Multi-Hop Question Answering on 2Wiki (Exact Match (EM))
46.9Exact Match (EM)Thinker-Instruct
Evaluation Results
| Method | Links | |
|---|---|---|
| Thinker-InstructTraining Pipeline=Synthesize → SFT, External Teacher=Qwen2.5-72B, Backbone=Qwen2.5-7B2026.05 | 46.9 | |
| SD-Search-InstructTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.05 | 44.1 | |
| SD-Search-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.05 | 43.8 | |
| MR-Search-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.05 | 43.6 | |
| SKILLGRAPHEvaluation Protocol=Zero-shot2026.05 | 43.4 | |
| EvolveREvaluation Protocol=Zero-shot2026.05 | 42 | |
| Search-R1-InstructTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.05 | 41.4 | |
| AutoRefine-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.05 | 40.5 | |
| SkillRLEvaluation Protocol=Zero-shot2026.05 | 40.3 | |
| Search-R1Evaluation Protocol=Zero-shot2026.05 | 40.1 | |
| StepSearch-BaseTraining Pipeline=Pure RL, External Teacher=GPT-4o, Backbone=Qwen2.5-7B2026.05 | 38.5 | |
| StepSearch-InstructTraining Pipeline=Pure RL, External Teacher=GPT-4o, Backbone=Qwen2.5-7B2026.05 | 36.6 | |
| ZeroSearchEvaluation Protocol=Zero-shot2026.05 | 35.2 | |
| Search-R1-BaseTraining Pipeline=Pure RL, External Teacher=None, Backbone=Qwen2.5-7B2026.05 | 27.2 | |
| Search-o1Evaluation Protocol=Zero-shot2026.05 | 27 | |
| RAGEvaluation Protocol=Zero-shot2026.05 | 23.2 | |
| CoTEvaluation Protocol=Zero-shot2026.05 | 22.6 | |
| Qwen2.5-7B-InstructEvaluation Protocol=Zero-shot2026.05 | 22.2 |