Question Answering on TriviaQA (factual accuracy)
49.2Factual AccuracyARREST-HB
Evaluation results
| Method | Links | |
|---|---|---|
| ARREST-HBLLM=LLaMa-3.1 8B2026.01 | 49.2 | |
| ARREST-HBLLM=LLaMa-2 7B2026.01 | 46.74 | |
| ITILLM=LLaMa-2 7B2026.01 | 46.18 | |
| Activation DecodingLLM=LLaMa-3.1 8B2026.01 | 44.17 | |
| Activation DecodingLLM=LLaMa-2 7B2026.01 | 42.57 | |
| DOLALLM=LLaMa-2 7B2026.01 | 42.3 | |
| ARREST-HBLLM=Vicuna 7B2026.01 | 35.59 | |
| Activation DecodingLLM=Vicuna 7B2026.01 | 29.33 | |
| Self-ReflectionLLM=Vicuna 7B2026.01 | 24.63 | |
| DOLALLM=Vicuna 7B2026.01 | 24.13 | |
| COVELLM=Vicuna 7B2026.01 | 16.02 | |
| ITILLM=Vicuna 7B2026.01 | 15.68 | |
| Base ModelLLM=Vicuna 7B2026.01 | 15.02 | |
| ITILLM=LLaMa-3.1 8B2026.01 | 14.68 | |
| DOLALLM=LLaMa-3.1 8B2026.01 | 14.3 | |
| COVELLM=LLaMa-2 7B2026.01 | 14.29 | |
| COVELLM=LLaMa-3.1 8B2026.01 | 14.29 | |
| Self-ReflectionLLM=LLaMa-3.1 8B2026.01 | 13.78 | |
| Self-ReflectionLLM=LLaMa-2 7B2026.01 | 13.36 | |
| Base ModelLLM=LLaMa-2 7B2026.01 | 12.55 | |
| Base ModelLLM=LLaMa-3.1 8B2026.01 | 10.07 |