Retrieval-Augmented Generation on KR1
84.4PC (MR)RAPS-DA
Evaluation Results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RAPS-DABackbone=Qwen2.5-14B-Instruct2026.06 | 84.4 | — | 85.7 | 88.2 | 60.3 | 55.8 | 42.6 | 78.5 | 75.3 | 38.5 | 47.8 | 17.6 | 60.4 | 76.9 | 52.9 | |
| RAPS-DABackbone=Qwen2.5-3B-Instruct2026.06 | 74.8 | — | 73.5 | 83.2 | 35.6 | 33.3 | 28.7 | 69.8 | 57.4 | 28.4 | 39.9 | 12.1 | 52 | 63.6 | 32.5 | |
| GRPO w/ RAGBackbone=Qwen2.5-14B-Instruct2026.06 | 73.6 | — | 74.7 | 78.9 | 38.9 | 31.7 | 36.5 | 74 | — | 34.7 | 39 | 14.7 | — | — | 35.7 | |
| GRPO w/ RAGBackbone=Qwen2.5-3B-Instruct2026.06 | 71 | — | 70.6 | 80 | 19.9 | 15.1 | 19.4 | 66.7 | 53.6 | 24.8 | 35.6 | 9.1 | 48.5 | 60.1 | 18.1 | |
| KR1Backbone=Qwen2.5-14B-Instruct2026.06 | 70.6 | — | 75.2 | 78.8 | 47.8 | 33.1 | 38.1 | 72.5 | — | 37 | 42.3 | 14.6 | — | — | 39.7 | |
| KR1Backbone=Qwen2.5-3B-Instruct2026.06 | 66.2 | — | 59.8 | 78.7 | 28.8 | 28.6 | 22 | — | 54.7 | — | — | — | — | — | 26.5 | |
| RAG promptingBackbone=Qwen2.5-14B-Instruct2026.06 | 63.2 | — | 63.9 | 73.3 | 22.2 | 13 | 28.1 | 61.9 | 70.5 | 22.9 | 22.5 | 6.9 | 42.1 | 66.2 | 21.1 | |
| RAG promptingBackbone=Qwen2.5-3B-Instruct2026.06 | 55.1 | — | 52.9 | 59.5 | 13.5 | 8.7 | 6.8 | 51.2 | 42.2 | 14.7 | 13 | 3.6 | 33.1 | 46.7 | 9.7 | |
| Query-onlyBackbone=Qwen2.5-14B-Instruct2026.06 | 30 | — | 26.5 | 35.8 | 30.6 | 28.9 | 39.4 | 30.9 | 70.8 | 25.4 | 26.7 | 5.7 | 25 | 50.9 | 33 | |
| Query-onlyBackbone=Qwen2.5-3B-Instruct2026.06 | 16.8 | — | 18.4 | 23.9 | 16 | 18.3 | 22.6 | 20.8 | 53.3 | 12.6 | 11.3 | 2.1 | 14.2 | 37 | 19 |