In-context retrieval on SQuAD
41.3AccuracyHybrid M2RNN
Evaluation Results
| Method | Links | |
|---|---|---|
| Hybrid M2RNNParameter scale=410M2026.03 | 41.3 | |
| Hybrid Mamba-2 + M2RNN-3Parameter scale=410M2026.03 | 40.4 | |
| Hybrid Gated DeltaNet + M2RNN-1Parameter scale=410M2026.03 | 39.8 | |
| Hybrid Gated DeltaNet + M2RNN-3Parameter scale=410M2026.03 | 39.3 | |
| Hybrid Mamba-2 + M2RNN-1Parameter scale=410M2026.03 | 38.6 | |
| Hybrid Mamba-2Parameter scale=410M2026.03 | 37.6 | |
| Hybrid Gated DeltaNetParameter scale=410M2026.03 | 37.6 | |
| Gated DeltaNetParameter scale=410M2026.03 | 34.8 | |
| M2RNNParameter scale=410M2026.03 | 34 | |
| Mamba-2Parameter scale=410M2026.03 | 31.2 | |
| GRUParameter scale=410M2026.03 | 15.8 | |
| Transformer++Parameter scale=410M2026.03 | 14 | |
| RNNParameter scale=410M2026.03 | 9.6 |