Continual Learning on Large Number of Tasks Benchmark (test)
78.1Performance (Order 1)PerTaskFT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| PerTaskFTBase model=T5-large, Reference Type=Non-CL Reference2024.10 | 78.1 | 78.1 | 78.1 | 78.1 | |
| MTLBase model=T5-large, Reference Type=Multi-task learning Reference2024.10 | 76.5 | 76.5 | 76.5 | 76.5 | |
| O-LoRABase model=T5-large2024.10 | 72.3 | 64.8 | 71.6 | 69.6 | |
| CLoRABase model=T5-large2024.10 | 70.7 | 65.6 | 68.2 | 68.1 | |
| LFPT5Base model=T5-large2024.10 | 70.4 | 68.2 | 69.1 | 69.2 | |
| IncLoRABase model=T5-large2024.10 | 63.3 | 58.5 | 61.7 | 61.2 | |
| L2PBase model=T5-large2024.10 | 57.5 | 53.8 | 56.9 | 56.1 | |
| ReplayBase model=T5-large2024.10 | 55 | 54.6 | 53.1 | 54.2 | |
| LwFBase model=T5-large2024.10 | 50.1 | 43.1 | 47.4 | 46.9 | |
| EWCBase model=T5-large2024.10 | 45.3 | 44.5 | 45.6 | 45.1 | |
| SeqFTBase model=T5-large2024.10 | 7.4 | 7.4 | 7.5 | 7.4 | |
| SeqLoRABase model=T5-large2024.10 | 2.3 | 0.6 | 1.9 | 1.6 |