Controlled Text Generation on Syntactic Control (Q = p) (test)
-22.71Log Probability p(y)Llama3-8B (5-shot) + SIS
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Llama3-8B (5-shot) + SISBackbone=Llama3-8B, Algorithm=SIS, Few-shot setting=5-shot2025.06 | -22.71 | -43.94 | 64.44 | 0.86 | 0.84 | 0.72 | |
| Llama3-8B (0-shot) + SMCBackbone=Llama3-8B, Algorithm=SMC, Few-shot setting=0-shot2025.06 | -22.94 | -54.58 | 58.18 | 0.89 | 0.86 | 0.74 | |
| Llama3-8B (0-shot)Backbone=Llama3-8B, Algorithm=None, Few-shot setting=0-shot2025.06 | -23.09 | -166.86 | 29.41 | 0.86 | 0.83 | 0.7 | |
| Llama3-8B (5-shot)Backbone=Llama3-8B, Algorithm=None, Few-shot setting=5-shot2025.06 | -23.12 | -133.95 | 35.33 | 0.88 | 0.84 | 0.7 | |
| Llama3-8B (0-shot) + SISBackbone=Llama3-8B, Algorithm=SIS, Few-shot setting=0-shot2025.06 | -23.34 | -65.66 | 56.42 | 0.83 | 0.83 | 0.72 | |
| Llama3-8B (5-shot) + SMCBackbone=Llama3-8B, Algorithm=SMC, Few-shot setting=5-shot2025.06 | -24.02 | -40.89 | 63.66 | 0.92 | 0.87 | 0.74 | |
| GPT2-largeBackbone=GPT2-large, Algorithm=None, Few-shot setting=None2025.06 | -52.64 | -176.8 | 12.31 | 0.9 | 0.85 | 0.71 | |
| GPT2-large + SISBackbone=GPT2-large, Algorithm=SIS, Few-shot setting=None2025.06 | -56.23 | -93.45 | 21.18 | 0.9 | 0.84 | 0.71 | |
| GPT2-large + SMCBackbone=GPT2-large, Algorithm=SMC, Few-shot setting=None2025.06 | -58.56 | -89.89 | 28.26 | 0.91 | 0.85 | 0.71 | |
| Llama3-8BBackbone=Llama3-8B, Algorithm=None, Few-shot setting=None2025.06 | -62.47 | — | 10.05 | 0.77 | 0.71 | 0.61 | |
| Llama3-8B + SISBackbone=Llama3-8B, Algorithm=SIS, Few-shot setting=None2025.06 | -79.79 | -53.86 | 16.45 | 0.92 | 0.85 | 0.72 | |
| Llama3-8B + SMCBackbone=Llama3-8B, Algorithm=SMC, Few-shot setting=None2025.06 | -117.88 | -64.12 | 16.12 | 0.91 | 0.86 | 0.75 |