Long-context Retrieval on RULER niah_single_2 zeroshot (test)
95.5AccuracyPhiMoE (FP16 baseline)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| PhiMoE (FP16 baseline)Precision=FP16, Shot count=zero-shot2026.04 | 95.5 | — | — | — | |
| NPUMoEPrecision=FP16, Shot count=zero-shot2026.04 | 95 | 21.77 | 37.49 | -0.52 |