Long-context retrieval and aggregation on RULER 4K context length
98.9NIAH-MVFOCUSFT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| FOCUSFT2026.05 | 98.9 | 97.8 | 100 | |
| Pretrained2026.05 | 98.4 | 98 | 99.6 | |
| Standard SFT2026.05 | 97.6 | 97.5 | 100 |
| Method | Links | |||
|---|---|---|---|---|
| FOCUSFT2026.05 | 98.9 | 97.8 | 100 | |
| Pretrained2026.05 | 98.4 | 98 | 99.6 | |
| Standard SFT2026.05 | 97.6 | 97.5 | 100 |