Question Answering on FinanceBench
45EMToken-Guard
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Token-GuardBackbone=Qwen3-8B2026.01 | 45 | 45.37 | |
| Token-GuardBackbone=Meta-Llama-3.1-8B-Instruct2026.01 | 30 | 30.8 | |
| Chain-of-ThoughtsBackbone=Qwen3-8B2026.01 | 29 | 35.12 | |
| Tree-of-ThoughtBackbone=Qwen3-8B2026.01 | 28 | 34.91 | |
| Guided DecodingBackbone=Qwen3-8B2026.01 | 21 | 23.56 | |
| BaseModelBackbone=Qwen3-8B2026.01 | 20 | 26.67 | |
| BaseModelBackbone=Meta-Llama-3.1-8B-Instruct2026.01 | 16 | 16 | |
| Guided DecodingBackbone=Meta-Llama-3.1-8B-Instruct2026.01 | 14 | 16.44 | |
| Chain-of-ThoughtsBackbone=Meta-Llama-3.1-8B-Instruct2026.01 | 11 | 11.01 | |
| Tree-of-ThoughtBackbone=Meta-Llama-3.1-8B-Instruct2026.01 | 10 | 14.44 | |
| Predictive DecodingBackbone=Meta-Llama-3.1-8B-Instruct2026.01 | 9 | 8.79 | |
| Predictive DecodingBackbone=Qwen3-8B2026.01 | 4 | 11.25 |