Document Retrieval on MuSiQue (test)
64Per-query Final RecallSARDI τc=0.95
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SARDI τc=0.95Retriever=BM25, K=15, Backbone=Qwen2.5-7B2026.06 | 64 | — | |
| AR ret@1Retriever=BM25, K=15, Backbone=Qwen2.5-7B2026.06 | 62 | — | |
| AR ret@10Retriever=BM25, K=15, Backbone=Qwen2.5-7B2026.06 | 61 | — | |
| SARDI τc=0.9Retriever=BM25, K=15, Backbone=Qwen2.5-7B2026.06 | 60 | — | |
| SARDI τc=0.95Retriever=BM25, K=10, Backbone=Qwen2.5-7B2026.06 | 59 | — | |
| AR ret@1Retriever=BM25, K=10, Backbone=Qwen2.5-7B2026.06 | 58 | — | |
| AR ret@10Retriever=BM25, K=10, Backbone=Qwen2.5-7B2026.06 | 56 | — | |
| SARDI τc=0.9Retriever=BM25, K=10, Backbone=Qwen2.5-7B2026.06 | 56 | — | |
| SARDI τc=0.95Retriever=BM25, K=7, Backbone=Qwen2.5-7B2026.06 | 56 | — | |
| SARDI τc=0.95Retriever=Dense, K=15, Backbone=Qwen2.5-7B2026.06 | 56 | — | |
| SARDI τc=0.9Retriever=BM25, K=7, Backbone=Qwen2.5-7B2026.06 | 55 | — | |
| SARDI τc=0.9Retriever=Dense, K=15, Backbone=Qwen2.5-7B2026.06 | 55 | — | |
| AR ret@10Retriever=Dense, K=15, Backbone=Qwen2.5-7B2026.06 | 54 | — | |
| AR ret@1Retriever=BM25, K=7, Backbone=Qwen2.5-7B2026.06 | 54 | — | |
| AR ret@1Retriever=Dense, K=15, Backbone=Qwen2.5-7B2026.06 | 54 | — | |
| AR ret@10Retriever=BM25, K=7, Backbone=Qwen2.5-7B2026.06 | 53 | — | |
| SARDI τc=0.9Retriever=Dense, K=10, Backbone=Qwen2.5-7B2026.06 | 53 | — | |
| SARDI τc=0.95Retriever=Dense, K=10, Backbone=Qwen2.5-7B2026.06 | 53 | — | |
| AR ret@10Retriever=Dense, K=10, Backbone=Qwen2.5-7B2026.06 | 51 | — | |
| AR ret@1Retriever=Dense, K=10, Backbone=Qwen2.5-7B2026.06 | 51 | — | |
| SARDI τc=0.9Retriever=Dense, K=7, Backbone=Qwen2.5-7B2026.06 | 50 | — | |
| SARDI τc=0.95Retriever=Dense, K=7, Backbone=Qwen2.5-7B2026.06 | 50 | — | |
| AR ret@1Retriever=Dense, K=7, Backbone=Qwen2.5-7B2026.06 | 48 | — | |
| AR ret@10Retriever=Dense, K=7, Backbone=Qwen2.5-7B2026.06 | 47 | — | |
| Search-R1Retriever=BM25, K=15, Training Protocol=RL-trained (PPO)2026.06 | 45 | — | |
| SARDI τc=0.95Retriever=BM25, K=3, Backbone=Qwen2.5-7B2026.06 | 44 | — | |
| AR ret@1Retriever=BM25, K=3, Backbone=Qwen2.5-7B2026.06 | 43 | — | |
| SARDI τc=0.9Retriever=BM25, K=3, Backbone=Qwen2.5-7B2026.06 | 43 | — | |
| AR ret@10Retriever=BM25, K=3, Backbone=Qwen2.5-7B2026.06 | 42 | — | |
| Search-R1Retriever=BM25, K=10, Training Protocol=RL-trained (PPO)2026.06 | 42 | — | |
| Search-R1Retriever=Dense, K=15, Training Protocol=RL-trained (PPO)2026.06 | 41 | — | |
| ReActRetriever=BM25, K=15, Backbone=Qwen2.5-7B2026.06 | 40 | — | |
| SARDI τc=0.95Retriever=Dense, K=3, Backbone=Qwen2.5-7B2026.06 | 40 | — | |
| SARDI τc=0.9Retriever=Dense, K=3, Backbone=Qwen2.5-7B2026.06 | 39 | — | |
| AR ret@10Retriever=Dense, K=3, Backbone=Qwen2.5-7B2026.06 | 38 | — | |
| AR ret@1Retriever=Dense, K=3, Backbone=Qwen2.5-7B2026.06 | 38 | — | |
| Search-R1Retriever=BM25, K=7, Training Protocol=RL-trained (PPO)2026.06 | 38 | — | |
| ReActRetriever=BM25, K=10, Backbone=Qwen2.5-7B2026.06 | 37 | — | |
| Search-R1Retriever=Dense, K=10, Training Protocol=RL-trained (PPO)2026.06 | 37 | — | |
| ReActRetriever=Dense, K=15, Backbone=Qwen2.5-7B2026.06 | 35 | — | |
| Search-R1Retriever=Dense, K=7, Training Protocol=RL-trained (PPO)2026.06 | 35 | — | |
| ReActRetriever=BM25, K=7, Backbone=Qwen2.5-7B2026.06 | 34 | — | |
| ReActRetriever=Dense, K=10, Backbone=Qwen2.5-7B2026.06 | 32 | — | |
| ReActRetriever=Dense, K=7, Backbone=Qwen2.5-7B2026.06 | 31 | — | |
| Search-R1Retriever=BM25, K=3, Training Protocol=RL-trained (PPO)2026.06 | 29 | — | |
| Search-R1Retriever=Dense, K=3, Training Protocol=RL-trained (PPO)2026.06 | 28 | — | |
| ReActRetriever=Dense, K=3, Backbone=Qwen2.5-7B2026.06 | 27 | — | |
| ReActRetriever=BM25, K=3, Backbone=Qwen2.5-7B2026.06 | 26 | — | |
| AR ret@adaptive (FLARE)Retriever=BM25, K=15, Backbone=Qwen2.5-7B2026.06 | 25 | — | |
| AR ret@adaptive (FLARE)Retriever=BM25, K=10, Backbone=Qwen2.5-7B2026.06 | 22 | — | |
| AR ret@adaptive (FLARE)Retriever=Dense, K=15, Backbone=Qwen2.5-7B2026.06 | 22 | — | |
| AR ret@adaptive (FLARE)Retriever=Dense, K=10, Backbone=Qwen2.5-7B2026.06 | 21 | — | |
| AR ret@adaptive (FLARE)Retriever=BM25, K=7, Backbone=Qwen2.5-7B2026.06 | 20 | — | |
| AR ret@adaptive (FLARE)Retriever=Dense, K=7, Backbone=Qwen2.5-7B2026.06 | 20 | — | |
| AR ret@adaptive (FLARE)Retriever=BM25, K=3, Backbone=Qwen2.5-7B2026.06 | 14 | — | |
| AR ret@adaptive (FLARE)Retriever=Dense, K=3, Backbone=Qwen2.5-7B2026.06 | 14 | — | |
| AdaptiveRAGRetriever=BM25, K=7, Backbone=Qwen2.5-7B2026.06 | 10 | — | |
| AdaptiveRAGRetriever=BM25, K=3, Backbone=Qwen2.5-7B2026.06 | 9 | — | |
| AdaptiveRAGRetriever=BM25, K=10, Backbone=Qwen2.5-7B2026.06 | 9 | — | |
| AdaptiveRAGRetriever=Dense, K=7, Backbone=Qwen2.5-7B2026.06 | 9 | — | |
| AdaptiveRAGRetriever=BM25, K=15, Backbone=Qwen2.5-7B2026.06 | 7 | — | |
| BM25Knowledge Representation Type=Unstructured Knowledge Representation Augmented2026.06 | — | 43.5 | |
| ContrieverKnowledge Representation Type=Unstructured Knowledge Representation Augmented2026.06 | — | 46.6 | |
| GritLM-7BKnowledge Representation Type=Embedding-based Knowledge Representation Augmented, Backbone=GritLM, Number of Parameters=7B2026.06 | — | 65.9 | |
| GTE-Qwen2-7B-InstructKnowledge Representation Type=Embedding-based Knowledge Representation Augmented, Backbone=Qwen2-7B-Instruct, Number of Parameters=7B2026.06 | — | 63.6 | |
| GTRKnowledge Representation Type=Unstructured Knowledge Representation Augmented2026.06 | — | 49.1 | |
| HippoRAG (reproduced)Knowledge Representation Type=Structured Knowledge Representation Augmented, Reproduction Status=reproduced2026.06 | — | 53.2 | |
| HippoRAG 2Knowledge Representation Type=Structured Knowledge Representation Augmented2026.06 | — | 74.7 | |
| HippoRAG 2 (reproduced)Knowledge Representation Type=Structured Knowledge Representation Augmented, Reproduction Status=reproduced2026.06 | — | 73.8 | |
| NV-Embed-v2 (7B)Knowledge Representation Type=Embedding-based Knowledge Representation Augmented, Number of Parameters=7B2026.06 | — | 69.7 | |
| RAPTORKnowledge Representation Type=Structured Knowledge Representation Augmented2026.06 | — | 57.8 | |
| SemFlowRAGKnowledge Representation Type=Structured Knowledge Representation Augmented2026.06 | — | 75.2 |