Question Answering on ChineseFactEval
70.4AccuracyHaluSearch (MCTS)
Evaluation Results
| Method | Links | |
|---|---|---|
| HaluSearch (MCTS)Backbone=Qwen2-7B-Instruct2025.01 | 70.4 | |
| BoNBackbone=Qwen2-7B-Instruct2025.01 | 64 | |
| SCBackbone=Qwen2-7B-Instruct2025.01 | 59.2 | |
| Direct GenerationBackbone=Qwen2-7B-Instruct2025.01 | 56 | |
| CoTBackbone=Qwen2-7B-Instruct2025.01 | 54.4 | |
| Self-RefineBackbone=Qwen2-7B-Instruct2025.01 | 48.8 | |
| HaluSearch (MCTS)Backbone=Llama3.1-8B-Instruct2025.01 | 40.8 | |
| SCBackbone=Llama3.1-8B-Instruct2025.01 | 36 | |
| BoNBackbone=Llama3.1-8B-Instruct2025.01 | 35.6 | |
| Self-RefineBackbone=Llama3.1-8B-Instruct2025.01 | 33.6 | |
| CoTBackbone=Llama3.1-8B-Instruct2025.01 | 32 | |
| Direct GenerationBackbone=Llama3.1-8B-Instruct2025.01 | 29.6 |