Continual Learning on G1
11.89APLoRAM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LoRAMrank=642026.05 | 11.89 | 10 | 1.89 | |
| LoRA-GArank=642026.05 | 11.75 | 10.91 | 0.84 | |
| Vanilla LoRArank=642026.05 | 10.29 | 11.06 | -0.77 | |
| LoRA-GARank=1282026.05 | 10.08 | 9.09 | 0.99 | |
| Vanilla LoRARank=1282026.05 | 9.78 | 9.05 | 0.73 | |
| LoRAMRank=1282026.05 | 9.68 | 8.88 | 0.79 | |
| Slicerank=64, c=1.00, baseline=Vanilla LoRA2026.05 | 1.27 | -1 | 2.27 | |
| SliceRank=128, Base Initializer=LoRAM, c=1.002026.05 | 0.56 | 0.02 | 0.54 | |
| SliceRank=128, Base Initializer=Vanilla LoRA, c=1.002026.05 | 0.46 | -0.15 | 0.6 | |
| SliceRank=128, Base Initializer=LoRAM, c=0.502026.05 | 0.33 | 0.38 | -0.05 | |
| SliceRank=128, Base Initializer=LoRAM, c=0.752026.05 | 0.23 | 0.17 | 0.05 | |
| SliceRank=128, Base Initializer=Vanilla LoRA, c=0.502026.05 | 0.22 | 0.21 | 0 | |
| SliceRank=128, Base Initializer=LoRA-GA, c=1.002026.05 | 0.16 | -0.18 | 0.34 | |
| SliceRank=128, Base Initializer=Vanilla LoRA, c=0.752026.05 | 0.12 | 0 | 0.11 | |
| SliceRank=128, Base Initializer=LoRA-GA, c=0.502026.05 | -0.08 | 0.18 | -0.26 | |
| SliceRank=128, Base Initializer=LoRA-GA, c=0.752026.05 | -0.18 | -0.03 | -0.15 | |
| Slicerank=64, c=1.00, baseline=LoRA-GA2026.05 | -0.19 | -0.85 | 0.67 | |
| Slicerank=64, c=1.00, baseline=LoRAM2026.05 | -0.32 | 0.07 | -0.39 | |
| Slicerank=64, c=0.50, baseline=Vanilla LoRA2026.05 | -0.75 | -2.03 | 1.28 | |
| Slicerank=64, c=0.75, baseline=Vanilla LoRA2026.05 | -0.89 | -1.99 | 1.1 | |
| Slicerank=64, c=0.50, baseline=LoRA-GA2026.05 | -2.21 | -1.88 | -0.33 | |
| Slicerank=64, c=0.50, baseline=LoRAM2026.05 | -2.35 | -0.96 | -1.38 | |
| Slicerank=64, c=0.75, baseline=LoRA-GA2026.05 | -2.35 | -1.84 | -0.51 | |
| Slicerank=64, c=0.75, baseline=LoRAM2026.05 | -2.48 | -0.92 | -1.56 |