Long-document Classification on IMDB (test)
0.8861AccuracyVanilla LSTM
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Vanilla LSTMParams (M)=27.83, Model Size (MB)=106.18, Avg Epoch Time (s)=57.12, Total Train Time (s)=285.62, Peak GPU (MB)=875.15, Best Epoch=2, Batch Size=32, MaxLen=5122025.12 | 0.8861 | 0.8861 | 0.8861 | 0.8861 | 0.952 | |
| QL-LSTMParams (M)=15.47, Model Size (MB)=59.03, Avg Epoch Time (s)=592.63, Total Train Time (s)=2963.17, Peak GPU (MB)=387.41, Best Epoch=1, Batch Size=16, MaxLen=5122025.12 | 0.8783 | 0.8791 | 0.8783 | 0.8782 | 0.9483 | |
| GRUParams (M)=27.31, Model Size (MB)=104.17, Avg Epoch Time (s)=46.13, Total Train Time (s)=230.67, Peak GPU (MB)=897.51, Best Epoch=2, Batch Size=32, MaxLen=5122025.12 | 0.8777 | 0.8779 | 0.8777 | 0.8777 | 0.9464 | |
| BiLSTMParams (M)=29.94, Model Size (MB)=113.64, Avg Epoch Time (s)=84.32, Total Train Time (s)=421.60, Peak GPU (MB)=1,004, Best Epoch=3, Batch Size=16, MaxLen=5122025.12 | 0.8559 | 0.8559 | 0.8559 | 0.8559 | 0.9319 |