Question Answering on HaluEval
68EMToken-Guard
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Token-GuardBackbone=Meta-Llama-3.1-8B-Instruct2026.01 | 68 | 78.54 | |
| Guided DecodingBackbone=Qwen3-8B2026.01 | 62 | 69.88 | |
| Token-GuardBackbone=Qwen3-8B2026.01 | 51 | 74.15 | |
| BaseModelBackbone=Qwen3-8B2026.01 | 43 | 59.83 | |
| Guided DecodingBackbone=Meta-Llama-3.1-8B-Instruct2026.01 | 42 | 57.41 | |
| Chain-of-ThoughtsBackbone=Meta-Llama-3.1-8B-Instruct2026.01 | 40 | 55.32 | |
| Tree-of-ThoughtBackbone=Qwen3-8B2026.01 | 39 | 57.33 | |
| Tree-of-ThoughtBackbone=Meta-Llama-3.1-8B-Instruct2026.01 | 38 | 56.02 | |
| Chain-of-ThoughtsBackbone=Qwen3-8B2026.01 | 34 | 53.21 | |
| BaseModelBackbone=Meta-Llama-3.1-8B-Instruct2026.01 | 32 | 42.16 | |
| Predictive DecodingBackbone=Meta-Llama-3.1-8B-Instruct2026.01 | 22 | 38 | |
| Predictive DecodingBackbone=Qwen3-8B2026.01 | 0 | 21.29 |