Exact Retrieval on Needle-in-a-Haystack (NIAH) 64K
97.6Average AccuracyBaseline SFT
Evaluation Results
| Method | Links | |
|---|---|---|
| Baseline SFT2026.04 | 97.6 | |
| MEAN0.1Anchor=MEAN, p=0.12026.04 | 97.53 | |
| MEAN0.2Anchor=MEAN, p=0.22026.04 | 96.47 | |
| MIN0.1Anchor=MIN, p=0.12026.04 | 96.33 | |
| MIN0.2Anchor=MIN, p=0.22026.04 | 95.2 |