Stylistic Text Generation on Shakespeare style references
1.75Judge ScoreSFT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SFTModel=Mistral-7B, Sparsity=50.46%, KL Comparison=SFT || LCDD2026.05 | 1.75 | 0.853 | |
| SFTModel=DeepSeek-8B, Sparsity=54.54%, KL Comparison=SFT || LCDD2026.05 | 1.73 | 1.52 | |
| LCDDModel=Mistral-7B, Sparsity=50.46%, KL Comparison=SFT || Trig2026.05 | 1.18 | 2.643 | |
| LCDDModel=DeepSeek-8B, Sparsity=54.54%, KL Comparison=SFT || Trig2026.05 | 0.98 | 4.301 | |
| SFTModel=Vicuna-7B, Sparsity=29.88%, KL Comparison=SFT || LCDD2026.05 | 0.8 | 0.636 | |
| SFTModel=Qwen3-0.6B, Sparsity=56.34%, KL Comparison=SFT || LCDD2026.05 | 0.79 | 0.265 | |
| LCDDModel=Qwen3-0.6B, Sparsity=56.34%, KL Comparison=SFT || Trig2026.05 | 0.44 | 2.405 | |
| LCDDModel=Vicuna-7B, Sparsity=29.88%, KL Comparison=SFT || Trig2026.05 | 0.26 | 1.702 | |
| BaseModel=Mistral-7B, Sparsity=50.46%2026.05 | 0.07 | — | |
| TrigModel=Mistral-7B, Sparsity=50.46%, KL Comparison=Base || Trig2026.05 | 0.07 | 2.052 | |
| BaseModel=Vicuna-7B, Sparsity=29.88%2026.05 | 0.05 | — | |
| TrigModel=Qwen3-0.6B, Sparsity=56.34%, KL Comparison=Base || Trig2026.05 | 0.04 | 0.446 | |
| TrigModel=Vicuna-7B, Sparsity=29.88%, KL Comparison=Base || Trig2026.05 | 0.04 | 0.094 | |
| BaseModel=Qwen3-0.6B, Sparsity=56.34%2026.05 | 0.03 | — | |
| BaseModel=DeepSeek-8B, Sparsity=54.54%2026.05 | 0.01 | — | |
| TrigModel=DeepSeek-8B, Sparsity=54.54%, KL Comparison=Base || Trig2026.05 | 0.005 | 0.531 |