In-context retrieval on SWDE
62.5AccuracyHybrid Gated DeltaNet + M2RNN-1
Evaluation Results
| Method | Links | |
|---|---|---|
| Hybrid Gated DeltaNet + M2RNN-1Parameter scale=410M2026.03 | 62.5 | |
| Hybrid Mamba-2Parameter scale=410M2026.03 | 61.7 | |
| Hybrid Mamba-2 + M2RNN-3Parameter scale=410M2026.03 | 61.7 | |
| Hybrid M2RNNParameter scale=410M2026.03 | 60.7 | |
| Hybrid Mamba-2 + M2RNN-1Parameter scale=410M2026.03 | 60.2 | |
| Hybrid Gated DeltaNetParameter scale=410M2026.03 | 58 | |
| Hybrid Gated DeltaNet + M2RNN-3Parameter scale=410M2026.03 | 56.5 | |
| Transformer++Parameter scale=410M2026.03 | 51.7 | |
| Gated DeltaNetParameter scale=410M2026.03 | 41.8 | |
| Mamba-2Parameter scale=410M2026.03 | 32.4 | |
| M2RNNParameter scale=410M2026.03 | 28.5 | |
| GRUParameter scale=410M2026.03 | 7.7 | |
| RNNParameter scale=410M2026.03 | 5 |