Language Modeling and Translation on General Language Preservation Task
41.4chrFLoRA-r16
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LoRA-r16Rank=16, Trainable parameters=~17M, Number of seeds=5, Backbone model=Qwen3-1.7B2026.06 | 41.4 | 4.54 | |
| LoRA-r16 + replay 1%Rank=16, Replay ratio=1%, Trainable parameters=~17M, Number of seeds=5, Backbone model=Qwen3-1.7B2026.06 | 41.4 | 4.5 | |
| OPLoRA-r16Rank=16, Trainable parameters=~17M, Number of seeds=5, Backbone model=Qwen3-1.7B2026.06 | 41.4 | 4.54 | |
| CorDA-k16-r16Rank=16, k=16, Trainable parameters=~17M, Number of seeds=5, Backbone model=Qwen3-1.7B2026.06 | 41.4 | 4.38 | |
| FProj-LRTrainable parameters=~17M, Number of seeds=5, Backbone model=Qwen3-1.7B2026.06 | 41.3 | 4.38 | |
| Fora-LRTrainable parameters=~17M, Number of seeds=5, Backbone model=Qwen3-1.7B2026.06 | 41.3 | 4.38 | |
| LoRA-r16 + replay 10%Rank=16, Replay ratio=10%, Trainable parameters=~17M, Number of seeds=5, Backbone model=Qwen3-1.7B2026.06 | 40.5 | 4.55 |