Continual Learning on SuperNI Standard CL Benchmark (test)
81.9Average PerformanceSPAT-LoRA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SPAT-LoRABackbone=Llama-2-13B-chat2025.12 | 81.9 | 83.2 | 82.4 | 80.1 | |
| Multi-TaskBackbone=Llama-2-13B-chat2025.12 | 81.4 | — | — | — | |
| SLAOBackbone=Llama-2-13B-chat2025.12 | 81 | 80.8 | 81.1 | 81.1 | |
| MagMaxBackbone=Llama-2-13B-chat2025.12 | 80.7 | 80.9 | 80.6 | 80.7 | |
| CorDABackbone=Llama-2-13B-chat2025.12 | 80 | 79.1 | 80.4 | 80.6 | |
| LoRM-BABackbone=Llama-2-13B-chat2025.12 | 79.7 | 78.4 | 80.2 | 80.4 | |
| IncLoRABackbone=Llama-2-13B-chat2025.12 | 79.1 | 78.3 | 79.6 | 79.5 | |
| SeqLoRABackbone=Llama-2-13B-chat2025.12 | 77.7 | 77.1 | 77.4 | 78.6 | |
| O-LoRABackbone=Llama-2-13B-chat2025.12 | 77.2 | 76.1 | 77.1 | 78.5 | |
| LoRM-ABBackbone=Llama-2-13B-chat2025.12 | 75.2 | 76.2 | 74.1 | 75.3 | |
| KnOTS(zero init)Backbone=Llama-2-13B-chat2025.12 | 73 | 71.9 | 73.1 | 73.9 | |
| LoRA-LEGOBackbone=Llama-2-13B-chat2025.12 | 72.7 | 73 | 72.3 | 72.9 | |
| OPCMBackbone=Llama-2-13B-chat2025.12 | 64.7 | 68.8 | 63.6 | 61.6 |