Long-context answering with citations on Dureader
73.4Citation RecallGLM-4
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| GLM-4Strategy=LAC-S2024.09 | 73.4 | 82.3 | 75 | |
| LongCite-8BStrategy=LAC-S2024.09 | 68.3 | 85.6 | 73.1 | |
| Claude-3-sonnetStrategy=LAC-S2024.09 | 67.7 | 89.2 | 75.5 | |
| LongCite-9BStrategy=LAC-S2024.09 | 67.6 | 89.2 | 74.4 | |
| GPT-4oStrategy=LAC-S2024.09 | 65.6 | 74.2 | 67.4 | |
| Mistral-Large-InstructStrategy=LAC-S2024.09 | 58.3 | 67 | 60.1 | |
| GLM-4-9B-chatStrategy=LAC-S2024.09 | 45.4 | 48.3 | 40.9 | |
| Llama-3.1-70B-InstructStrategy=LAC-S2024.09 | 38.2 | 46 | 35.4 | |
| Llama-3.1-8B-InstructStrategy=LAC-S2024.09 | 22 | 25.1 | 17 |