Multi-hop Question Answering on Average (test)
51.3EMARBOR
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ARBORBackbone=Qwen3-14B2026.06 | 51.3 | 60 | 62.8 | |
| GRPOBackbone=Qwen3-14B2026.06 | 49.8 | 59 | 60.8 | |
| DAPOBackbone=Qwen3-14B2026.06 | 49.2 | 58 | 59.1 | |
| ARBORBackbone=Qwen3-8B2026.06 | 48.9 | 57.9 | 61.1 | |
| ARBORBackbone=Qwen3-4B2026.06 | 47.9 | 56.7 | 59.6 | |
| DAPOBackbone=Qwen3-8B2026.06 | 47.1 | 55.1 | 56.7 | |
| GRPOBackbone=Qwen3-8B2026.06 | 46.9 | 55.5 | 56.9 | |
| GRPOBackbone=Qwen3-4B2026.06 | 46.1 | 54.9 | 55.6 | |
| DAPOBackbone=Qwen3-4B2026.06 | 46.1 | 55.1 | 56.1 | |
| TIR PromptingBackbone=Qwen3-4B2026.06 | 29.7 | 38.1 | 43.6 | |
| TIR PromptingBackbone=Qwen3-14B2026.06 | 26.2 | 33.2 | 35.2 | |
| TIR PromptingBackbone=Qwen3-8B2026.06 | 15.7 | 20.5 | 22.7 |