Long-context retrieval on NIAH-Single-2 256K context
93.4AccuracyPre-train
Evaluation Results
| Method | Links | |
|---|---|---|
| Pre-trainModel=HypeNet-5B2026.06 | 93.4 | |
| +QK-RestoreModel=HypeNet-5B2026.06 | 85.4 | |
| +SFTModel=HypeNet-5B2026.06 | 83.6 | |
| Pre-trainModel=HypeNet-9B2026.06 | 67.2 | |
| Pre-trainModel=HypeNet-2B2026.06 | 56.2 | |
| +SFTModel=HypeNet-2B2026.06 | 41.8 | |
| +QK-RestoreModel=HypeNet-2B2026.06 | 40.8 | |
| Pre-trainModel=Jet-Nemotron-2B2026.06 | 22.2 | |
| +QK-RestoreModel=HypeNet-9B2026.06 | 19.6 | |
| +SFTModel=Jet-Nemotron-2B2026.06 | 11.6 | |
| +SFTModel=HypeNet-9B2026.06 | 9.4 | |
| +QK-RestoreModel=Jet-Nemotron-2B2026.06 | 7.6 |