Training Efficiency on Distillation (train)
150Total Tokens ProcessedBitMamba-2
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BitMamba-2Architecture=Mamba-2, Precision=1.58-bit2026.04 | 150 | — | |
| Bi-MambaArchitecture=Mamba-1, Precision=1-bit2026.04 | 105 | 32 | |
| DF-SSMArchitecture=Mamba-2, Precision=1-bit + int8 LoRA2026.04 | 32 | 1 |