In-context retrieval on TriviaQA
56.7AccuracyHybrid Mamba-2 + M2RNN-3
Evaluation Results
| Method | Links | |
|---|---|---|
| Hybrid Mamba-2 + M2RNN-3Parameter scale=410M2026.03 | 56.7 | |
| Hybrid Mamba-2 + M2RNN-1Parameter scale=410M2026.03 | 55.7 | |
| Hybrid M2RNNParameter scale=410M2026.03 | 55.5 | |
| Hybrid Gated DeltaNet + M2RNN-3Parameter scale=410M2026.03 | 54.9 | |
| Hybrid Gated DeltaNet + M2RNN-1Parameter scale=410M2026.03 | 54.6 | |
| Hybrid Gated DeltaNetParameter scale=410M2026.03 | 53.4 | |
| Transformer++Parameter scale=410M2026.03 | 52.4 | |
| M2RNNParameter scale=410M2026.03 | 52.3 | |
| Gated DeltaNetParameter scale=410M2026.03 | 51.7 | |
| Hybrid Mamba-2Parameter scale=410M2026.03 | 51.7 | |
| Mamba-2Parameter scale=410M2026.03 | 50.5 | |
| GRUParameter scale=410M2026.03 | 41.5 | |
| RNNParameter scale=410M2026.03 | 29.7 |