Dialogue Utterance Generation on User Fidelity Experiment Dataset
42.7ROUGE-Lutt-only-14B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| utt-only-14BParameter Count=14B, Training Mode=Utterance-only fine-tuning2026.06 | 42.7 | 15.9 | 1.03 | |
| CogWM-14BParameter Count=14B, Training Mode=Joint Optimization2026.06 | 41.7 | 15 | 1.01 | |
| DS-V4-Pro2026.06 | 28.8 | 6 | 1.81 | |
| Dual-LLMBackbone=GPT-5.52026.06 | 25.3 | 5.4 | 2.12 | |
| GPT-5.52026.06 | 23.7 | 4.3 | 1.82 | |
| Qwen3-14BParameter Count=14B2026.06 | 19.6 | 3.3 | 2.26 |