Benchmarking various Deep Learning models such as BERT, ALBERT, BiLSTMs on the task of sentence entailment using two datasets - MultiNLI and SNLI.
These correspond to the model benchmarks on 100k training samples of the MultiNLI dataset.
| Model | Validation accuracy | Validation loss | Training accuracy | Training loss |
|---|---|---|---|---|
| BiLSTM | 0.6104 | 0.8718 | 0.6014 | 0.8754 |
| BERT | 0.7906 | 0.8490 | 0.9690 | 0.0916 |
| ALBERT | 0.7110 | 0.7003 | 0.7405 | 0.6356 |