Implementation of FusionNet for SQuAD There seems to be an implementation by the original authors of the paper at FusionNet-NLI. You might want to check that out
- Basic Model
- Top question word tuning
- Dropout
- Embedding dropout
- LSTM dropouts
- Attention dropouts
- Linear transform dropouts
- Masking
- fusion masking
- rnn masking
- Loss as per paper
- Disk caching
- Optimizer
- Data Preparation with Spacy
- Tensorboard
- Single model
- Ensemble
- Demo
| Tag | F1 | EM |
|---|
- Dropout details are provided at openreview
- Attention dimensions are confirmed to be 250 for all fusions