Multi-hop Question Answering on Bamboogle out-of-domain (Exact Match)
46.4Exact MatchEM+Info-Think+Think-Ans
Evaluation Results
| Method | Links | |
|---|---|---|
| EM+Info-Think+Think-AnsFramework=VERITAS-R1, Reward Components=EM+Info-Think+Think-Ans2025.10 | 46.4 | |
| EM+Info-ThinkFramework=VERITAS-R1, Reward Components=EM+Info-Think2025.10 | 45.6 | |
| ReSearch-7B-BaseModel Scale=7B, Model Variant=Base2025.10 | 43.2 | |
| Search-R1-7B-Base-PPOModel Scale=7B, Training Algorithm=PPO2025.10 | 43.2 | |
| ReSearch-7B-InstructModel Scale=7B, Model Variant=Instruct2025.10 | 42.4 | |
| EM+Think-AnsFramework=VERITAS-R1, Reward Components=EM+Think-Ans2025.10 | 41.6 | |
| Search-R1-7B-Base-PPO w/ FormatModel Scale=7B, Training Algorithm=PPO, Input Formatting=Included2025.10 | 40.3 | |
| DeSA-7B-Instrct-GRPOModel Scale=7B, Training Algorithm=GRPO2025.10 | 39.5 | |
| Rejection SamplingReasoning Strategy=Rejection Sampling2025.10 | 35.5 | |
| DeSA-3B-Instruct-GRPOModel Scale=3B, Training Algorithm=GRPO2025.10 | 34.7 | |
| Search-o1Reasoning Strategy=Search-o12025.10 | 29.6 | |
| R1-baseModel Variant=Base2025.10 | 29.6 | |
| R1-instructModel Variant=Instruct2025.10 | 29.3 | |
| CoTReasoning Strategy=Chain-of-Thought2025.10 | 23.2 | |
| IRCoTReasoning Strategy=Iterative Retrieval Chain-of-Thought2025.10 | 22.4 | |
| RAGReasoning Strategy=Retrieval-Augmented Generation2025.10 | 20.8 | |
| Direct InferenceReasoning Strategy=Direct Inference2025.10 | 12 | |
| SFTTraining Algorithm=Supervised Fine-Tuning2025.10 | 11.2 |