Long-context Question Answering on 2WikiMQA Fixed Chunk 2048
52.53QA ScoreBaseline
Evaluation Results
| Method | Links | |
|---|---|---|
| BaselineModel=ChatGLM, Setting=Fixed Chunk (2048)2026.03 | 52.53 | |
| Our + ReorderModel=ChatGLM, Setting=Fixed Chunk (2048)2026.03 | 51.76 | |
| BaselineModel=Qwen, Setting=Fixed Chunk (2048)2026.03 | 51.61 | |
| OurModel=Qwen, Setting=Fixed Chunk (2048)2026.03 | 50.89 | |
| OurModel=ChatGLM, Setting=Fixed Chunk (2048)2026.03 | 50.64 | |
| Our + ReorderModel=Qwen, Setting=Fixed Chunk (2048)2026.03 | 47.73 | |
| OurModel=LLaMA, Setting=Fixed Chunk (2048)2026.03 | 46.35 | |
| BaselineModel=LLaMA, Setting=Fixed Chunk (2048)2026.03 | 45.88 | |
| Our + ReorderModel=LLaMA, Setting=Fixed Chunk (2048)2026.03 | 44.55 | |
| CacheBlendModel=Qwen, Setting=Fixed Chunk (2048)2026.03 | 44.17 | |
| EPIC (15%)Model=ChatGLM, Setting=Fixed Chunk (2048)2026.03 | 44.01 | |
| No RecomputeModel=ChatGLM, Setting=Fixed Chunk (2048)2026.03 | 43.7 | |
| EPIC (15%)Model=Qwen, Setting=Fixed Chunk (2048)2026.03 | 43.21 | |
| CacheBlendModel=ChatGLM, Setting=Fixed Chunk (2048)2026.03 | 42.26 | |
| CacheBlendModel=LLaMA, Setting=Fixed Chunk (2048)2026.03 | 41.31 | |
| EPIC (15%)Model=LLaMA, Setting=Fixed Chunk (2048)2026.03 | 40.87 | |
| No RecomputeModel=LLaMA, Setting=Fixed Chunk (2048)2026.03 | 39.69 | |
| No RecomputeModel=Qwen, Setting=Fixed Chunk (2048)2026.03 | 39.48 |