Continual Learning on 15-Task Sequence Benchmark (C-STANCE to QQP)
45.68OP (%)TreeLoRA
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| TreeLoRAFoundation model=meta-llama / Llama-3.2-1B-Instruct2025.06 | 45.68 | 6.41 | 251 | |
| O-LoRAFoundation model=meta-llama / Llama-3.2-1B-Instruct2025.06 | 44.02 | 10.99 | 679 | |
| HiDeLoRAFoundation model=meta-llama / Llama-3.2-1B-Instruct2025.06 | 42.38 | 11.23 | 683 | |
| FIXFoundation model=meta-llama / Llama-3.2-1B-Instruct2025.06 | 41.32 | — | — | |
| DualPromptFoundation model=meta-llama / Llama-3.2-1B-Instruct2025.06 | 41.29 | 15.58 | 411 | |
| L2PFoundation model=meta-llama / Llama-3.2-1B-Instruct2025.06 | 41.05 | 14.92 | 403 | |
| SeqLoRAFoundation model=meta-llama / Llama-3.2-1B-Instruct2025.06 | 40.71 | 15.72 | 721 | |
| GEMFoundation model=meta-llama / Llama-3.2-1B-Instruct2025.06 | 35.48 | 18.33 | 2,235 | |
| OGDFoundation model=meta-llama / Llama-3.2-1B-Instruct2025.06 | 32.52 | 21.32 | 1,921 | |
| EWCFoundation model=meta-llama / Llama-3.2-1B-Instruct2025.06 | 31.46 | 22.22 | 13,058 |