Instruction Hierarchy Adherence on IHEval Aligned Multi-turn
88.8AccuracyPrompting
Evaluation Results
| Method | Links | |
|---|---|---|
| PromptingModel=Qwen3-8B-Instruct2026.06 | 88.8 | |
| IHDec + VerIHModel=Qwen3-8B-Instruct2026.06 | 87.3 | |
| VerIHModel=Qwen3-8B-Instruct2026.06 | 86.29 | |
| VanillaModel=Qwen3-8B-Instruct2026.06 | 85.6 | |
| IHDecModel=Qwen3-8B-Instruct2026.06 | 83.5 | |
| IHDec + PromptingModel=Qwen3-4B-Instruct2026.06 | 82.2 | |
| PromptingModel=Qwen3-4B-Instruct2026.06 | 78.6 | |
| IHDec + PromptingModel=Qwen3-8B-Instruct2026.06 | 77.4 | |
| IHDecModel=Qwen3-4B-Instruct2026.06 | 76.7 | |
| VanillaModel=Qwen3-4B-Instruct2026.06 | 75.1 | |
| IHDecModel=Llama-3.1-8B-Instruct2026.06 | 68.38 | |
| VanillaModel=Llama-3.1-8B-Instruct2026.06 | 65.9 | |
| IHDec + PromptingModel=Llama-3.1-8B-Instruct2026.06 | 57.3 | |
| PromptingModel=Llama-3.1-8B-Instruct2026.06 | 53.9 | |
| IHDec + ISEModel=Llama-3.1-8B-Instruct2026.06 | 37.6 | |
| ISEModel=Llama-3.1-8B-Instruct2026.06 | 36 |