Question Answering on HotpotQA (test) (Unreliability Rate Metric)
38.8Unreliability RateKbSD
Evaluation Results
| Method | Links | |
|---|---|---|
| KbSDBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 38.8 | |
| KbSDBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 41.6 | |
| KbPOBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 51.8 | |
| ReACTBackbone=Qwen2.5-7B, Paradigm=Agentic RAG2026.06 | 57.19 | |
| KbPOBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 57.6 | |
| TCRAGBackbone=Qwen2.5-7B, Paradigm=Agentic RAG2026.06 | 59.17 | |
| Search-R1Backbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 61.11 | |
| Search-R1Backbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 62.76 | |
| ARPOBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 63.52 | |
| FL-RAGBackbone=Qwen2.5-7B, Paradigm=Naive RAG2026.06 | 65.58 | |
| ReACTBackbone=Qwen2.5-3B, Paradigm=Agentic RAG2026.06 | 65.63 | |
| ReSearchBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 66.04 | |
| ReSearchBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 69.61 | |
| Mem1Backbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 70.02 | |
| FS-RAGBackbone=Qwen2.5-7B, Paradigm=Naive RAG2026.06 | 70.79 | |
| COTBackbone=Qwen2.5-7B, Paradigm=No RAG2026.06 | 70.9 | |
| AEPOBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 71.29 | |
| FL-RAGBackbone=Qwen2.5-3B, Paradigm=Naive RAG2026.06 | 73.22 | |
| BaseBackbone=Qwen2.5-7B, Paradigm=No RAG2026.06 | 73.37 | |
| IRCOTBackbone=Qwen2.5-7B, Paradigm=Agentic RAG2026.06 | 73.71 | |
| FS-RAGBackbone=Qwen2.5-3B, Paradigm=Naive RAG2026.06 | 74.15 | |
| ARPOBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 74.8 | |
| IRCOTBackbone=Qwen2.5-3B, Paradigm=Agentic RAG2026.06 | 75.5 | |
| BaseBackbone=Qwen2.5-3B, Paradigm=No RAG2026.06 | 75.92 | |
| COTBackbone=Qwen2.5-3B, Paradigm=No RAG2026.06 | 76.18 | |
| TCRAGBackbone=Qwen2.5-3B, Paradigm=Agentic RAG2026.06 | 78.06 | |
| Mem1Backbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 79.85 | |
| AEPOBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 86.15 |