Stylized Dialogue on 148-query Average across 9 styles (test)
4.463Context RelevanceSFR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SFRTemperature=0.8, Backbone=Qwen3-32B, LLM-judge=Qwen3-235B2026.05 | 4.463 | 4.666 | 4.401 | 4.881 | |
| SFTTemperature=0.8, Backbone=Qwen3-32B, LLM-judge=Qwen3-235B2026.05 | 4.427 | 4.661 | 4.243 | 4.86 | |
| DeepSeek-R1-promptTemperature=0.8, Backbone=DeepSeek-R1 (frozen), LLM-judge=Qwen3-235B2026.05 | 4.368 | 4.604 | 4.329 | 4.811 |