Jailbreak on Jailbreak
65.5MMLUOriginal
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| OriginalQuantization=FP32, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 65.5 | 58.69 | 77 | |
| OriginalQuantization=BF16, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 65.48 | 58.68 | 77 | |
| ELQQuantization=FP32, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 65.25 | 49.23 | 173 | |
| ELQQuantization=BF16, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 65.16 | 49.71 | 173 | |
| OriginalQuantization=INT8, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 64.76 | 56.42 | 96 | |
| ACLQuantization=INT8, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 64.71 | 37.72 | 9,558 | |
| OriginalQuantization=NF4, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 64.34 | 56.83 | 58 | |
| ELQQuantization=INT8, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 64.18 | 41.1 | 9,115 | |
| Q-MisalignQuantization=INT8, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 64.12 | 41.23 | 9,346 | |
| ACLQuantization=NF4, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 63.95 | 36.77 | 9,462 | |
| Q-MisalignQuantization=NF4, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 63.62 | 40.44 | 9,077 | |
| ACLQuantization=BF16, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 63.58 | 53.28 | 96 | |
| ACLQuantization=FP32, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 63.56 | 51.58 | 58 | |
| ELQQuantization=NF4, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 63.55 | 40.42 | 8,846 | |
| Q-MisalignQuantization=FP32, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 62.47 | 60.33 | 19 | |
| Q-MisalignQuantization=BF16, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 60.21 | 59.8 | 77 | |
| OriginalQuantization=FP4, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 59.36 | 55.72 | 173 | |
| ACLQuantization=FP4, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 58.34 | 36.49 | 9,135 | |
| ELQQuantization=FP4, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 57.6 | 41.53 | 8,692 | |
| Q-MisalignQuantization=FP4, Model=Qwen2.5-3B-Instruct, Zero-shot=true2026.01 | 57.58 | 41.51 | 8,231 |