Multi-Hop Question Answering on HotpotQA (EM, ACC)
44.6Exact Match (EM)IGRPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| IGRPOBackbone=Qwen2.5-7B-Instruct, Type=Tree-based2026.07 | 44.6 | — | |
| IGPOBackbone=Qwen2.5-7B-Instruct, Type=Chain-based2026.07 | 42.9 | — | |
| GiGPOBackbone=Qwen2.5-7B-Instruct, Type=Chain-based2026.07 | 41.6 | — | |
| Tree-GRPOBackbone=Qwen2.5-7B-Instruct, Type=Tree-based2026.07 | 39.9 | — | |
| IGRPOBackbone=Qwen2.5-3B-Instruct, Type=Tree-based2026.07 | 38.9 | — | |
| StepSearchBackbone=Qwen2.5-7B-Instruct, Type=Chain-based2026.07 | 38.6 | — | |
| DTKG2025.10 | 38.2 | 85.8 | |
| TOG2025.10 | 37.1 | 83.5 | |
| Search-R1Backbone=Qwen2.5-7B-Instruct, Type=Chain-based2026.07 | 37 | — | |
| GiGPOBackbone=Qwen2.5-3B-Instruct, Type=Chain-based2026.07 | 36.9 | — | |
| IGPOBackbone=Qwen2.5-3B-Instruct, Type=Chain-based2026.07 | 35.7 | — | |
| COT2025.10 | 35.6 | 81 | |
| KGR2025.10 | 35.5 | 83.5 | |
| Original2025.10 | 35 | 68.2 | |
| ZeroSearchBackbone=Qwen2.5-7B-Instruct, Type=Chain-based2026.07 | 34.6 | — | |
| StepSearchBackbone=Qwen2.5-3B-Instruct, Type=Chain-based2026.07 | 34.5 | — | |
| CRITIC2025.10 | 34 | 78 | |
| Search-R1Backbone=Qwen2.5-3B-Instruct, Type=Chain-based2026.07 | 32.4 | — | |
| Tree-GRPOBackbone=Qwen2.5-3B-Instruct, Type=Tree-based2026.07 | 31.1 | — | |
| ZeroSearchBackbone=Qwen2.5-3B-Instruct, Type=Chain-based2026.07 | 27.4 | — | |
| R1-InstructBackbone=Qwen2.5-3B-Instruct, Type=RL Training2026.07 | 23.7 | — | |
| R1-InstructBackbone=Qwen2.5-7B-Instruct, Type=RL Training2026.07 | 20.8 | — |