CS229 homework in ACM class.
I used python=3.7, torch=1.10.0, cudatoolkit=11.1.
Overall, batch_size=32, initial learening_rate=2e-5.
For CoSENT, I set
You can modify the train/test data in "train.tsv"/"test.tsv". Results will be saved under "submission.csv"/"CoSent_submission.csv"/"xlnet_submission.csv".
Run python main.py (default model is RoBERTa).
Run python xlnet_main.py to use XLNet.
Run python CoSent_main.py (default model is RoBERTa + CoSENT).
| Method | BERT (w/o data augmentation) | BERT | RoBERTa | CoSENT |
|---|---|---|---|---|
| Performance | 73.451% | 80.088% | 88.053% | 76.991% |