Continual Learning on NumGLUE-cm
36.25Average AccuracySETA
Evaluation results
| Method | Links | |
|---|---|---|
| SETABackbone=LLaMA-2 7B, Batch size=16, Learning rate=1e-42026.01 | 36.25 | |
| I-LoRABackbone=LLaMA-2 7B, Batch size=16, Learning rate=1e-42026.01 | 33.7 | |
| EWCBackbone=LLaMA-2 7B, Batch size=16, Learning rate=1e-42026.01 | 25.5 | |
| PPBackbone=LLaMA-2 7B, Batch size=16, Learning rate=1e-42026.01 | 25.3 | |
| L2PBackbone=LLaMA-2 7B, Batch size=16, Learning rate=1e-42026.01 | 22.7 | |
| Seq-TrainBackbone=LLaMA-2 7B, Batch size=16, Learning rate=1e-42026.01 | 21.9 | |
| A-GEMBackbone=LLaMA-2 7B, Batch size=16, Learning rate=1e-42026.01 | 19.8 | |
| ERBackbone=LLaMA-2 7B, Batch size=16, Learning rate=1e-42026.01 | 18.7 | |
| GEMBackbone=LLaMA-2 7B, Batch size=16, Learning rate=1e-42026.01 | 12.7 |