Attribute-controlled Text Generation on Code-UltraFeedback Relative Positive Representative Target
90.3DiversityPrompting
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PromptingBackbone=Phi-4-mini2025.10 | 90.3 | 1.47 | 946 | |
| BaseBackbone=Phi-4-mini2025.10 | 90.2 | 1.57 | — | |
| Re-ControlBackbone=Llama-3.2-3B2025.10 | 88 | 2.21 | 754 | |
| PromptingBackbone=Llama-3.2-3B2025.10 | 87.9 | 2.21 | 680 | |
| BaseBackbone=Llama-3.2-3B2025.10 | 87.6 | 2.29 | — | |
| ITIBackbone=Phi-4-mini2025.10 | 78.9 | 1.55 | 1,649 | |
| Re-ControlBackbone=Phi-4-mini2025.10 | 78.6 | 1.43 | 1,725 | |
| MAT-SteerBackbone=Llama-3.2-3B2025.10 | 77.8 | 2.41 | 1,363 | |
| ITIBackbone=Llama-3.2-3B2025.10 | 74.1 | 2.62 | 1,272 | |
| MAT-SteerBackbone=Phi-4-mini2025.10 | 70 | 1.43 | 1,892 | |
| PRE-CONTROLBackbone=Phi-4-mini2025.10 | 68.8 | 1.42 | 2,616 | |
| PRE-CONTROLBackbone=Llama-3.2-3B2025.10 | 61.4 | 2.08 | 1,746 |