Factuality Evaluation on TruthfulQA
94.3MC2GER-Steer
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GER-SteerBackbone=Gemma-2-9B-it2026.03 | 94.3 | — | 63.9 | — | 68 | — | |
| LDPBackbone=Gemma-2-9B-it2026.03 | 94.1 | — | 61 | — | 67 | — | |
| ACTBackbone=Gemma-2-9B-it2026.03 | 93.7 | — | 63.4 | — | 67.7 | — | |
| VanillaBackbone=Gemma-2-9B-it2026.03 | 93 | — | 59.5 | — | 64 | — | |
| RePEBackbone=Gemma-2-9B-it2026.03 | 93 | — | 61.4 | — | 66.1 | — | |
| GER-SteerBackbone=Llama-3.1-8B-Instruct2026.03 | 92.8 | — | 47 | — | 50.7 | — | |
| ACTBackbone=Llama-3.1-8B-Instruct2026.03 | 92.4 | — | 46.4 | — | 50.3 | — | |
| NL-ITIBackbone=Gemma-2-9B-it2026.03 | 92.4 | — | 61.8 | — | 66.9 | — | |
| NL-ITIBackbone=Llama-3.1-8B-Instruct2026.03 | 92.3 | — | 48 | — | 52.1 | — | |
| Angular SteeringBackbone=Gemma-2-9B-it2026.03 | 92.1 | — | 62.5 | — | 67.9 | — | |
| VanillaBackbone=Llama-3.1-8B-Instruct2026.03 | 92 | — | 42.3 | — | 46 | — | |
| LDPBackbone=Llama-3.1-8B-Instruct2026.03 | 92 | — | 42.1 | — | 45 | — | |
| Angular SteeringBackbone=Llama-3.1-8B-Instruct2026.03 | 91.8 | — | 47.1 | — | 51.3 | — | |
| RePEBackbone=Llama-3.1-8B-Instruct2026.03 | 90.3 | — | 44.7 | — | 49.5 | — | |
| CAABackbone=Llama-3.1-8B-Instruct2026.03 | 89 | — | 45.4 | — | 51.1 | — | |
| GER-SteerBackbone=Qwen2.5-7B2026.03 | 85 | — | 42.5 | — | 50 | — | |
| Angular SteeringBackbone=Qwen2.5-7B2026.03 | 83.7 | — | 40.8 | — | 48.7 | — | |
| ACTBackbone=Qwen2.5-7B2026.03 | 83.1 | — | 39.6 | — | 47.6 | — | |
| NL-ITIBackbone=Qwen2.5-7B2026.03 | 80.7 | — | 39.7 | — | 49.3 | — | |
| CAABackbone=Gemma-2-9B-it2026.03 | 80 | — | 50.4 | — | 63 | — | |
| CAABackbone=Qwen2.5-7B2026.03 | 79 | — | 34.8 | — | 44 | — | |
| RePEBackbone=Qwen2.5-7B2026.03 | 75 | — | 35.3 | — | 47 | — | |
| LDPBackbone=Qwen2.5-7B2026.03 | 70.1 | — | 21.7 | — | 31.1 | — | |
| SLEDBackbone=Qwen-3-14B-Base2024.11 | 68.99 | 40 | 36.23 | — | — | — | |
| Qwen-3-14B-Base2024.11 | 68.65 | 38.1 | 36.16 | — | — | — | |
| SLEDBackbone=Mixtral-8×7B-IT2024.11 | 68.52 | 41.73 | 37.7 | — | — | — | |
| VanillaBackbone=Qwen2.5-7B2026.03 | 68 | — | 20.4 | — | 30 | — | |
| SLEDBackbone=gpt-oss-20b2024.11 | 67.69 | 36.71 | 35.26 | — | — | — | |
| gpt-oss-20b2024.11 | 67.24 | 34.43 | 34.41 | — | — | — | |
| HICDBackbone=Mistral-7B-v0.32025.03 | 65.63 | 49.75 | 37.92 | — | — | — | |
| DoLaBackbone=Qwen-3-14B-Base2024.11 | 65.09 | 34.43 | 33.17 | — | — | — | |
| Mistral-7B-v0.3Backbone=Mistral-7B-v0.32025.03 | 63.23 | 49.71 | 37.15 | — | — | — | |
| Mixtral-8×7B-IT2024.11 | 62.51 | 37.94 | 35.25 | — | — | — | |
| DoLaBackbone=gpt-oss-20b2024.11 | 61.72 | 28.99 | 30.33 | — | — | — | |
| HICDBackbone=LLaMA-3-8B-Instruct2025.03 | 59.79 | 40.21 | 34.45 | — | — | — | |
| SLEDBackbone=Mixtral-8×7B2024.11 | 57.26 | 35.86 | 32.96 | — | — | — | |
| LLaMA-3-8B-InstructBackbone=LLaMA-3-8B-Instruct2025.03 | 56.91 | 39.31 | 30.43 | — | — | — | |
| HRTNRate=4.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 56.46 | 36.47 | — | — | — | — | |
| DoLaBackbone=LLaMA2-7b2025.03 | 55.91 | 34.51 | 28.81 | — | — | — | |
| RTNRate=4.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 55.77 | 36.84 | — | — | — | — | |
| GPTQRate=3.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 55.21 | 36.11 | — | — | — | — | |
| 13b-CDBackbone=LLaMA2-7b2025.03 | 54.87 | 28.15 | 29.75 | — | — | — | |
| Huffman-GPTQRate=3.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 54.73 | 36.11 | — | — | — | — | |
| GPTQRate=4.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 54.55 | 36.35 | — | — | — | — | |
| BF16Rate=16.000, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 54.5 | 36.35 | — | — | — | — | |
| WaterSICRate=3.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 53.95 | 36.6 | — | — | — | — | |
| WaterSICRate=4.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 53.8 | 35.99 | — | — | — | — | |
| HICDBackbone=LLaMA-7b2025.03 | 53.71 | 25.45 | 26.52 | — | — | — | |
| HRTNRate=3.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 53.68 | 35.13 | — | — | — | — | |
| Huffman-GPTQRate=4.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 53.36 | 35.01 | — | — | — | — | |
| RTNRate=3.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 52.76 | 34.03 | — | — | — | — | |
| AlpacaBackbone=LLaMA-7b2025.03 | 52.47 | 22.88 | 25.19 | — | — | — | |
| DoLaBackbone=LLaMA-7b2025.03 | 52.21 | 31.95 | 28.17 | — | — | — | |
| HRTNRate=2.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 51.78 | 30.48 | — | — | — | — | |
| HICDBackbone=LLaMA2-7b2025.03 | 51.28 | 23.99 | 25.89 | — | — | — | |
| SLEDBackbone=Gemma-2B2024.11 | 50.2 | 25.21 | 26.94 | — | — | — | |
| Mixtral-8×7B2024.11 | 49.98 | 35.13 | 34.17 | — | — | — | |
| SLEDBackbone=Gemma-7B2024.11 | 49.88 | 32.31 | 25.22 | — | — | — | |
| WaterSICRate=2.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 49.11 | 32.8 | — | — | — | — | |
| Huffman-GPTQRate=2.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 49.01 | 31.09 | — | — | — | — | |
| DoLaBackbone=Gemma-2B2024.11 | 48.97 | 26.07 | 26.55 | — | — | — | |
| Gemma-7B2024.11 | 47.63 | 31.58 | 22.75 | — | — | — | |
| RTNRate=2.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 47.33 | 24.11 | — | — | — | — | |
| HICDBackbone=Qwen-7B2025.03 | 47.3 | 25.94 | 27.29 | — | — | — | |
| Qwen-7BBackbone=Qwen-7B2025.03 | 46.95 | 30.59 | 25.06 | — | — | — | |
| GPTQRate=2.125, Evaluation Protocol=Zero-shot, Backbone Model=Qwen3-8B2026.03 | 46.91 | 28.4 | — | — | — | — | |
| Llama-3.3-70BI(M)=0.39, Branch=early2026.05 | 46.47 | 29.74 | — | — | — | — | |
| Qwen3-14BI(M)=2.57, Branch=mix2026.05 | 45.08 | 36.47 | — | — | — | — | |
| GPT-OSS-120BI(M)=5.66, Branch=mix2026.05 | 44.68 | 35.74 | — | — | — | — | |
| Qwen3-30B-A3BI(M)=3.51, Branch=mix2026.05 | 44.6 | 34.88 | — | — | — | — | |
| DeepSeek-R1-Q-32BI(M)=1.18, Branch=mix2026.05 | 44.38 | 33.9 | — | — | — | — | |
| VanillaBackbone=LLaMA2-7b2025.03 | 43.3 | 28.51 | 22.4 | — | — | — | |
| GPT-OSS-20BI(M)=3.48, Branch=mix2026.05 | 43.27 | 32.44 | — | — | — | — | |
| Phi-4-ReasoningI(M)=0.27, Branch=early2026.05 | 43.21 | 34.15 | — | — | — | — | |
| DoLaBackbone=Gemma-7B2024.11 | 43.14 | 25.21 | 26.13 | — | — | — | |
| Gemma-3-1BI(M)=6.54, Branch=mix2026.05 | 41.99 | 27.78 | — | — | — | — | |
| Mixtral-8×7BI(M)=1.30, Branch=mix2026.05 | 41.83 | 29.62 | — | — | — | — | |
| VanillaBackbone=LLaMA-7b2025.03 | 41.21 | 23.62 | 19.33 | — | — | — | |
| Ministral-3-14B-RI(M)=0.57, Branch=early2026.05 | 41.19 | 29.25 | — | — | — | — | |
| 13b-CDBackbone=LLaMA-7b2025.03 | 41.01 | 24.4 | 19.03 | — | — | — | |
| Gemma-3-27BI(M)=5.18, Branch=mix2026.05 | 40.91 | 28.4 | — | — | — | — | |
| LLaMA-30BI(M)=0.13, Branch=early2026.05 | 40.21 | 24.97 | — | — | — | — | |
| Gemma-3-4BI(M)=11.91, Branch=mix2026.05 | 40 | 25.21 | — | — | — | — | |
| LLaMA-13BI(M)=0.16, Branch=early2026.05 | 39.44 | 23.01 | — | — | — | — | |
| DoLaBackbone=Mixtral-8×7B-IT2024.11 | 39.17 | 32.19 | 33.76 | — | — | — | |
| LLaMA-7BI(M)=0.21, Branch=early2026.05 | 39 | 21.66 | — | — | — | — | |
| Gemma-2B2024.11 | 37.16 | 23.38 | 17.42 | — | — | — | |
| DoLaBackbone=Mixtral-8×7B2024.11 | 35.91 | 32.44 | 33.68 | — | — | — | |
| TRACEBase Model=GPT-OSS-20B, I(M)=3.48, Branch=mix2026.05 | 9.14 | 8.08 | — | — | — | — | |
| TRACEBase Model=Phi-4-Reasoning, I(M)=0.27, Branch=early2026.05 | 7.31 | 5.14 | — | — | — | — | |
| TRACEBase Model=Qwen3-30B-A3B, I(M)=3.51, Branch=mix2026.05 | 6.93 | 9.42 | — | — | — | — | |
| TRACEBase Model=GPT-OSS-120B, I(M)=5.66, Branch=mix2026.05 | 6.35 | 4.9 | — | — | — | — | |
| TRACEBase Model=Gemma-3-4B, I(M)=11.91, Branch=mix2026.05 | 6.17 | 5.39 | — | — | — | — | |
| TRACEBase Model=Ministral-3-14B-R, I(M)=0.57, Branch=early2026.05 | 6.01 | 13.34 | — | — | — | — | |
| TRACEBase Model=Qwen3-14B, I(M)=2.57, Branch=mix2026.05 | 4.68 | 5.14 | — | — | — | — | |
| TRACEBase Model=DeepSeek-R1-Q-32B, I(M)=1.18, Branch=mix2026.05 | 4.47 | 8.32 | — | — | — | — | |
| TRACEBase Model=Mixtral-8×7B, I(M)=1.30, Branch=mix2026.05 | 3.97 | 10.65 | — | — | — | — | |
| TRACEBase Model=Gemma-3-1B, I(M)=6.54, Branch=mix2026.05 | 3.68 | 2.2 | — | — | — | — | |
| TRACEBase Model=Gemma-3-27B, I(M)=5.18, Branch=mix2026.05 | 2.89 | 1.71 | — | — | — | — | |
| TRACEBase Model=LLaMA-7B, I(M)=0.21, Branch=early2026.05 | 2.86 | 6.12 | — | — | — | — |