Long-context retrieval on NIAH-Single-1 128K
100AccuracyPre-train
Evaluation results
| Method | Links | |
|---|---|---|
| Pre-trainModel=Jet-Nemotron-2B2026.06 | 100 | |
| +SFTModel=Jet-Nemotron-2B2026.06 | 100 | |
| +QK-RestoreModel=Jet-Nemotron-2B2026.06 | 100 | |
| Pre-trainModel=HypeNet-5B2026.06 | 100 | |
| +SFTModel=HypeNet-5B2026.06 | 100 | |
| +QK-RestoreModel=HypeNet-5B2026.06 | 100 | |
| Pre-trainModel=HypeNet-9B2026.06 | 100 | |
| +SFTModel=HypeNet-9B2026.06 | 100 | |
| +QK-RestoreModel=HypeNet-9B2026.06 | 100 | |
| +QK-RestoreModel=HypeNet-2B2026.06 | 99.4 | |
| +SFTModel=HypeNet-2B2026.06 | 99.2 | |
| Pre-trainModel=HypeNet-2B2026.06 | 97.2 |