Multi-Hop Question Answering on MuSiQue (EM & LLM)
21.7EMSearch-R1 (GRPO/OF)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Search-R1 (GRPO/OF)Backbone=Qwen3-8B, FT=Full2026.06 | 21.7 | 29.6 | |
| DACBackbone=Qwen3-8B, FT=LoRA2026.06 | 21.1 | 28.5 | |
| Search-R1 (PPO/OF)Backbone=Qwen3-8B, FT=Full2026.06 | 19.4 | 26.2 | |
| Search-R1 (GRPO/O)Backbone=Qwen3-8B, FT=Full2026.06 | 18.3 | 26.1 | |
| DACBackbone=Qwen2.5-7B-Instruct, FT=LoRA2026.06 | 17.2 | 23.7 | |
| Search-R1 (PPO/O)Backbone=Qwen3-8B, FT=Full2026.06 | 16.5 | 22 | |
| Search-R1 (GRPO/OF)Backbone=Qwen2.5-7B-Instruct, FT=Full2026.06 | 15.8 | 23.1 | |
| Search-R1 (PPO/O)Backbone=Qwen2.5-7B-Instruct, FT=Full2026.06 | 15.6 | 21.6 | |
| Search-R1 (GRPO/O)Backbone=Qwen2.5-7B-Instruct, FT=Full2026.06 | 15.6 | 21.8 | |
| Search-R1 (PPO/OF)Backbone=Qwen2.5-7B-Instruct, FT=Full2026.06 | 12.7 | 19.6 | |
| IRCoTBackbone=Qwen3-8B, FT=-2026.06 | 12 | 17.7 | |
| Search-o1Backbone=Qwen2.5-7B-Instruct, FT=-2026.06 | 11.6 | 18.5 | |
| Search-o1Backbone=Qwen3-8B, FT=-2026.06 | 11.3 | 22.1 | |
| Naive RAGBackbone=Qwen3-8B, FT=-2026.06 | 7 | 11 | |
| Naive RAGBackbone=Qwen2.5-7B-Instruct, FT=-2026.06 | 6.4 | 9.9 | |
| CoTBackbone=Qwen3-8B, FT=-2026.06 | 5.8 | 10.9 | |
| CoTBackbone=Qwen2.5-7B-Instruct, FT=-2026.06 | 5.3 | 10.2 | |
| IRCoTBackbone=Qwen2.5-7B-Instruct, FT=-2026.06 | 4.8 | 11.1 | |
| DirectBackbone=Qwen2.5-7B-Instruct, FT=-2026.06 | 4.3 | 7.9 | |
| DirectBackbone=Qwen3-8B, FT=-2026.06 | 3.4 | 7.8 |