Search-augmented Multi-turn Question Answering on MuSiQue (EM, F1)
17.8Exact Match (EM)IGPO (gold-sup.)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| IGPO (gold-sup.)Backbone=Qwen3-8B2026.05 | 17.8 | 27.6 | |
| GRPO (gold-sup.)Backbone=Qwen3-8B2026.05 | 14.1 | 22.4 | |
| TTRLBackbone=Qwen3-8B2026.05 | 13.3 | 19.2 | |
| GRPO (gold-sup.)Backbone=Qwen3-4B2026.05 | 12.1 | 18.3 | |
| EMPOBackbone=Qwen3-8B2026.05 | 12 | 19.3 | |
| IGPO (gold-sup.)Backbone=Qwen3-4B2026.05 | 11.3 | 18.4 | |
| SIOPBackbone=Qwen3-8B2026.05 | 10.7 | 20.3 | |
| SIOPBackbone=Qwen3-4B2026.05 | 10 | 17.9 | |
| TTRLBackbone=Qwen3-4B2026.05 | 9 | 15.7 | |
| EMPOBackbone=Qwen3-4B2026.05 | 9 | 15.9 |