Safety Evaluation on Sorry-Bench (Ref%, Acc%)
65.7Ref %ALIGNBEAM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ALIGNBEAMDraft model=Finance-Llama3-8B, Domain=finance, LBM-DROP=true, alpha=0.5, N=6, K=3, seed=422026.06 | 65.7 | 77.3 | |
| ALIGNBEAMDraft model=Qwen2-7B-Base, Domain=same-vocab ctrl, LBM-DROP=true, alpha=0.5, N=6, K=3, seed=422026.06 | 63.4 | 79.8 | |
| ALIGNBEAMDraft model=MedLlama-3-8B, Domain=medical, LBM-DROP=true, alpha=0.5, N=6, K=3, seed=422026.06 | 63.2 | 72.3 | |
| ALIGNBEAMDraft model=Llama-3.1-8B, Domain=general, LBM-DROP=true, alpha=0.5, N=6, K=3, seed=422026.06 | 61.4 | 74.3 | |
| ALIGNBEAMDraft model=Macro average, LBM-DROP=true, alpha=0.5, K=3, seed=422026.06 | 59.7 | 73.9 | |
| ALIGNBEAMDraft model=DeepSeek-Math-7B-Instruct, Domain=math, LBM-DROP=true, alpha=0.5, N=20, K=3, seed=422026.06 | 45 | 65.7 |