Long-context Answering with Citations on Longbench-Chat
0.62Citation RecallLongCite-8B
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| LongCite-8BStrategy=LAC-S2024.09 | 0.62 | 0.797 | 0.674 | — | — | — | — | — | — | — | — | — | |
| LongCite-9BStrategy=LAC-S2024.09 | 0.576 | 0.781 | 0.636 | — | — | — | — | — | — | — | — | — | |
| Claude-3-sonnetStrategy=LAC-S2024.09 | 0.52 | 0.678 | 0.551 | — | — | — | — | — | — | — | — | — | |
| GLM-4Strategy=LAC-S2024.09 | 0.476 | 0.539 | 0.471 | — | — | — | — | — | — | — | — | — | |
| GPT-4oStrategy=LAC-S2024.09 | 0.467 | 0.535 | 0.467 | — | — | — | — | — | — | — | — | — | |
| GLM-4-9B-chatStrategy=LAC-S2024.09 | 0.259 | 0.205 | 0.167 | — | — | — | — | — | — | — | — | — | |
| Llama-3.1-70B-InstructStrategy=LAC-S2024.09 | 0.258 | 0.32 | 0.232 | — | — | — | — | — | — | — | — | — | |
| Mistral-Large-InstructStrategy=LAC-S2024.09 | 0.198 | 0.239 | 0.19 | — | — | — | — | — | — | — | — | — | |
| Llama-3.1-8B-InstructStrategy=LAC-S2024.09 | 0.141 | 0.195 | 0.124 | — | — | — | — | — | — | — | — | — | |
| GLM-42024.09 | — | — | — | 61.2 | 67.5 | 60.2 | 47.6 | 53.9 | 47.1 | 46.1 | 29.1 | 30.8 | |
| LongCite-8B2024.09 | — | — | — | 79.6 | 88.9 | 82.6 | 62 | 79.7 | 67.4 | 59.6 | 39.5 | 42 | |
| LongCite-9B2024.09 | — | — | — | 72.8 | 84.2 | 75.8 | 57.6 | 78.1 | 63.6 | 64.2 | 45.1 | 47.1 |