Attribute-controlled Text Generation on Code-UltraFeedback Negative Representative Target Score (test)
0.874DiversityBase
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| BaseBackbone=Llama-3.2-3B2025.10 | 0.874 | 2.95 | — | |
| PromptingBackbone=Phi-4-mini2025.10 | 0.869 | 3.64 | 2.15 | |
| BaseBackbone=Phi-4-mini2025.10 | 0.868 | 3.65 | — | |
| PromptingBackbone=Llama-3.2-3B2025.10 | 0.865 | 2.85 | 6.06 | |
| ITIBackbone=Phi-4-mini2025.10 | 0.623 | 3.66 | 4.54 | |
| Re-ControlBackbone=Phi-4-mini2025.10 | 0.614 | 3.53 | 6.92 | |
| Re-ControlBackbone=Llama-3.2-3B2025.10 | 0.607 | 2.78 | 6.57 | |
| MAT-SteerBackbone=Llama-3.2-3B2025.10 | 0.48 | 2.59 | 16.67 | |
| ITIBackbone=Llama-3.2-3B2025.10 | 0.441 | 2.83 | 6.79 | |
| PRE-CONTROLBackbone=Llama-3.2-3B2025.10 | 0.44 | 1.95 | 30.68 | |
| MAT-SteerBackbone=Phi-4-mini2025.10 | 0.318 | 2.89 | 8.38 | |
| PRE-CONTROLBackbone=Phi-4-mini2025.10 | 0.279 | 2.8 | 22.34 |