Continual Learning on GSM8K, MedMCQA, and SQuAD Continual Learning sequence v2
61Average PerformanceKL
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| KLModel=Qwen2.52026.04 | 61 | 0.9 | 2.8 | 5 | |
| Moderate-GiModel=Qwen2.52026.04 | 60.9 | 1.8 | 1.8 | 2.7 | |
| BaselineModel=Qwen2.52026.04 | 59.1 | 1.7 | 1.7 | 11.8 | |
| RandomModel=Qwen2.52026.04 | 58.9 | 0.4 | 0.4 | 10.6 | |
| O-LoRAModel=Qwen32026.04 | 58.8 | 12.4 | 12.4 | 15.3 | |
| Moderate-GiModel=Qwen32026.04 | 58.1 | 4.3 | 4.3 | 5.6 | |
| O-LoRAModel=Qwen2.52026.04 | 55.7 | 8.8 | 11.4 | 21.4 | |
| RandomModel=Qwen32026.04 | 55 | 19.8 | 19.8 | 23.2 | |
| KLModel=Qwen32026.04 | 54.8 | 15.5 | 15.5 | 26 | |
| BaselineModel=Qwen32026.04 | 54.2 | 18.5 | 18.5 | 32.5 | |
| O-LoRAModel=LLaMA-3.12026.04 | 47.9 | 6.4 | 4.5 | — | |
| RandomModel=LLaMA-3.12026.04 | 46.8 | 0.8 | 0.8 | 4.6 | |
| BaselineModel=LLaMA-3.12026.04 | 46.5 | 0.2 | 0.2 | 8.4 | |
| KLModel=LLaMA-3.12026.04 | 46.4 | 1.1 | 1.1 | 8.7 | |
| Moderate-GiModel=LLaMA-3.12026.04 | 46.4 | 1.7 | 1.7 | 5.4 |