Helpfulness Evaluation on Alpaca Eval
90Alpaca Eval (%)No Defense
Evaluation Results
| Method | Links | |
|---|---|---|
| No DefenseBase LLM=ChatGPT2025.08 | 90 | |
| Self-ReminderBase LLM=ChatGPT2025.08 | 90 | |
| Self-ExaminationBase LLM=ChatGPT2025.08 | 90 | |
| ICDBase LLM=ChatGPT2025.08 | 88 | |
| Context FilteringBase LLM=ChatGPT2025.08 | 88 | |
| No DefenseBase LLM=Llama22025.08 | 62 | |
| Context FilteringBase LLM=Llama22025.08 | 60 | |
| No DefenseBase LLM=Vicuna2025.08 | 59 | |
| Context FilteringBase LLM=Vicuna2025.08 | 57 | |
| Self-ReminderBase LLM=Vicuna2025.08 | 56 | |
| Self-ExaminationBase LLM=Vicuna2025.08 | 56 | |
| Self-ReminderBase LLM=Llama22025.08 | 55 | |
| Safe DecodingBase LLM=Llama22025.08 | 52 | |
| ICDBase LLM=Vicuna2025.08 | 51 | |
| Safe DecodingBase LLM=Vicuna2025.08 | 50 | |
| Intention AnalysisBase LLM=Vicuna2025.08 | 33 | |
| ICDBase LLM=Llama22025.08 | 21 | |
| AMA ReweightingStrategy=Reweighting2025.05 | 17.77 | |
| UltraFeedback Specialist2025.05 | 17.59 | |
| AMA Resamplingoptimization_strategy=resampling2025.05 | 16.88 | |
| MOPO2025.12 | 16.33 | |
| AMA Reweightingoptimization_strategy=reweighting2025.05 | 16.21 | |
| Chatbot Arena 2024 Specialist2025.05 | 16.05 | |
| Standard Uniform2025.05 | 15.51 | |
| PARM2025.12 | 14.71 | |
| RiC2025.12 | 13.15 | |
| AMA ResamplingStrategy=Resampling2025.05 | 13.04 | |
| Model Averaging2025.05 | 12.28 | |
| Model Averaging2025.05 | 12.21 | |
| Standard UniformSampling=Uniform2025.05 | 11.78 | |
| Standard2025.05 | 11.32 | |
| DPO on D_J2025.12 | 10.99 | |
| CodeUltraFeedback Specialist2025.05 | 9.93 | |
| Zephyr-7b-sft-full2025.05 | 8.64 | |
| Zephyr-7b-sft-full2025.05 | 8.64 | |
| MODPO2025.12 | 7.34 | |
| Standard2025.05 | 6.9 | |
| SafeRLHF Specialist2025.05 | 6.66 | |
| SafeRLHF Specialist2025.05 | 6.66 | |
| Self-ExaminationBase LLM=Llama22025.08 | 5 | |
| Intention AnalysisBase LLM=ChatGPT2025.08 | 4 | |
| Intention AnalysisBase LLM=Llama22025.08 | 1 |