Multi-document Question Answering on Loong
68Average ScoreBaseline
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| BaselineContext=Full Context2026.03 | 68 | 0.273 | 249.1 | |
| SPD-RAG2026.03 | 58.1 | 0.103 | 564.1 | |
| ProxyCoT-RLModel=Qwen3-4B-Instruct2026.04 | 42.51 | 40.83 | — | |
| Normal RAG2026.03 | 33 | 0.08 | 412.5 | |
| Agentic RAG2026.03 | 32.8 | 0.098 | 334.7 | |
| Qwen3-4B-InstructTraining=Zero-shot2026.04 | 24.91 | 37.76 | — | |
| ProxyCoT-RLModel=Gemma3-4B-IT2026.04 | 24.32 | 32.05 | — | |
| Gemma3-4B-ITTraining=Zero-shot2026.04 | 3.55 | 25.85 | — |