Re-implementation of Bidirectional Attention Flow for Machine Comprehension (ICLR 2017)
- python=3.7.0
- pytorch=1.1.0
- torchtext
- tensorboardX
- tqdm
- nltk
- Squad1.1 where every question has its answer.
MODEL_NAME="model"
python main.py --model_name ${MODEL_NAME}\
--GPU 0\
--train_batch_size 60\
--train_file train-v1.1.json\
--dev_batch_size 100\
--dev_file dev-v1.1.json
- Squad 2.0 where some questions don't have their answers within the passages.
MODEL_NAME="model"
python main.py --model_name ${MODEL_NAME}\
--GPU 0\
--train_batch_size 60\
--train_file train-v2.0.json\
--dev_batch_size 100\
--dev_file dev-v2.0.json\
--no_answer True
| checkout | dataset | EM | F1 |
|---|---|---|---|
| edca354e8b74d895c6f5a363f31aefab267dbf89 | SQuAD1.1 | 65.13 | 75.53 |
| edca354e8b74d895c6f5a363f31aefab267dbf89 | SQuAD2.0 (no_answer False) | 32.87 | 38.31 |
| SQuAD2.0 | ? | ? |