Long-context retrieval and reasoning on LV-Eval (Context Lengths)
58.82Performance (16k Context)Qwen2-72B-Instruct
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Qwen2-72B-InstructYARN+DCA=false2024.07 | 58.82 | 56.7 | 42.92 | 31.79 | 2.88 | |
| Qwen2-72B-InstructYARN+DCA=true2024.07 | 58.82 | 56.7 | 53.03 | 48.83 | 42.35 | |
| Qwen2-7B-InstructYARN+DCA=false2024.07 | 49.77 | 46.93 | 28.03 | 11.01 | 0.55 | |
| Qwen2-7B-InstructYARN+DCA=true2024.07 | 49.77 | 46.93 | 42.14 | 36.64 | 34.72 | |
| ChatGLM4-9B-1M2024.07 | 46.4 | 43.23 | 42.92 | 40.41 | 36.95 |