Instruction Hierarchy Adherence on IHEval Aligned Single-turn
84.8AccuracyVanilla
Evaluation Results
| Method | Links | |
|---|---|---|
| VanillaModel=Qwen3-8B-Instruct2026.06 | 84.8 | |
| VerIHModel=Qwen3-8B-Instruct2026.06 | 84.59 | |
| PromptingModel=Qwen3-8B-Instruct2026.06 | 83.6 | |
| IHDec + VerIHModel=Qwen3-8B-Instruct2026.06 | 81.8 | |
| IHDec + PromptingModel=Qwen3-8B-Instruct2026.06 | 78.35 | |
| IHDecModel=Qwen3-8B-Instruct2026.06 | 77.2 | |
| IHDecModel=Qwen3-4B-Instruct2026.06 | 76.7 | |
| PromptingModel=Qwen3-4B-Instruct2026.06 | 75.9 | |
| IHDec + PromptingModel=Qwen3-4B-Instruct2026.06 | 75.9 | |
| IHDecModel=Llama-3.1-8B-Instruct2026.06 | 73.53 | |
| VanillaModel=Qwen3-4B-Instruct2026.06 | 71 | |
| VanillaModel=Llama-3.1-8B-Instruct2026.06 | 66.05 | |
| IHDec + PromptingModel=Llama-3.1-8B-Instruct2026.06 | 61.8 | |
| PromptingModel=Llama-3.1-8B-Instruct2026.06 | 56.5 | |
| IHDec + ISEModel=Llama-3.1-8B-Instruct2026.06 | 55.1 | |
| ISEModel=Llama-3.1-8B-Instruct2026.06 | 42.08 |