Fine-tuning for knowledge acquisition and abstention preservation on RWD
1FT ScoreFull-FT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Full-FTBase Model=Llama3-8B-Instruct2025.06 | 1 | 0 | 0 | |
| LoRABase Model=Llama3-8B-Instruct2025.06 | 1 | 0 | 0 | |
| R-tuningBase Model=Llama3-8B-Instruct2025.06 | 1 | 0 | 0 | |
| Exp. ReplayBase Model=Llama3-8B-Instruct2025.06 | 1 | 0.691 | 0.654 | |
| SEATBase Model=Llama3-8B-Instruct2025.06 | 1 | 0.977 | 0.977 | |
| Full-FTBase Model=Qwen2.5-7B-Instruct2025.06 | 1 | 0 | 0 | |
| LoRABase Model=Qwen2.5-7B-Instruct2025.06 | 1 | 0.031 | 0.058 | |
| EWCBase Model=Qwen2.5-7B-Instruct2025.06 | 1 | 0.042 | 0.037 | |
| CLoRABase Model=Qwen2.5-7B-Instruct2025.06 | 1 | 0.089 | 0.267 | |
| R-tuningBase Model=Qwen2.5-7B-Instruct2025.06 | 1 | 0.005 | 0.068 | |
| SEATBase Model=Qwen2.5-7B-Instruct2025.06 | 1 | 0.909 | 1 | |
| Exp. ReplayBase Model=Qwen2.5-7B-Instruct2025.06 | 0.997 | 0.764 | 0.78 | |
| EWCBase Model=Llama3-8B-Instruct2025.06 | 0.995 | 0.01 | 0 | |
| CLoRABase Model=Llama3-8B-Instruct2025.06 | 0.989 | 0 | 0 |