Text Generation on TruthfulQA (test)
2.8LCTG Error RateMARKERGEN
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MARKERGENBackbone=Llama3.1-70B2025.02 | 2.8 | 4.48 | — | |
| MARKERGENBackbone=Llama3.1-8B2025.02 | 3.82 | 4.25 | — | |
| MARKERGENBackbone=Qwen2.5-32B2025.02 | 4.48 | 4.54 | — | |
| ImplicitBackbone=Llama3.1-8B2025.02 | 7.21 | 4.22 | — | |
| MARKERGENBackbone=Qwen2.5-14B2025.02 | 7.59 | 4.43 | — | |
| ImplicitBackbone=Llama3.1-70B2025.02 | 7.64 | 4.46 | — | |
| ImplicitBackbone=Qwen2.5-32B2025.02 | 8.7 | 4.45 | — | |
| MARKERGENBackbone=Qwen2.5-7B2025.02 | 9.08 | 4.33 | — | |
| ImplicitBackbone=Qwen2.5-7B2025.02 | 16.7 | 4.29 | — | |
| ImplicitBackbone=Qwen2.5-14B2025.02 | 17.9 | 4.44 | — |