Long-context recall on NIAH Single 2
1Recall @ 32K ContextQwen3
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen3Param=1.7B, Token=-, Evaluation Protocol=Evaluated with YaRN, Architecture=teacher, no RNNs2026.01 | 1 | 0.988 | 0.248 | 0.192 | |
| KL-LS (GDN)Param=3B, Token=25B2026.01 | 0.994 | 0.496 | 0.282 | 0.104 | |
| HypeNet + HALOParam=2B, Token=2.3B2026.01 | 0.952 | 0.996 | 0.978 | 0.862 | |
| Jet-NemotronParam=2B, Token=400B2026.01 | 0.942 | 0.65 | 0 | 0 |