Long-context Answering with Citations on LongBench-Cite Average
72Citation F1LongCite-8B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LongCite-8BStrategy=LAC-S2024.09 | 72 | 85 | |
| LongCite-9BStrategy=LAC-S2024.09 | 69.2 | 91 | |
| Claude-3-sonnetStrategy=LAC-S2024.09 | 67.2 | 132 | |
| GPT-4oStrategy=LAC-S2024.09 | 65.6 | 220 | |
| GLM-4Strategy=LAC-S2024.09 | 65.4 | 169 | |
| Mistral-Large-InstructStrategy=LAC-S2024.09 | 51.5 | 132 | |
| Llama-3.1-70B-InstructStrategy=LAC-S2024.09 | 40.4 | 174 | |
| GLM-4-9B-chatStrategy=LAC-S2024.09 | 27.2 | 96 | |
| Llama-3.1-8B-InstructStrategy=LAC-S2024.09 | 19.7 | 100 |