The Apache 2.0 License applies to both the code and the data.
• Please contact Mandar Joshi (<first-name>[email protected]) for suggestions and comments.

• Python 3. You should be able to run the evaluation scripts using Python 2.7 if you take care of unicode in utils.utils.py.
• BiDAF requires Python 3 -- check the original repository for more details.

• tensorflow (only if you want to run BiDAF, verified on r0.11)
• nltk
• tqdm

The dataset file parameter refers to files in the qa directory of the data (e.g., wikipedia-dev.json). For file format, check out the sample directory in the repo.

python3 -m evaluation.triviaqa_evaluation --dataset_file samples/triviaqa_sample.json --prediction_file samples/sample_predictions.json

• If you have a SQuAD model and want to run on TriviaQA, please refer to utils.convert_to_squad_format.py