Question Answering on TriviaQA (test) (Unreliability Rate (%))
28.9Unreliability RateKbSD
Evaluation Results
| Method | Links | |
|---|---|---|
| KbSDBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 28.9 | |
| KbPOBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 32.6 | |
| KbSDBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 34.8 | |
| KbPOBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 39.9 | |
| Search-R1Backbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 44.09 | |
| Search-R1Backbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 44.92 | |
| TCRAGBackbone=Qwen2.5-7B, Paradigm=Agentic RAG2026.06 | 45.22 | |
| ReACTBackbone=Qwen2.5-7B, Paradigm=Agentic RAG2026.06 | 45.45 | |
| ReSearchBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 46.07 | |
| ARPOBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 46.34 | |
| Mem1Backbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 48.96 | |
| TCRAGBackbone=Qwen2.5-3B, Paradigm=Agentic RAG2026.06 | 49.54 | |
| IRCOTBackbone=Qwen2.5-7B, Paradigm=Agentic RAG2026.06 | 50.57 | |
| COTBackbone=Qwen2.5-7B, Paradigm=No RAG2026.06 | 50.67 | |
| IRCOTBackbone=Qwen2.5-3B, Paradigm=Agentic RAG2026.06 | 50.81 | |
| BaseBackbone=Qwen2.5-7B, Paradigm=No RAG2026.06 | 50.92 | |
| FL-RAGBackbone=Qwen2.5-3B, Paradigm=Naive RAG2026.06 | 51.5 | |
| ReSearchBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 51.75 | |
| AEPOBackbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 52.22 | |
| ReACTBackbone=Qwen2.5-3B, Paradigm=Agentic RAG2026.06 | 53.96 | |
| FS-RAGBackbone=Qwen2.5-3B, Paradigm=Naive RAG2026.06 | 54.62 | |
| FL-RAGBackbone=Qwen2.5-7B, Paradigm=Naive RAG2026.06 | 57.34 | |
| ARPOBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 59.84 | |
| COTBackbone=Qwen2.5-3B, Paradigm=No RAG2026.06 | 60.38 | |
| BaseBackbone=Qwen2.5-3B, Paradigm=No RAG2026.06 | 61.16 | |
| FS-RAGBackbone=Qwen2.5-7B, Paradigm=Naive RAG2026.06 | 64.98 | |
| Mem1Backbone=Qwen2.5-3B, Paradigm=RL-based Agentic RAG2026.06 | 66.91 | |
| AEPOBackbone=Qwen2.5-7B, Paradigm=RL-based Agentic RAG2026.06 | 82.47 |