Long-context Retrieval on NIAH Single 2
100Accuracy (1024 tokens)GDN
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GDNModel (1.5B)=1.5B2026.05 | 100 | 93.8 | 49.8 | |
| Mamba-2Model (1.5B)=1.5B2026.05 | 100 | 53.8 | 11.8 | |
| Looped GDNModel (1.5B)=1.5B, Looped=true2026.05 | 100 | 96.4 | 53.2 | |
| Looped Mamba-2Model (1.5B)=1.5B, Looped=true2026.05 | 100 | 57.1 | 13.5 | |
| Looped Hybrid (GDN+DSA)Model (1.5B)=1.5B, Looped=true2026.05 | 100 | 100 | 77.6 | |
| Looped Hybrid (Full+GDN)Model (1.5B)=1.5B, Looped=true2026.05 | 100 | 100 | 81 | |
| Looped TransformerModel (1.5B)=1.5B, Looped=true2026.05 | 94.6 | 100 | 0 | |
| TransformerModel (1.5B)=1.5B2026.05 | 92.2 | 100 | 0 |