Long-context retrieval on Probing (val)
85.4Document AvgFILM-7B
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| FILM-7BParameters=7B2024.04 | 85.4 | 6.1 | 83.3 | 18.7 | 89 | 16.8 | 85.9 | 13.9 | |
| GPT-4-TurboType=Proprietary2024.04 | 81.3 | 31.7 | 66.1 | 46.5 | 89.6 | 18 | 79 | 32.1 | |
| Mistral-7B-Instruct-v0.2Parameters=7B, Version=v0.22024.04 | 74.2 | 32.1 | 20.3 | 59.5 | 47.5 | 77 | 47.3 | 56.2 | |
| LongAlign-13B-64KParameters=13B, Context length=64K2024.04 | 71.7 | 13.4 | 50.8 | 40.8 | 82.9 | 27 | 68.5 | 27.1 | |
| InternLM2-chat-7BParameters=7B2024.04 | 68.8 | 18.7 | 50.2 | 44.1 | 61.2 | 57.1 | 60.1 | 40 | |
| InternLM2-chat-20BParameters=20B2024.04 | 66.4 | 27.2 | 63.4 | 45.5 | 74.9 | 57.2 | 68.2 | 43.3 | |
| LongAlign-7B-64KParameters=7B, Context length=64K2024.04 | 65.3 | 16.9 | 39.3 | 56 | 55 | 36.2 | 53.2 | 36.4 | |
| Mistral-7B-Instruct-v0.1Parameters=7B, Version=v0.12024.04 | 44.8 | 29.9 | 6.8 | 53.2 | 8.8 | 74.5 | 20.1 | 52.5 |