Continual Learning on TRACE Llama2-7B-chat
55.2Average AccuracyOGP
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| OGPType=OptSt, LoRA rank=16, seeds=32026.04 | 55.2 | 10.8 | 8.7 | 2.3 | 5.4 | |
| Replay 5%Type=ER, Budget=5%, LoRA rank=16, seeds=32026.04 | 54.8 | 9.4 | 8.6 | 2.4 | 5.1 | |
| GORPType=Proj, LoRA rank=16, seeds=32026.04 | 53.9 | 12.7 | 10.1 | 2.1 | 6.8 | |
| Replay 1%Type=ER, Budget=1%, LoRA rank=16, seeds=32026.04 | 52.6 | 11.2 | 11.2 | 2.3 | 6.4 | |
| InfLoRAType=PEFT, LoRA rank=16, seeds=32026.04 | 51.8 | 12.9 | 11.2 | 2 | 7.4 | |
| Adam-NSCLType=Proj, LoRA rank=16, seeds=32026.04 | 51.4 | 13.2 | 11.8 | 1.9 | 7.8 | |
| MoFOType=OptSt, LoRA rank=16, seeds=32026.04 | 49.9 | 15.8 | 13.9 | 1.5 | 9.7 | |
| EWCType=Reg, LoRA rank=16, seeds=32026.04 | 49.2 | 14.6 | 14.6 | 1.8 | 9.1 | |
| Seq FT (LoRA)Type=—, LoRA rank=16, seeds=32026.04 | 46.4 | 21.8 | 18.3 | 1.4 | 12.4 |