Knowledge Retrieval and Integration on KR1 protocol (in-distribution)
89.3PC-MRRAPS-DA
Evaluation Results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RAPS-DABackbone=Qwen2.5-7B-Instruct2026.06 | 89.3 | — | 87 | 80.9 | 51.5 | 48.2 | 44.6 | 87.3 | 69.9 | 33.9 | 45.5 | 20.1 | 59.5 | 78.6 | 48.1 | |
| TIPBackbone=Qwen2.5-7B-Instruct2026.06 | 81.8 | — | 79.1 | 76.8 | 33.3 | 36.5 | 41 | 76.2 | 52.9 | 31.7 | 44.5 | 15.2 | 54.9 | 64.5 | 36.9 | |
| RAPS-DABackbone=Llama3.1-8B-Instruct2026.06 | 79.5 | — | 83.7 | 84.4 | 67.1 | 52.9 | 55.7 | 77.8 | 52.6 | 40.2 | 49.5 | 18.3 | 59.2 | 65.2 | 58.6 | |
| GRPO w/ RAGBackbone=Llama3.1-8B-Instruct2026.06 | 78 | — | 79.7 | 82.6 | 41.6 | 35.7 | 39.3 | 76.6 | 47.6 | 34.8 | 41.2 | 16.6 | 55.5 | 62.1 | 38.8 | |
| GRPO w/ RAGBackbone=Qwen2.5-7B-Instruct2026.06 | 77.6 | — | 77.4 | 80 | 26.9 | 19.7 | 26 | 75.3 | 66.5 | 27.9 | 34 | 11.8 | 51.4 | 70.9 | 24.2 | |
| TIPBackbone=Llama3.1-8B-Instruct2026.06 | 75.4 | — | 78.1 | 45.8 | 62.5 | 58.9 | 41.3 | 65.5 | 36.4 | 35.2 | 43.7 | 15.2 | 48.9 | 51 | 54.2 | |
| KR1Backbone=Qwen2.5-7B-Instruct2026.06 | 75.1 | — | 75.5 | 80.9 | 43.9 | 37.3 | 29.4 | 76.3 | 67.6 | 31.4 | — | 12 | 52.1 | 72 | 36.9 | |
| KR1Backbone=Llama3.1-8B-Instruct2026.06 | 73.8 | — | 80.2 | 80 | 55.4 | 41.1 | 44.6 | 73.7 | 49.6 | 37.1 | 45.4 | 14.7 | 55.2 | 61.6 | 47 | |
| SFTBackbone=Llama3.1-8B-Instruct2026.06 | 72.9 | — | 79.2 | 73.8 | 42.6 | 35.5 | 35.9 | 70.1 | 47.2 | 33.6 | 38.2 | 13.4 | 51.9 | 58.6 | 38 | |
| SFTBackbone=Qwen2.5-7B-Instruct2026.06 | 72 | — | 77.7 | 74.7 | 24.9 | 21.1 | 22 | 68.5 | 66.6 | 30.1 | 32.2 | 11.8 | 49.7 | 67.5 | 22.6 | |
| AstuteBackbone=Llama3.1-8B-Instruct2026.06 | 65.8 | — | 64.9 | 78 | 17 | 9.1 | 17.3 | 59.8 | 40.1 | 1.6 | 30.3 | 9.6 | 41.7 | 50 | 14.4 | |
| RAG promptingBackbone=Qwen2.5-7B-Instruct2026.06 | 65.7 | — | 66.4 | 74.3 | 13.5 | 8.1 | 11.3 | 59.5 | 62.2 | 20.4 | 22.5 | 6.4 | 42.6 | 60.9 | 10.9 | |
| AstuteBackbone=Qwen2.5-7B-Instruct2026.06 | 65.5 | — | 66 | 77.6 | 12.8 | 7.1 | 10.3 | 54.2 | 56.7 | 17.9 | 20.4 | 6.3 | 42.3 | 55.5 | 10.1 | |
| RAG promptingBackbone=Llama3.1-8B-Instruct2026.06 | 64.8 | — | 62 | 76.4 | 22.9 | 16.3 | 24.9 | 61.2 | 39.2 | 24.4 | 23.5 | 8.2 | 43.2 | 50.2 | 21.4 | |
| CK-PLUGBackbone=Qwen2.5-7B-Instruct2026.06 | 64.7 | — | 66.5 | 78.7 | 11.6 | 8.1 | 7.9 | 55 | 55 | 22.7 | 24.8 | 6.2 | 43.9 | 55 | 9.2 | |
| OPCDBackbone=Qwen2.5-7B-Instruct2026.06 | 64.7 | — | 64.4 | 62.8 | 40.2 | 37 | 15.8 | 56.5 | 18.1 | 24.6 | 34.5 | 8.8 | 43.3 | 37.3 | 31 | |
| OPCDBackbone=Llama3.1-8B-Instruct2026.06 | 61.8 | — | 59.5 | 70.2 | 46.7 | 40.4 | 32.9 | 54.2 | 25.6 | 27.8 | 38.7 | 8.5 | 44.4 | 39.9 | 40 | |
| CK-PLUGBackbone=Llama3.1-8B-Instruct2026.06 | 54.8 | — | 58.5 | 69.7 | 12.1 | 9.1 | 17.3 | 42 | 31.5 | 22.4 | 24.6 | 5.2 | 39.2 | 36.8 | 12.8 | |
| Query-onlyBackbone=Llama3.1-8B-Instruct2026.06 | 29.4 | — | 26.2 | 39.9 | 27.1 | 27.6 | 42.6 | 32.1 | 43.3 | 20.7 | 21 | 6.2 | 23.9 | 37.7 | 32.5 | |
| Query-onlyBackbone=Qwen2.5-7B-Instruct2026.06 | 27.7 | — | 24.7 | 31.7 | 25.9 | 25.8 | 32.3 | 29.7 | 64.5 | 20.9 | 25.5 | 4.4 | 22.5 | 47.1 | 28 |