Negotiation on CRAD (test)
100Success RateIQL+SFT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| IQL+SFTBackbone=Qwen2.5-7B-Instruct2026.05 | 100 | 69.8 | 10.1 | |
| IQLBackbone=Qwen3.5-Plus2026.05 | 95 | 63.6 | 10.4 | |
| IQL+JPOBackbone=Qwen2.5-7B-Instruct2026.05 | 95 | 51.7 | 14.1 | |
| IQL+SFT+JPOBackbone=Qwen2.5-7B-Instruct, Note=Full EmoDistill pipeline2026.05 | 90 | 72.2 | 15 | |
| Random2026.05 | 85 | 40.6 | 13.9 | |
| Vanilla (LLM)Backbone=Qwen3.5-Plus2026.05 | 50 | 5 | 8.5 | |
| Vanilla (SLM)Backbone=Qwen2.5-7B-Instruct2026.05 | 25 | 8.8 | 11.2 |