Sentiment Classification on IMDB character-level Long Range Arena (LRA) (val)
70.19AccuracyHKT-Small
Evaluation Results
| Method | Links | |
|---|---|---|
| HKT-SmallSequence Length (T)=1,024, Number of Seeds=3, Model Dimension (d)=128, Overhead=1.313x2026.04 | 70.19 | |
| MHA (LRA)Sequence Length (T)=4,096, Overhead=1.00x2026.04 | 65.9 | |
| MHA (retrained)Sequence Length (T)=1,024, Number of Seeds=3, Overhead=1.00x2026.04 | 62.72 |