Question Answering on SQuAD (2048 tokens)
49.4AccuracyLooped Transformer
Evaluation results
| Method | Links | |
|---|---|---|
| Looped TransformerModel (1.5B)=1.5B, Looped=true, Iterations=T=4, Mixer=Pure2026.05 | 49.4 | |
| Looped Hybrid (Full+GDN)Model (1.5B)=1.5B, Looped=true, Iterations=T=4, Mixer=Hybrid, Hybrid Ratio=1:12026.05 | 48.9 | |
| Looped Hybrid (GDN+DSA)Model (1.5B)=1.5B, Looped=true, Iterations=T=4, Mixer=Hybrid, Hybrid Ratio=1:12026.05 | 48 | |
| TransformerModel (1.5B)=1.5B, Looped=false, Mixer=Pure2026.05 | 46.6 | |
| Looped GDNModel (1.5B)=1.5B, Looped=true, Iterations=T=4, Mixer=Pure2026.05 | 41.8 | |
| Looped Mamba-2Model (1.5B)=1.5B, Looped=true, Iterations=T=4, Mixer=Pure2026.05 | 40.5 | |
| GDNModel (1.5B)=1.5B, Looped=false, Mixer=Pure2026.05 | 40 | |
| Mamba-2Model (1.5B)=1.5B, Looped=false, Mixer=Pure2026.05 | 39.1 |